Optimisation, before it locks.
Also called: DFM review · Value-engineering pass · Design refinement · Cost-down loop
Refining an engineered design to hit critical targets, cost, performance, assembly, manufacturability, before it locks for production.
Optimisation is a refinement loop, not a one-off. You take a working design and trim cost, assembly time and waste against the targets that matter, while protecting the one number that must not move. Stop when the gains go small and the risk grows.
What this activity is
Optimisation is the discipline of taking a design that already works and making it better against the targets that decide whether the product is viable. By the time you reach it, the engineering is sound: the thing performs, it is safe, it can be made. The question optimisation answers is sharper. Can it be made cheaper, assembled faster, shipped lighter, or built with less waste, without losing the performance that justifies the price? It is a deliberate pass, run against named targets, not a vague tidy-up.
The key word is loop. You do not optimise once and declare victory. You pick a target, make a change, measure the effect on cost and on the performance you care about, and decide whether to keep it. Then you do it again on the next target. Each pass trades something: a cheaper component might cost you a tolerance, a faster assembly might cost you a fixing. The work is holding those trades in view at once, rather than chasing a single number into a corner. In my experience the discipline is knowing which number is allowed to move and which one is not.
The number you protect
Every product has one target that defines it. Move it and you no longer have the same product. Optimisation works because you name that number first, fence it off, and then push on everything around it. You let cost fall, assembly time shrink and wattage drop, but the defining performance figure stays inside its band. Cross that line in search of a cheaper bill of materials and you have not optimised the product, you have quietly redesigned it into a worse one.
How to run it: what good looks like
The clearest way to see optimisation is to watch it run against a real bill of materials. Here is the proofing box we took through the loop, so you can see the trades rather than a generic checklist.
Notice the fenced number sits at the top, not the bottom. Every row beneath it earns its keep only because the hold stayed inside ±0.5°C. Optimise without naming that line first and you save pennies while losing the product.
- Cost cut hard, with no check on the temperature hold.
- A cheaper sensor swapped in, untested against the band.
- Changes stacked with no measurement between them.
- The loop run once, then declared finished.
- The defining number fenced and measured every pass.
- One change at a time, with its effect recorded.
- Trades made explicit: cost against tolerance, speed against fixings.
- Stopped when the gains went small and the risk grew.
Where it fits
Optimisation sits inside Stage 07 Engineer, after the design is working and before it locks for production. It depends on a sound engineered design and clear targets to push against, and it feeds a tighter, cheaper, more makeable design into the standards, testing and conformity work that closes out Engineer. Skip it and you carry avoidable cost and assembly waste straight into Stage 08 Develop, where every saving you missed is now baked into tooling and far more expensive to claw back.
What it can do
It turns a design that merely works into one that works and pays. A disciplined pass can move a bill of materials from its high end toward its floor, cut minutes off every build, and trim wattage and waste, all while the defining performance figure stays inside its band. Done before lock, those savings compound across the whole production run.
What it can’t do
It can’t rescue a design that doesn’t work yet. Optimisation refines a sound thing; it does not invent soundness. If the product can’t hold its target before you start trimming, no amount of cost-down will fix that, and pushing on regardless just hides the fault under a cheaper bill of materials. Fix the engineering first, then optimise.
See the full 10-stage process →
Common mistakes
The frequent failure is optimising one number into a corner: cost falls, but the temperature hold quietly drifts and nobody measured it. The second is stacking changes without testing between them, so when the band breaks you can’t tell which change broke it. The third is treating it as a one-off rather than a loop, and stopping too early or, worse, too late, when the gains have gone small and each new change adds more risk than saving.
Not sure which target on your product is the one you must never move? Start the Free Sprint → and find the defining number before you cut a penny.
Your optimisation checklist
Project notes: trimming the bill of materials
▸ From the notebook · optional reading
Running the proofing box through the optimisation loop with Dan and Anna Hartley in Stockport, and the change we nearly kept that we shouldn’t have.
3 min read · click to open
The box worked. It held 26°C ±0.5°C, drew under 30W, and the Sourdough School testers were happy. But the bill of materials sat at the £55 high end of our range, and at a £149 price the margin was thinner than Dan and Anna wanted before a first run of 500 to 1,000 units. So we ran the loop.
Where the money was
The first pass was the obvious one. The Manchester PCB house had given us a control board and a separate sensor daughterboard joined by a small harness. Folding them into a single board removed two connectors, the harness, and the labour to fit them. The EMC behaviour had to be re-checked, and it was, but the saving was real and the hold never twitched. That alone took a useful slice off the BOM.
The change we backed out
The Stoke-on-Trent ceramic body was heavier than it needed to be, so we re-spec’d a thinner wall. Material and firing energy both fell. But a thinner wall meant less thermal mass, so I insisted we measure rather than assume. We did, and the hold stayed inside the band, just. Then someone proposed a cheaper sensor to push the BOM further. On paper it saved another pound or two. On the bench it widened the swing to ±0.7°C. We backed it straight out. That is the line: the ±0.5°C is the product, and a cheaper bill of materials that loses it is not a saving.
We stopped with the BOM down toward the £38 floor, two screwdriver steps gone from assembly, and draw still under 30W. We could have chased another fifty pence, but the gains had gone small and each new change carried more risk to the hold than it was worth. Knowing when to stop the loop is as much the skill as knowing how to run it. Then the design locked, and we moved on.
— Engineer stage, project notes, 2026
— Next stage → Stage 08 · Develop