Value Engineering, 360°

1.1 · Value is a ratio

Everything in this course rests on one equation: Value = Function ÷ Cost. Function is what the product must do — its performance, reliability and appeal. Cost is everything it takes to deliver that function across the lifecycle. A product is not "expensive" or "cheap" in isolation; it is good or bad value relative to the function it delivers. Value Engineering is the discipline of improving that ratio deliberately.

The method was born at General Electric in 1947, when Lawrence D. Miles noticed that wartime material substitutions — forced by shortages — often produced parts that worked better and cost less. His insight: teams had been buying parts when they should have been buying functions. Ask "what does it do, and what else could do it?" and whole design spaces open up.

Seventy-five years on, the trajectory only points one way: from a wartime workaround to a named engineering discipline, to professional societies and certifications, into federal law and European standards — and now into an AI-accelerated practice. Methods that merely fashion come and go in a decade; methods that work compound for generations. This one works.

1.2 · Five ways to improve value

• Same function, lower cost — classic cost-out (most VAVE work lives here)
• More function, same cost — value-up for competitiveness
• More function, lower cost — the VE ideal; happens more often than you'd think
• Function grows faster than cost — justified premiumisation
• Trim unvalued function, cut cost sharply — data-backed de-contenting

Note what is not on the list: cutting function customers value to save cost. That is how cost-cutting destroys brands — and it is precisely what the function-first method prevents.

1.3 · VA, VE and VM

Value Analysis (VA) applies the method to products already in production — teardown, question, re-engineer, implement as running changes. Value Engineering (VE) applies it during design, before cost is locked in — and since roughly 80% of lifecycle cost is committed by early design decisions, a pound of effort in design returns what ten return later. Value Management (VM) is the governance layer that makes both permanent: targets, funnels, cadence, capability. Together the practice is called VAVE.

1.4 · The four types of value

• Use value — what the product does; its work-performing functions
• Esteem value — what makes it desirable: brand, finish, perceived quality
• Exchange value — what it can be traded for: price power, resale
• Cost value — the sum of material, labour, overhead and lifecycle cost

Esteem value is real value — customers pay for it. The VE question is never "kill the chrome"; it is "does this chrome deliver more esteem than it costs?"

1.5 · Value, price and cost — untangling the trio

Three words engineers constantly blur: price is what the market will pay — set by competition and perceived value, largely outside your control. Cost is what you spend to deliver — set by your design and supply choices, largely inside your control. Value is the ratio the customer experiences. The market caps the price; physics floors the cost; everything between is margin plus the headroom VE exists to capture. When someone says "we can't afford that feature", translate it: the function's cost exceeds the price the market pays for it — now it's an engineering problem with three levers, not a budget complaint.

1.6 · Why the economics never stop favouring VE

The cost-commitment curve is the discipline's cornerstone: by the time concept design is frozen, roughly 70–80% of lifecycle cost is committed while less than 10% has actually been spent. Every later phase can only optimise inside decisions already made — which is why an hour of function analysis at a design gate outperforms a month of price negotiation after tooling kick-off. Add the arithmetic of scale: a study costing a few person-weeks that removes even 5% from a high-volume product returns its cost hundreds of times over. This is why VE is one of the few methodologies written into law: US federal agencies are required to apply it (OMB Circular A-131), and federal-aid highway projects above statutory thresholds must undergo VE analysis before construction approval.

2.1 · Scope and sponsorship

Every failed VE study fails before it starts. Preparation takes 2–4 weeks and begins with a sponsor — someone senior enough to commit resources and act on decisions — plus a written scope: which product, which cost baseline, what target (typically 10–20% of the addressed cost), what's explicitly out of bounds. An unscoped study wanders; an unsponsored one produces slides.

2.2 · The team

Six to ten people, cross-functional by design: design engineering, manufacturing, purchasing, quality, finance, service — plus key suppliers for commodity insight. Homogeneous teams produce homogeneous ideas; the friction between functions is where the ideas live. The team is led by a trained facilitator who owns the process, not the content — ideally certified (VMA/AVS/CVS, covered in Module 12).

2.3 · The data pack

• Costed BOM — every part with material, process, labour and overhead cost, validated by finance. This is non-negotiable.
• Quality history — warranty Paretos, scrap and rework data, field-failure modes
• Voice of customer — requirements, satisfaction data, feature-usage evidence
• Volumes & forecasts — because every lever's economics depend on quantity
• Competitor samples — procured early; teardown insight feeds every phase
• Physical parts — the actual product in the room. Teams ideate on what they can touch.

2.4 · Choosing the right first product

Programmes live or die on the first study's credibility, so pick the target deliberately. Score candidates on: spend (annual volume × unit cost — enough zeros to matter), design authority (you can actually change it), stability (2+ years of production life left to harvest savings), pain (margin pressure or warranty trouble creates pull), and data (a costed BOM exists or can be built in two weeks). A mid-complexity, high-volume product your own engineers designed is the ideal first target; the flagship with frozen tooling and a defensive chief engineer is the worst.

2.5 · Roles that make or break the study

• Sponsor — owns the target, attends kick-off and Phase 6, clears roadblocks. No sponsor, no study.
• Facilitator — owns the process and the clock; deliberately NOT the deepest product expert (experts defend designs).
• Finance partner — validates the baseline and every savings claim; involving them from day one prevents the classic "those numbers aren't real" ambush in Phase 6.
• Idea champions — assigned per surviving idea in Phase 4; they carry proposals through development.
• Scribe — captures every idea verbatim with its function tag; memory is not a capture system.

2.6 · Freeze the baseline, then set the target

Before the workshop, freeze a dated cost baseline — a specific BOM revision, volume basis and currency — and agree the measurement rules (gross vs. net, how commodity swings are neutralised). Savings claimed against a moving baseline convince no one. Then set the target from evidence: a benchmark gap, a margin requirement, or the value-mismatch total from a prior scan — not a round number pulled from the ceiling.

2.7 · The workshop week — a proven shape

• Day 1 — Information & functions — cost walk, line walk, product on the table; function analysis begins while the data is fresh
• Day 2 — FAST, worth & targets — the FAST diagram, function–cost matrix and Value Index; the team leaves knowing exactly where the mismatches are
• Day 3 — Creative — full-day divergence on the target functions: brainwriting, SCAMPER, TRIZ; judgement banned until sundown
• Day 4 — Evaluation & champions — triage, screens and the weighted matrix; every survivor leaves with a named champion and a gate date
• Day 5 — Develop & present — business cases drafted, risks flagged, and the sponsor decision meeting closes the week with a decision log

Compressed three-day formats work for smaller scopes (merge days 1–2 and 4–5); split-week formats — two days, a fortnight's gap for data-chasing, then two more — suit complex products. What never survives compression is the separation of phases: the day creative thinking and evaluation happen in the same session, both die.

3.1 · Know the thing cold

The Information Phase assembles the complete picture before anyone proposes anything. Miles' founding questions structure it: What is it? What does it do? What does it cost? What is it worth? What else could do the job? The discipline of answering the first four before touching the fifth is what separates a structured VE study from a brainstorm.

3.2 · The cost walk

Walk the costed BOM as a team: material, conversion, labour, overhead, logistics, warranty. Then Pareto it — typically 20% of parts carry 80% of cost, and those parts get the analytical firepower. Complement with a process walk: visit the line, watch the assembly, time the operations. Cost that looks reasonable on a spreadsheet often looks absurd on the shop floor.

3.3 · Requirements, not assumptions

Capture what customers actually require — via QFD, Kano analysis or plain structured interviews — and tag each requirement as must-have, performance, or delighter. Later, when someone claims "customers need this", the team checks the evidence, not the volume of the voice.

3.4 · Structuring cost data so it talks

A costed BOM answers "what does each part cost?" — but the Information Phase needs "why does it cost that?" Split every part into material / conversion / labour / overhead, and be suspicious of allocated overhead: standard-costing systems smear indirect cost by labour hours or material value, which can make a simple part look expensive and an expensive process look cheap. Compute cost density metrics — £/kg against material benchmarks, £ per function against the worth estimates coming in Phase 2 — because density exposes outliers that absolute Paretos hide.

3.5 · The Kano model in five minutes

• Must-be — expected silently (brakes work). Absence enrages; excellence earns nothing. Deliver at minimum adequate cost.
• Performance — more is better, linearly (fuel economy, capacity). Customers consciously trade money for these — spend here matches price.
• Attractive — unexpected delighters. High esteem-value leverage, but validate with data before spending.
• Indifferent — customers genuinely don't care. The prime de-spec hunting ground.
• Reverse — some customers actively dislike (complexity, extra buttons). Removing these raises value while cutting cost — the perfect VE move.

Mapping requirements to Kano categories converts the vague instruction "don't hurt the customer" into a surgical one: protect must-be and performance, interrogate attractive, attack indifferent and reverse.

3.6 · Building the data room — where the truth already lives

Most of what the Information Phase needs already exists inside the business — scattered, unowned, and never joined up. The high-yield sources: warranty claim text (what actually fails, in the customer's words — the richest function-language data you own), service-part sales (the parts bought as spares are the parts that break — a ranked reliability Pareto nobody compiled), line rework and scrap logs (where the design fights the factory daily), customer returns commentary (esteem and usability failures that never reach engineering), the spend cube (price variance for near-identical parts across plants and suppliers), and prior study archives (parked ideas from the last wave are pre-screened seeds for this one). Assign each source an owner in the preparation phase with a one-week deadline — and note that this mining is precisely what LLMs now accelerate best, clustering ten thousand warranty claims into function-language Paretos in an afternoon (Module 11 shows how). The team that walks in with six joined-up data sources doesn't argue about opinions; it argues about which fact to attack first.

4.1 · What is a function? (Start here)

Look at the pen on your desk. Most people see a thing: a tube, a spring, a clip, some ink. A value engineer sees jobs: the ink makes marks, the spring stores energy, the clip secures the pen to a pocket. A function is the job a thing does — not the thing itself. That one shift in seeing is the heart of Value Engineering.

Why does it matter so much? Because customers never buy parts — they buy jobs done. Nobody wants a heating element; they want hot water. Nobody wants six screws; they want two parts held together. The moment you describe a product as jobs instead of parts, you free yourself to ask the most powerful question in engineering: "what else could do this job — better, or cheaper?"

Remember Lawrence Miles' insight from Module 1: teams were buying parts when they should have been buying functions. Phase 2 is where that idea becomes a working method.

4.2 · The two-word rule: active verb + measurable noun

Value Engineering writes every function as exactly two words: an active verb (a doing word) plus a measurable noun (a thing you can count, weigh, or test). "Heat water." "Transmit torque." "Support load." Two words. Always.

Why so strict? Three reasons, and each one earns money:

• Two words force clarity. If you need a sentence, you haven't understood the job yet. "It provides a secure mounting interface for the assembly" becomes, honestly: position component.
• Two words remove the solution. "Fasten flange" secretly assumes bolts. "Join components" opens welding, adhesives, snap-fits — or redesigning so there is nothing to join. The vaguer the verb–noun pair sounds, the wider the door to better ideas.
• Two words can be measured. "Conduct current" — you can measure amps. "Support load" — you can measure newtons. If you can measure it, you can put a cost and a worth on it. If you can't measure it, you can't engineer it.

The acid test for any function statement: could two engineers independently agree how to measure whether the job is being done? "Emit light" — yes, lumens. "Provide illumination solutions" — no. Rewrite it.

4.3 · The banned verbs (and how to spot a fake function)

Some verbs sound professional but say nothing. Ban them: provide, allow, enable, facilitate, ensure, support (as in "support the user"), improve, assist. They fail the measurement test, and they usually smuggle a part into the sentence. "Provide mounting" is just a bracket wearing a disguise. "Enable operation" is the whole product wearing a disguise.

Fake functions come in four flavours — learn to spot each one:

• The banned verb — "provide heat" → say generate heat.
• The hidden solution — "fasten with screws" → the screws are one answer, not the job. Say join components.
• The unmeasurable noun — "create convenience" → convenience isn't measurable. What actually happens? reduce effort, save time.
• The product in disguise — "boil kettle" → the kettle is the product, not the job. Say heat water.

4.4 · The function family — every type, one by one

Not all functions are equal, and the differences are where the money hides. Here is the whole family, each with a plain-English test you can apply in seconds:

• Basic function — the reason the product exists. Test: remove it — is the product now pointless? A kettle that cannot heat water is a jug. There is exactly one basic function per study scope; if you find three, you are really looking at three studies.
• Secondary functions — jobs that exist only because of how this particular design achieves the basic function. The kettle's element must resist corrosion only because we chose a metal element immersed in water. Choose a different design and many secondary functions simply vanish — this is why most removable cost lives here.
• Required secondary functions — demanded by law, codes or safety standards: suppress interference, resist flame. Test: would a regulator object if it vanished? You cannot delete these — but you can achieve them a cheaper way. Non-negotiable in existence, fully negotiable in how.
• Esteem functions — jobs of desirability: convey quality, please eye. The soft-close hinge, the satisfying click, the gloss finish. Test: would the product still work without it, but sell worse? Esteem value is real value — measure it with market data, not opinions.
• Unwanted functions — jobs the design does that nobody asked for: the motor generates heat, the pump creates noise. Each unwanted function usually breeds a whole family of corrective parts and cost (heat sinks, insulation, dampers). Kill the unwanted function and its children die with it.
• Unnecessary functions — functions that support nothing on the logic path. Pure gold: delete them and nothing is lost. They are usually left over from an older design generation nobody re-questioned.
• All-time functions — jobs performed constantly across the whole product, like ensure safety and convey esteem. On a FAST diagram they float above the main path.
• Higher-order & assumed functions — the objective your product serves (prepare beverage) sits outside the left scope line; things you take as given (supply power) sit outside the right. They anchor the logic but are not studied.

4.5 · The FAST diagram — watch it build, step by step

FAST stands for Function Analysis System Technique — a map that arranges your functions by logic instead of by geometry. It runs on two tiny questions: moving right asks HOW? a function is achieved; moving left asks WHY? it exists. If the chain reads correctly in both directions, your team truly understands the product. If it doesn't, a card is in the wrong place — and you just found a misunderstanding before it cost you money.

Press Next step and watch a real FAST diagram build itself for an electric kettle:

Two practical notes. First, when one function needs several functions together, stack them vertically (an AND); when there are alternative routes, branch them (an OR). Second, there are two flavours of FAST: the technical FAST you just watched, and the customer FAST, which starts from customer-need verbs (assure dependability, assure convenience) — useful when the scope is a whole product rather than a subsystem.

4.6 · The function library — your verb & noun toolbox

Stuck for the right two words? Use this library. Filter by category, steal the verb–noun pair, and adapt the noun to your product. Every entry is written the way a function should be: active verb, measurable noun.

4.7 · Loading the money: function–cost matrix, worth, and the Value Index

Logic alone doesn't save money — so now we load economics onto it. Take every part's cost and share it out across the functions that part serves. Rows add up to the component's cost; columns add up to each function's total cost. For the first time you can see what you are actually paying per job — and it is always a surprise.

Next, estimate each function's worth: the lowest cost that could still reliably do the job. Estimate it honestly, four ways: (1) the cheapest existing solution in any industry (a neon lamp indicates status for £0.15, whatever your current PCB does); (2) the physics floor — the minimum material and energy the job fundamentally needs; (3) your historical best; (4) expert triangulation when data is thin. Never set worth equal to current cost — that is the classic self-deception that guarantees the study finds nothing.

Finally divide: Value Index = Cost ÷ Worth. The VI turns feelings into a ranked target list:

• VI ≤ 1.2 — healthy; leave it alone
• VI 1.2–2.0 — watch list; challenge at the next design change
• VI > 2.0 — attack; this function goes straight to the Creative Phase

Example: a kettle's "indicate status" function costs £0.60, but a simple neon lamp does the same job for £0.15 — VI = 4.0. At a million units a year, that single row of arithmetic is worth £450,000. VE picks its battles with division, not debate.

5.1 · The most fragile moment in the whole job plan

You have just finished Phase 2. You know exactly which functions cost too much — the kettle's "indicate status" function costs £0.60 and is worth £0.15. Now comes the question the whole study exists for: "how else could we do this?" And here is the uncomfortable truth: the answer is fragile. New ideas arrive weak, half-formed and easy to kill. A raised eyebrow kills one. A quick "we tried that in 2019" kills another. The boss speaking first kills ten, because everyone quietly rearranges their thinking around what the boss said. The Creative Phase is not about being talented — it is about building a protective bubble where fragile ideas survive long enough to grow up.

One more mindset shift before the techniques: you are not looking for the answer. A strong workshop produces 200–400 raw ideas, and only about one in eight survives evaluation. That is not failure — that is the maths. If you want 25 implemented improvements, you need a couple of hundred raw ideas to start with. Quantity is the strategy, and every technique in this module is simply a different pump for producing it.

5.2 · Osborn's four rules — the law of the room

Alex Osborn, the advertising executive who invented brainstorming in the 1940s, wrote four rules. They sound soft. They are not — they are the operating system of the phase, and one visible violation switches the room back into judging mode for the rest of the day.

• 1 · Defer all judgement. No criticism, no praise either — "great idea!" tells everyone else theirs wasn't. Evaluation has its own phase; it can wait a day. In the room it sounds like: an engineer says "make the bracket from cardboard", and nobody laughs — someone writes it down.
• 2 · Welcome wild ideas. The silly idea is rarely the answer, but it stretches the space where the answer lives. "Cardboard bracket" is absurd — but it provokes "how much load does this bracket actually see?", and that question finds the 40% weight saving.
• 3 · Go for quantity. Set a number and chase it: "ten more ideas for 'contain water' before the break." The first five ideas are the obvious ones everyone already had. The gems hide in ideas 15 to 30, after the easy answers run out.
• 4 · Build on ideas — "yes, and…". Ideas are seeds, not verdicts. Someone says "snap-fit instead of screws"; the builder adds "yes, and if it snaps from the top, the robot can assemble it in one motion." Combination is where breakthroughs actually come from.

5.3 · Aim every technique at a function, never at a part

This is the difference between VE ideation and ordinary cost-down brainstorming. Ask a room "how do we make this shaft cheaper?" and you get thin answers: cheaper steel, looser tolerance, ask the supplier for 5%. The existing design sits in everyone's head like an anchor. Ask instead "how else can we transmit torque?" — the function, with the part deliberately out of sight — and the room produces splines, press-fits, polygon shafts, adhesives, friction welds, overmoulds. Ten different physics, not three discounts. Every technique below gets pointed at your worst value-index functions from Phase 2, one function at a time, written at the top of the board.

5.4 · Classic brainstorming — running it so it actually works

Everyone thinks they know brainstorming; most meetings called "brainstorms" are just discussions. The working version has mechanics: a facilitator who enforces the four rules and does not contribute ideas; one function on the board at a time; 25-minute sprints with real breaks, because idea production collapses after half an hour; every idea captured verbatim with a function tag — paraphrasing kills the wild edges; and the most senior person in the room speaks last in every round, because whatever they say first becomes the anchor. When the flow dries up, the facilitator injects a provocation — "how would IKEA do this?", "what if it had to cost nothing?", "what does a £2 toy that does this look like?" — and the pump restarts. Fast, social and energising, but it favours loud voices, which is exactly why the next technique exists.

5.5 · Brainwriting 6-3-5 — the introvert's superpower

Six people, three ideas each, five minutes — in silence. Each person writes three ideas for the target function on a sheet, then passes the sheet to the neighbour. Next round, you read what arrived and write three more — new ideas, or better versions of what your colleague started. After six rounds every sheet has visited everyone: 6 × 3 × 6 = 108 ideas in thirty silent minutes.

Watch one idea travel: round 1, Priya writes "snap-fit cover". Round 2, Marek reads it and adds "snap-fit with moulded-in hinge — lid never gets lost". Round 3, Chen adds "make the hinge the cable guide too — deletes a clip". One seed became a part-count reduction nobody had at the start. Because it is silent, the loudest voice adds exactly three ideas per round — the same as the quietest graduate, who often knows the product best. Use 6-3-5 to open every creative session: it guarantees volume, equality and written capture before any discussion starts.

5.6 · SCAMPER — the mutation checklist

SCAMPER is seven questions you fire at an existing design, one letter at a time: Substitute, Combine, Adapt, Modify or magnify, Put to other use, Eliminate, Reverse. It is mechanical on purpose — when the room is tired and the flow has stopped, the checklist keeps producing. Ask each question about the part, the material, the process, even the packaging. The strongest letter in VE is E — eliminate: "what if this part simply didn't exist? which neighbour could do its job?" Try each letter on a torch in the explorer below, then fire the same questions at your own product.

5.7 · TRIZ in twenty minutes, not twenty days

TRIZ's core claim, distilled from studying millions of patents: inventive problems are contradictions, and the same ~40 principles resolve them across every industry. A technical contradiction is "improving A worsens B" — stiffer but heavier, faster but hotter. Look up the two parameters in the contradiction matrix and it returns the principles inventors historically used. Example: strength vs. weight of a moving part points to Principle 1 (Segmentation — ribbed or hollow structures), 8 (Anti-weight — counterbalance), 15 (Dynamics — make it adaptive) and 40 (Composite materials). A physical contradiction is "A must be X and not-X" (hot AND cold, present AND absent) — resolved by separation: in time (hot during brazing, cold in use), in space (rigid here, flexible there), or by condition. And the north star is ideality — the Ideal Final Result asks: "how would this function perform itself, with no part, no cost, no harm?" Snap-fits are the IFR of fastening; that question alone generates ideas.

5.8 · Morphological analysis — the combination machine

When a product has several sub-functions, break it into them and list every known solution for each — then combine. For the torch: provide light (LED array, single high-power LED, laser diode, electroluminescent panel), store energy (AA cells, rechargeable li-ion, supercapacitor, hand-crank dynamo), control output (push switch, twist bezel, proximity sensor). That little table already holds 4 × 4 × 3 = 48 possible torch concepts — including combinations nobody has built, like "single LED + hand-crank + twist bezel" (the emergency torch that never needs batteries). Most cells combine into nonsense; you are hunting the three or four fresh pairings worth carrying into evaluation. Morphological analysis is the slowest technique in the kit and the most systematic — use it when the function structure is rich and you suspect the winning combination has simply never been tried.

5.9 · Analogy and biomimicry — steal from other worlds

Someone, somewhere, already delivers your function for less — often in a different industry, sometimes in nature. The questions are simple: "which industry does this function at one-tenth of our cost?" and "how does nature do it?" The track record is famous. Velcro came from burrs stuck to a dog's fur. The Shinkansen bullet train's nose copies the kingfisher's beak — a bird that enters water without a splash — which cut the train's tunnel boom and its energy use. Drag-reducing surfaces copy shark skin. And at workshop scale: a £2 toy transmits torque with a moulded polygon and no key at all; a drinks carton "contains fluid" for pennies; a bicycle brake modulates huge friction forces with stamped parts. Pick your function, ask "who else has this problem, but cheaper?", and spend twenty minutes in their world.

5.10 · AI-assisted ideation — the newest pump

The newest tool in the kit: prime an AI assistant with the FAST model, the BOM and the constraints, and it will generate and cluster hundreds of candidate ideas in minutes — including cross-industry solutions no one in the room has seen. A working prompt looks like: "Here are our functions and costs. For the function 'dissipate heat', currently a £2.10 aluminium heatsink: list 20 alternative ways to deliver this function, grouped by physics, including approaches from consumer electronics, automotive and aerospace." The division of labour is clear: machines diverge, humans judge. AI raises the floor of idea volume; the team still owns feasibility, novelty-checking and selection.

5.11 · The techniques are a toolkit, not a menu

Sequence them: open with brainwriting 6-3-5 (silent, equal, fast volume), move to function-prompted brainstorming to build on the written seeds, apply SCAMPER to the stubborn functions, bring TRIZ when a genuine contradiction blocks progress, use morphological analysis when a rich function structure hides untried combinations, and close with analogy rounds ("how does a £2 toy solve this? how does nature?"). Quotas help: "ten more ideas for 'contain water' before lunch" reliably produces eight mediocre ideas and two gems — which is exactly the deal you want.

6.1 · The morning after — 380 sticky notes and a decision to make

The creative day went well. The wall holds 380 ideas: brilliant ones, silly ones, duplicates, and a few that would get someone fired. Now what? Two failure modes wait for you here. Analysis paralysis: trying to deep-study all 380, running out of steam by idea 60, and quietly picking the familiar ones. Gut-feel triage: the chief engineer walks the wall and keeps what they already liked — which makes the whole creative phase theatre. The Evaluation Phase is the escape from both: a funnel of four passes, each cheap and fast at the top, each more careful as the list shrinks. Roughly one idea in eight survives to development — and that ratio is healthy, not disappointing.

6.2 · Pass 1 — triage: go, grow or park

First pass is fast and shallow — about a minute per idea, done as a group. Three piles only:

• Go — clearly feasible, clearly valuable. "Replace six screws with two snap-fits" goes straight through.
• Grow — promising but immature. "Some kind of integrated hinge…?" needs shaping before it can be judged. It advances, flagged for strengthening, often by combining it with its neighbours.
• Park — not now. Wrong timing, wrong volumes, needs a platform change. Parked is not deleted: the parked list is the seed stock for next year's study, and half of today's "obvious" winners were parked in a previous wave.

Two rules keep triage honest. Ideas are judged as written, kindly — triage kills the unworkable, not the unfamiliar. And nobody says "that will never work here" — the only legal kill reasons at this stage are physics, safety or law. An hour of triage typically cuts 380 ideas to about 120.

6.3 · Pass 2 — group, merge, and knock out

Before any scoring, group the survivors by theme — fastening, material, architecture, packaging. Duplicates merge (eight versions of "use fewer screws" become one strong idea with eight authors), and half-ideas combine into whole ones: "snap-fit cover" plus "moulded hinge" plus "cable guide in the lid" is one integrated-cover concept worth three times any single note. Then apply the knockout screens — the non-negotiables that no score can trade against: safety and regulatory compliance, brand-defining requirements, contractual customer commitments. A £3 saving that violates a homologation rule is not a low score; it is out. Grouping typically leaves 60–80 distinct, screened concepts.

6.4 · Pass 3a — the effort–impact grid, in ten minutes

Now the first ranking tool — deliberately crude. Two questions per idea: how big is the prize? (impact: savings, quality, weight) and how hard is it to get? (effort: engineering hours, tooling cost, validation time). Plot every idea on the 2×2:

• Quick wins (high impact, low effort) — do these first; they fund the programme's credibility. Example: delete the second label nobody reads — £0.04 per unit, one drawing change.
• Big bets (high impact, high effort) — the strategic projects. Example: integrate the bracket into the housing moulding — £0.80 per unit but new tooling and full revalidation.
• Fill-ins (low impact, low effort) — do them when convenient, batch them into running changes.
• Money pits (low impact, high effort) — park them, politely, forever.

The grid takes ten minutes with sticky notes and instantly shows the shape of your programme: a wall full of fill-ins means the creative phase stayed too timid; a wall full of big bets means this year's savings target is in trouble.

6.5 · Pass 3b — the weighted decision matrix, with real numbers

For the ideas that matter — the quick wins and big bets — move from crude to explicit. A weighted matrix has five steps. One: agree the criteria (four to six, no more). Say: net savings, technical risk, time to implement, investment. Two: weight them to total 100% — say 40 / 30 / 20 / 10. Three: score every concept 1–9 on every criterion. Four: multiply and add. Five: read the ranking — then argue with it, because the argument is the value.

Worked run on "transmit torque": the spline shaft scores 8 on savings, 7 on risk (proven), 9 on speed, 8 on investment → weighted 7.9. The integrated forging scores 9 on savings but 4 on risk, 3 on speed, 4 on investment → weighted 5.8. The spline wins today. But watch the sensitivity: give risk 15% and savings 55%, and the forging jumps ahead. That is not a flaw — that is the matrix telling you the decision depends on your appetite, which is exactly the conversation the sponsor needs to have. Try it live in the lab below.

6.6 · Criteria that don't lie

The matrix is only as honest as its criteria, so fix them before anyone sees the idea list — criteria chosen afterwards get bent toward someone's favourite. Keep them to four to six; beyond that, weights turn into noise. Watch for double counting: "annual savings" and "payback" both measure money — pick one and pair it with investment. Make finance own the savings scale, so Phase 6 cannot be ambushed with "those numbers aren't real". And never score your own idea — authors swap concepts with a neighbour.

6.7 · Pass 3c — Pugh: when scores are false precision

Sometimes you cannot honestly score 1–9 — early concepts are too foggy. The Pugh matrix trades precision for honesty. Pick a datum — normally the current design — and judge every concept against it, criterion by criterion, with just three marks: better (+), same (S), worse (−). Datum: the current keyed shaft. The spline runs +cost, +assembly, S-risk, −tooling → net +1. The plastic overmould runs ++cost, −risk, −validation → net 0, but with the biggest single upside on the board. Pugh's real power is the second pass: hybridise. Ask of every minus, "can this concept borrow a plus from another?" — can the spline's tooling problem take the overmould's approach? Concepts converge upward instead of merely being ranked. Use Pugh for early concept selection; use the weighted matrix when the numbers deserve trust.

6.8 · Pass 4 — pressure-test the winners

The short list — maybe fifteen concepts — gets one last cheap test before development spends real money on it: what could go wrong, and would we see it coming? For each winner ask three questions. Does it touch a safety or homologation boundary? (If yes, the validation cost just tripled — rescore it.) Does it depend on one supplier, one machine or one person? (Single points of failure need a plan B priced in.) And what is the reversal cost if it fails in the field? A £0.30 saving with a recall-shaped downside is not a saving. This is a ten-minute FMEA mindset, not a full FMEA — the point is that no idea reaches Phase 5 carrying an invisible showstopper.

6.9 · From ranked list to executable scenarios

Ideas interact. A material substitution may kill a process idea — or enable it. So run a quick compatibility check, then bundle survivors by implementation vehicle: running changes this year, the mid-cycle refresh, the next-generation platform. Many teams present the bundles as three scenarios — conservative / balanced / stretch — so the sponsor chooses an ambition level, not a list of parts. Finally the honesty test: count the engineering hours the scenarios need against hours that actually exist. A shortlist beyond your capacity is a wish list — better to implement eight ideas than approve thirty.

6.10 · Champions — because unowned ideas are already dead

The phase ends with names, not lists. Every surviving concept gets a champion: one person who carries it into development, reports its status, and fights its corner when priorities collide. Not a committee — a name. The champion gets the idea's file (the function it serves, its score, its risks, its worth), a date for the Phase 5 development gate, and the authority to recruit help. Experience is blunt here: ideas with champions get developed; ideas assigned to "the team" appear again, untouched, in next year's workshop.

7.1 · From idea to proposal

Champions convert shortlisted ideas into implementable value proposals. Each answers five questions: what exactly changes, what it saves (validated by finance), what it costs to implement (tooling, validation, engineering hours), what could go wrong, and when it lands. An idea with those five answers is a business decision; without them it's still a sticky note.

7.2 · Engineering validation

De-risk before you commit: CAE simulation to verify a thinner wall or relaxed safety factor, design FMEA to expose new failure modes, a DVP&R (design verification plan & report) scaled to the change's risk. For sourcing ideas, a should-cost model or supplier quotation validates the savings claim. Regulated industries add the change-control question early: does this need re-certification, re-validation, PPAP?

7.3 · The business case

The universal arithmetic: annual saving = per-unit saving × volume; payback = one-time cost ÷ annual saving. Most running-change proposals should pay back in under 12 months; well-run VE studies return better than 10:1 overall. Present savings honestly: gross vs. net of implementation cost, and dated — a saving that lands in 18 months is worth less than one landing this quarter.

7.4 · The validation ladder

Match rigour to risk, not to habit: desk calculation → CAE/simulation → rig test → line trial → field pilot. A packaging spec change may need one step; a safety-factor reduction needs all five. Two rules: climb only as high as the risk class demands (over-validation quietly kills payback), and start long-lead items (tooling, supplier qualification) in parallel behind clear kill-gates rather than in series after every test passes.

7.5 · FMEA in ten minutes

Every VE change can create failure modes the current design never had. Score each candidate mode on Severity × Occurrence × Detection (1–10 each; the product is the RPN) — but score the delta against the current design, not absolutes: a thinner wall that raises occurrence from 2 to 4 on a severity-8 mode needs an action; one that goes 2 to 3 on severity-2 needs a note. High-severity or regulated systems escalate to a full cross-functional DFMEA — that's not bureaucracy, it's what makes the saving safe to bank.

7.6 · Money mechanics finance will sign

Net saving = gross piece-cost saving − amortised one-time costs (tooling, validation, engineering hours, scrapped inventory, requalification). Payback = one-time ÷ annual net — the right metric for stable-volume running changes; use NPV at the company hurdle rate when volumes ramp or decay. And date everything: savings start at the implementation date, not the approval date — a proposal without a date has a business case of zero.

8.1 · The decision meeting

Present to the sponsor and decision board: the value logic (functions and mismatches), the numbers (savings, cost, payback), the risks and their mitigations, and the asks (resources, decisions, dates). The output is a decision log — explicit go/no-go per proposal. A presentation that ends without decisions is a rehearsal, not a Phase 6.

8.2 · The savings funnel

Approved ideas enter a staged funnel: L1 Idea → L2 Validated → L3 Approved → L4 Implemented. Each stage has entry criteria (L2 requires finance-verified savings; L4 requires the change in production with PPAP/validation complete). The funnel is reviewed monthly: every idea has an owner, a stage and a date — or it is killed. Savings only "count" at L4, audited against the frozen baseline so inflation and mix changes can't blur the result.

8.3 · Close the loop

Post-implementation, feed the lessons back: update design guidelines, cost standards, preferred-parts lists and the idea bank. This institutional memory is what makes wave two cheaper and faster than wave one — and what eventually turns VE from a project into a capability.

8.4 · The one-pager that wins decisions

Boards decide on one page per proposal: the function and its value mismatch → what changes (picture beats paragraph) → net saving, one-time cost, payback → top risk and its mitigation → the ask, the owner, the date. Then pre-wire: walk the two most sceptical stakeholders through it before the meeting; surprises make executives defensive, and defensive executives say no. The meeting itself produces a decision log — proposal, verdict, owner, date, signature. Applause is not an output.

8.5 · Savings accounting that survives audit

Rules agreed in Phase 0, applied ruthlessly here: measure against the frozen baseline; neutralise commodity and volume swings (index-adjust so a copper spike doesn't erase a real design saving); report the gross → net bridge openly; count only at L4 — implemented in production and audited by finance; and reconcile with purchasing's ledger so the same pound isn’t claimed twice. Every credible programme's reputation rests on this paragraph.

8.6 · Closing the loop

Feed every implemented idea back into the system: update design rules (so the next programme never re-adds the deleted bracket), refresh cost standards with the new should-costs, tag the idea bank (parked ideas seed the next wave), and run a 30-minute retro per wave: what stalled, why, and which stage of the funnel leaked. Wave two should always be cheaper and faster than wave one — if it isn't, the loop isn't closing.

8.7 · Where ideas go to die — the five killers

Here is the uncomfortable industry truth this phase exists to fight: in weakly-governed programmes, 40–60% of approved workshop ideas never reach production. They don't die dramatically — they fade. The five killers, and their designed-in antidotes: no owner (an idea assigned to "the team" belongs to nobody — one named champion per idea, set in Phase 4); no funded capacity (approval without engineering hours is a wish — the decision meeting books the hours, not just the intent); the change-control surprise (requalification, PPAP or re-certification costs surface late and kill the payback — ask the change-control question in Development, not after approval); baseline drift (nobody can prove the saving landed, so finance stops believing — freeze the baseline in Phase 0); and the champion leaves (single-point knowledge walks out the door — the funnel review reassigns orphaned ideas within a month). None of these is bad luck. All five are governance failures with known fixes — which is why the monthly funnel review, boring as it sounds, is where programmes are actually won.

9.1 · Why should-cost — one number against the truth

Every purchased part arrives with an information problem. The supplier knows their material buy price, their cycle times, their scrap rate, their plant utilisation and their margin. You know one number — the quote. Negotiating from that position is guessing: push too little and you overpay for years; push too hard, blind, and you force a good supplier below water and inherit their quality problems. Should-costing restores the balance. By rebuilding the part's cost from physics and public market rates — what the material must weigh, what the machine must cost per hour, what a fair margin looks like — you walk into the negotiation knowing roughly what the supplier knows. The conversation changes instantly: from "give us 5%" (arbitrary, resisted) to "your quote implies a 55%-utilised press shop — is that right?" (specific, answerable). One is haggling; the other is engineering.

9.2 · Six lever families

• Product design — part-count reduction, DFMA, material substitution, tolerance optimisation, de-speccing, platforms, safety-factor right-sizing
• Materials — grade optimisation, recyclates, near-net-shape blanks, nesting & yield, scrap monetisation
• Manufacturing — process substitution, automation, cycle-time/OEE, yield, tooling strategy, make-vs-buy, footprint
• Sourcing — should-cost negotiation, LPP, competitive RFQs, bundling, best-cost-country, supplier VAVE, indexation
• Packaging & logistics — pack spec, returnables, cube utilisation, mode shift
• Complexity & lifecycle — SKU rationalisation, commonality, warranty-cost design

Rule of thumb: where the cost sits decides which family bites. BOM-heavy products (like most automotive) → sourcing + design; labour-heavy → DFMA + automation; volatile-material products → indexation first.

9.3 · Anatomy of a should-cost

A quoted price is one opaque number; a should-cost model decomposes it into negotiable layers: raw material (mass × market price ÷ yield), process & machine (cycle time × machine-hour rate), direct labour, scrap & yield, overhead (driven by plant utilisation), SG&A, and a fair margin. Each layer is built from physics and market rates, and each is a separate conversation with the supplier. Fact-based negotiation with a cleansheet typically recovers 5–15% on quoted prices — without squeezing the supplier's legitimate profit.

9.4 · Linear Performance Pricing

When you buy a family of similar parts, regress price against the dominant cost driver (mass, power, area). Parts above the regression line are paying more per unit of driver than their siblings — a ready-made negotiation target list, produced from data you already own.

9.5 · Building a cleansheet in seven steps

• 1 · Scope — part drawing, material spec, annual volume, sourcing region. Cost is meaningless without these four.
• 2 · Material — net mass ÷ process yield = gross mass, × market price from a commodity index (never a list price).
• 3 · Process route — the operations a competent maker would use, with cycle times from physics and machine specs, not folklore.
• 4 · Machine rate — (depreciation + energy + floorspace + maintenance) ÷ (annual hours × OEE). Utilisation assumptions move this ±30%.
• 5 · Labour — operators per machine × regional loaded wage; automation level is the design variable.
• 6 · Overhead & SG&A — regional percentage benchmarks, sanity-checked; the layer where idle plants hide.
• 7 · Margin — a fair commodity-typical profit. Then sanity-check the total against LPP curves and real quotes.

9.6 · LPP with real numbers

Plot your eight stamped brackets: price (y) against mass (x). Regression: price ≈ £0.85 + £1.9/kg. Six parts hug the line; bracket D sits 18% above it, bracket F 25%. Before celebrating, check specs — F has a special coating (justified); D doesn't (target). One hour with data you already own produced a negotiation list worth more than a month of meetings. That's LPP: cheap, fast, and brutally objective.

9.7 · Negotiation choreography

Never email a cleansheet with "explain the gap" — that's how you get a defensive supplier and a worse relationship. Instead: share the method, invite them to a joint workshop, walk the layers together, and split the conversation three ways — raw material gets indexed in the contract (nobody negotiates the copper price), conversion gets engineered together (cycle times, yields, packaging), margin gets respected. Close with a gain-share clause so their ideas keep coming. Suppliers who feel audited hide information; suppliers who feel partnered volunteer it.

9.8 · AI in should-costing — from days to minutes

The seven layers haven't changed, but the speed of building them has. Feature-based AI engines read a 3D CAD model, recognise its manufacturing features, pick a feasible process route and return a regional should-cost in under a minute — Module 11 opens the engine up in detail. Spend analytics runs LPP automatically across an entire commodity, flagging every outlier instead of the eight parts you had time to plot. Large language models draft the negotiation fact pack — the layer table, the questions, the counter-arguments — from your cleansheet in minutes. What AI does not change: the model is only as honest as its assumptions, and a negotiation is still won by the engineer who understands the physics behind every layer. Master the manual cleansheet first — this module — and the AI tools become a force multiplier instead of a black box you can't defend across the table.

10.1 · Why teardown is VE's mirror

Phase 2 taught you to ask "what is this function worth?" — and the honest answer is hard to find from inside your own company. Your cost systems tell you what things do cost, not what they could cost. Your engineers defend designs they created. Your suppliers quote what the market will bear. The one source of evidence nobody can argue with is a competitor's product delivering the same function for less — sitting on your bench, taken apart, weighed and costed. That is what teardown provides: not opinion, but physical proof of what is possible.

And here is the liberating part: every product on the market is an open textbook. The moment a product is sold openly, anyone may buy one and study it — an entire industry exists to do exactly that. Your competitors are almost certainly reading your products this way. The only question is whether you read theirs.

10.2 · What it costs, and what it returns

Teardown feels expensive — buy competitor products just to destroy them? Do the arithmetic before deciding. A mid-size teardown study — two competitor units, four weeks of a small team, the lab time — might cost £30,000 to £60,000 all-in. If it surfaces ideas worth even £1.50 per unit on a product built at 200,000 units a year, that is £300,000 every year — a five-to-tenfold return in year one, repeating annually. This is why carmakers pay millions for subscription teardown databases: the economics are not close. Choose targets deliberately: the market leader (what does "best" cost?), the price disruptor (how is it so cheap?), and the closest rival (where exactly do we lose?). One of each teaches more than three of the same.

10.3 · Step 1 — Acquire and document

The units arrive. The most important rule of the whole process applies now, before any tool comes out: you get exactly one first teardown. Every state you change is gone forever. So the first day is cameras and scales, not screwdrivers: photograph every face and label, weigh the complete product, measure the overall dimensions, record serial numbers, build dates and market specification. Professional labs photograph against a neutral background with a scale reference in every frame, so any measurement can be re-taken from the photo months later. Boring? Completely. But every argument the study will ever face — "are you sure that's the EU spec?", "was that seal already broken?" — is settled by this archive.

10.4 · Step 2 — Systematic disassembly, against the clock

Disassembly is not demolition — it is assembly observed in reverse. Work level by level: closures off, interior out, systems separated, subassemblies opened, exactly as the exploded view above shows. And capture time for every operation, because the stopwatch is reading their factory: a product that comes apart in 48 seconds went together in roughly a minute of paid labour, and one that needs 92 seconds and three tool changes cost its maker twice as much to assemble. Log every fastener (count, type, size — each unique fastener is purchasing complexity), every clip, every "why is this here?" moment. The disassembly log is a competitor's process sheet, reconstructed.

10.5 · Step 3 — Digital BOM capture

Now the wall of parts becomes data. Every component is logged against a schema agreed before the first screw turned: part ID, assembly level, mass, material (verified with a spark test, XRF gun or burn test — never guessed), manufacturing process evidence (parting lines, tool marks, draft angles tell you mould counts and machine types), supplier markings, fastener count, disassembly time. Photograph every part on the lightbox. This digital BOM is the asset the whole study rests on — findings fade, opinions argue, but a complete costed BOM of a competitor product keeps answering questions for years. Keep the physical wall of parts too: both products laid out side by side, level by level, teaches more in an hour than the spreadsheet does in a week.

10.6 · Step 4 — Should-cost their parts, honestly

Cleansheet the competitor BOM using Module 9's method — but at their assumptions, not yours: their likely manufacturing region, their volumes (estimable from sales data), their process choices as evidenced by the parts themselves. Band every estimate honestly (±15%) and document the assumptions. The output that matters is the delta table: their door hinge £1.10 versus yours £2.30 is a robust finding even though both numbers carry error bars — deltas survive uncertainty that absolutes don't. Rank the deltas, and the study's priorities write themselves.

10.7 · Step 5 — Harvest and transfer

A teardown that ends with a report is called teardown tourism. The real product is ideas in your funnel: every meaningful delta becomes a creative-phase seed ("they deliver this function with a snap-fit — what would that take on our platform?"), every clever solution gets an adopt / adapt / leapfrog decision, and the quantified findings feed the cost walk from Module 7 — current cost on the left, target on the right, teardown-evidenced blocks in between. The bench work is finished only when the findings have owners, dates and a place in the savings funnel.

10.8 · Six dimensions — benchmarking is bigger than piece cost

• Cost — the delta table: where do they spend less, part by part? The £1.20 hinge gap.
• Functional — performance per pound: torque/£, lumens/£, litres of boot space per £. The value ratio, not the absolute.
• Design — architecture choices: their one-piece moulded chassis versus your four-part welded one; integration versus modularity.
• Process — what the parts say about how the best plants make them: weld counts, tool marks, cycle-time clues.
• Feature — feature sets against price ladders: who over-serves the market, who charges for what customers ignore (Kano, Module 3, applied to rivals).
• Patent / IP — expired patents are free ideas; active ones are constraints to design around. Both are readable in the parts.

Price alone tells you almost nothing — a cheaper rival might be buying share with negative margin. Six lenses together tell you why the numbers are what they are, and benchmarking answers VE's hardest question — what is a function worth? — with the best evidence there is.

10.9 · Running the physical lab

You need less than you think: a solid bench, calibrated scales, torque drivers, callipers, a lightbox, and the agreed data schema. Photograph every state before changing it; scan or CT anything you may want to measure after destructive steps — those are one-way doors. Label bags and bins by assembly level from the start: the three hours of discipline saves three weeks of "which bracket was this?" later.

10.10 · AI in the teardown lab

Teardown is being transformed by AI faster than almost any other VE activity, because it is drowning in exactly the data AI is good at. Computer vision now recognises and classifies photographed parts, pre-filling BOM lines that an engineer only verifies. CT scanning with AI segmentation digitises a complete assembly without touching a screw — internal geometry, wall thicknesses and material boundaries extracted into a 3D model, so the "one first teardown" can be non-destructive. Large language models classify thousands of captured BOM lines into commodity groups and draft the finding narratives. Auto-should-cost engines take part geometry and spit out cycle times and cost estimates in minutes rather than days. And digital teardown libraries mean you benchmark twenty vehicles from a laptop before buying one. The division of labour is the same as ever: machines count and estimate; engineers judge what it means and what to do about it.

10.11 · Ethics and legality — the bright lines

Teardown of products bought on the open market is lawful, standard industry practice. The bright lines: no misappropriated confidential information — a competitor's drawing from a shared supplier is poison; refuse it and say why. Patents are not secrets — they are public documents you should actively read — but copying a patented solution needs a licence or a design-around. Learn freely, copy carefully, and involve counsel when adopting, not when analysing.

11.1 · The digital cost stack

• AI should-cost engines — feature-based tools read 3D CAD, simulate the manufacturing process and output cycle time, tooling and piece cost per region before any supplier quote
• CT-scan & 3D teardown — industrial computed tomography digitises competitor products non-destructively: wall thicknesses, hidden joints, internal architecture
• Spend cubes — classified spend joined with BOM data exposes price variance for identical parts across plants and suppliers
• Generative design & topology optimisation — algorithms explore thousands of geometry variants at minimum material
• Additive manufacturing — part consolidation, tool-less low volumes, conformal-cooled tooling
• Digital twin / CAE — simulation-validated margin removal; virtual DOE replaces physical trials

11.2 · LLMs in the workshop

Large language models now assist across the job plan: mining warranty text and quotes in the Information Phase, generating and clustering ideas in the Creative Phase, drafting business cases in Development, and powering cost knowledge graphs that link functions ↔ parts ↔ costs ↔ suppliers so no lesson is lost between waves. The practical stance: AI compresses analysis and divergence; humans own judgement, validation and decisions. Teams that treat AI output as a starting draft move roughly twice as fast; teams that treat it as an answer ship mistakes faster.

11.3 · What doesn't change

No tool replaces the discipline: functions before solutions, worth before ideas, validation before implementation, P&L before applause. The stack multiplies a good method — it cannot rescue a skipped one.

11.4 · Inside an AI should-cost engine

Feature-based engines work in four stages: geometry extraction (the CAD model is parsed into manufacturing features — holes, bends, bosses, pockets), routing selection (the engine picks a feasible process chain for the material and geometry), physics simulation (cycle times computed from tonnage, melt volumes, tool paths), and regional economics (machine rates, wages and overheads from maintained data libraries — aPriori, for instance, maintains 92 regional cost libraries). Two consequences: quality-in decides quality-out (a sloppy CAD model costs wrongly), and outputs are directional — perfect for ranking design options and preparing negotiations, never a substitute for the final quote.

11.5 · Prompt patterns that work in VAVE

The reliable ideation pattern: role + function + constraints + quantity + ranking — "You are a cost engineer. Generate 25 ways to 'transmit torque' in a 40 Nm automotive application, manufacturable at 200k/year, ranked by likely unit cost; flag any that typically fail durability." Use the same structure to cluster teardown notes, draft business cases, and translate warranty text into function language. Two hard rules: verify every number the model outputs (LLMs are fluent, not calibrated), and never paste confidential BOMs or drawings into public tools — use your company's approved instances.

11.6 · Cost and carbon — one lever, two savings

The same physics that drives cost drives CO₂e: mass, energy intensity, scrap, logistics distance. Add a kgCO₂e column to the cleansheet and most cost levers reveal a carbon dividend — lightweighting, yield improvement, recycled content, nearshoring. With carbon border mechanisms and customer scope-3 reporting spreading, the VE study that quantifies both is answering next year's question this year.

11.7 · AI across the job plan, phase by phase

• Phase 1 · Information — LLMs cluster thousands of warranty claims and service notes into function-language Paretos; spend analytics builds the cost baseline in days, not weeks
• Phase 2 · Function Analysis — a drafted function tree from the BOM in minutes; the team's job shifts to correcting it, which is faster and sharper than starting blank
• Phase 3 · Creative — 25 ranked ideas per function as a warm-up before the human session; cross-industry analogy search on demand
• Phase 4 · Evaluation — auto-clustering of duplicate ideas; first-pass feasibility flags; the humans keep the scoring
• Phase 5 · Development — drafted business cases with the arithmetic pre-filled; AI should-cost for instant savings estimates per idea
• Phase 6 · Presentation — live funnel dashboards; auto-drafted decision logs and savings-audit trails

Notice the pattern: in every phase AI does the drafting, counting and clustering, and the team does the judging, deciding and owning. The phase gates, the decision rights and the audit trail stay exactly where the six-phase job plan put them.

11.8 · Sequencing your own stack — crawl, walk, run

Crawl — start with the data you already own: a spend cube and LPP across your top commodities needs no new tools, finds outliers in weeks, and funds everything that follows. Walk — pilot an AI should-cost engine on your top 50 purchased parts; measure it against real quotes for a quarter before trusting it; add a digital teardown-library subscription if you benchmark regularly. Run — approved LLM copilots wired into the workshop itself, CT scanning for non-destructive teardown, and a cost knowledge graph linking functions ↔ parts ↔ costs ↔ suppliers so no lesson is lost between waves. The failure mode to avoid is buying the "run" stack first: tools amplify a working method — bolted onto no method, they produce beautifully-formatted guesses. Sequence capability first, software second.

11.9 · Guardrails — keeping AI-assisted cost work auditable

Four rules make the difference between a force multiplier and a liability. Verify every number — LLMs are fluent, not calibrated; any figure that reaches a business case gets a human check against a source. Protect the data — BOMs, drawings and supplier prices never enter public tools; use your company's approved instances, and treat a pasted drawing like an emailed one. Keep a human signature — every AI-drafted business case, should-cost and decision log carries a named engineer who checked it; "the model said so" survives no audit. Watch for drift — re-benchmark AI cost estimates against real quotes on a cadence, because rate cards age and models change silently. Teams that adopt these four rules early move fast and stay credible with finance — which, as Module 8 taught, is the only speed that counts.

12.1 · From event to operating system

One-off studies decay; programmes compound. The operating system: a monthly funnel review (owners, stages, dates), quarterly wave planning (next products, next teardowns), an annual value roadmap per product line tied to margin plans, and continuous capability building. KPIs that keep it honest: funnel coverage (% of product cost under active study, target ≥60%), idea conversion rate, cost transparency (% of BOM with a should-cost, target ≥80%), savings velocity, average payback, and the reuse index.

12.2 · What to expect

A first structured wave on an unworked product typically finds 8–15% of product cost; mature programmes sustain 3–5% annually; study ROI routinely exceeds 10:1. The biggest failure mode is not weak ideas — it is weak follow-through, which is exactly what the funnel governance exists to prevent.

12.3 · Standards and certification

The SAVE International Value Methodology Standard defines the job plan used worldwide; EN 12973 codifies Value Management in Europe. Practitioners certify through SAVE as VMA (Value Methodology Associate — entry level), AVS (Associate Value Specialist) and CVS (Certified Value Specialist — the professional benchmark for leading studies). TRIZ has its own certification levels; both credentials signal a practitioner who runs the method, not just the meeting.

12.4 · Operating models that scale

Three archetypes: a central CoE (deep expertise, tools, standards — but becomes a bottleneck), fully embedded champions in each product line (scales, but skills dilute), and the hybrid that wins in practice — a small central value office owning method, training, tooling and the savings ledger, with certified champions embedded in every line. Report it neutrally: a value office under engineering optimises design and ignores sourcing; under purchasing, the reverse. CFO or COO sponsorship keeps both levers honest.

12.5 · Targets, cascades and incentives

Cascade market-back: product-line margin plans → annual cost roadmaps → subsystem targets at design gates — cost as a spec with the same authority as mass. Incentives need care: reward audited, net, implemented savings (or people will game gross paper numbers), credit idea originators as well as implementers, and never let the savings target exceed the engineering capacity you actually funded — a target without capacity is a morale programme, in reverse.

12.6 · Growing practitioners deliberately

Capability ladders beat hero dependence: awareness (everyone — this course's job) → practitioner (runs analysis in their own product) → facilitator (leads studies; each wave co-facilitated by a learner) → certified specialist (VMA → AVS → CVS through SAVE International). Add a community of practice that meets monthly and an internal "value day" each year where teams present implemented ideas — recognition is the cheapest retention tool a programme has.

13.1 · Prevention beats correction

Everything so far in this course removes cost that already exists. This module is about never letting it in. By the time a concept design is frozen, 70–80% of lifecycle cost is committed while almost none has been spent — so the cheapest VAVE workshop is the one the target made unnecessary. Target costing (Toyota's genka kikaku, developed from the 1960s onward) inverts the traditional logic. Cost-plus asks: what will it cost, and what price does that force? Target costing asks: what will the market pay, and what cost does that allow?

The equation is disarmingly simple: Allowable cost = target market price − required margin. Everything hard about target costing is organisational, not arithmetic: holding the line when engineering says "it can't be done for that", and cascading the number so every team owns a piece of it.

13.2 · Allowable cost, drifting cost, and the value gap

Three numbers run the process. The allowable cost comes from the market, top-down. The drifting cost is the current bottom-up estimate of what the design as drawn would actually cost — it moves with every design decision. The distance between them is the value gap, and closing it is a design activity: function analysis, DFMA, spec challenge, supplier co-design — the entire toolkit of Modules 4–11 pointed forward instead of backward.

Discipline comes from the cardinal rule (Cooper & Slagmulder): the target cost may never be exceeded. If a feature must be added, its cost must be found elsewhere — a different feature trimmed, a design simplified, a make-buy changed. What the rule really bans is the quiet drift where every team adds "just 2%" and the product arrives 15% over. If the gap truly cannot close, the honest moves are to change the product's content or not launch — not to pretend the margin will appear later.

13.3 · The cascade: from vehicle to part to supplier

A single top-level number is a wish. A cascade is a plan. The vehicle-level allowable cost is decomposed to systems (body, powertrain, chassis, electrical…), weighted by function value and benchmark data, then to subsystems and parts, and finally into supplier target prices — shared openly with sourcing partners along with the cost model behind them. Two practices make the cascade survive contact with reality: give each level a small unallocated management reserve (concepts change), and make every target's owner a named engineer, not a department.

13.4 · Design-to-cost: run cost like weight

Aerospace programmes track mass with an owner per subsystem, a live status at every gate, and a recovery plan the moment an allocation is exceeded. Design-to-cost applies exactly that machinery to cost: cost is a specification with the same authority as mass, performance or safety. At each design gate the drifting cost is reported against the target; a gate with a red cost status and no recovery plan does not pass. Practical instruments: trade-off curves (cost vs. performance per concept, so decisions are chosen, not discovered), tolerance–cost curves (tightening a tolerance typically raises machining cost exponentially — challenge every decimal place), and a cost model that updates with the CAD, so engineers see the £ consequence of a design choice the week they make it, not at the next quote round.

13.5 · After SOP: the kaizen handoff

Target costing ends at start of production; kaizen costing picks up from there — continuous, incremental reduction against a yearly cost-down target on the running product. The two are one system: target costing decides where the cost curve starts; kaizen costing bends it further down. A mature product organisation runs both, plus periodic VAVE waves (Module 12) when step-change is needed. One warning: kaizen targets applied blindly, year after year, are how quality death-spirals begin — audited function and quality gates apply to cost-downs exactly as they did in the Development Phase (Module 7).