Optioneering & optimization
Stop drawing one scheme and defending it. Generate a hundred, score them against your goals, and spend your judgement on the trade-offs instead.
A developer asked for the best layout. The right answer was 'best at what?' — and that question is your whole job.
A studio is pitching for a residential plot in Hyderabad. The developer wants maximum saleable units. The architect wants daylight and decent open space. The budget wants the cheapest structure. In the old way you'd draw two or three schemes over a week and argue about them. With TestFit, the architect generates _hundreds_ of layouts against zoning, unit-mix and parking in real time, then ranks them by yield, by parking ratio, by open space. The machine surfaces the trade-off curve in an afternoon. But which point on that curve is _right_ — that is not a number the software can give. That is the designer's call, and it always will be.
Diverge by the hundred, rank by the goal, judge the trade-off
Generate many, score against explicit goals, compare
Optioneering is the discipline of generating many design options and ranking them against measurable goals — cost, daylight, unit yield, parking ratio, open space — instead of lovingly developing one scheme and hoping it's good.
The AI's contribution is volume and speed: it produces options far faster than you can draw, and it scores them on the metrics you name. TestFit does this for multi-family and site feasibility, generating layouts in real time from zoning, unit-mix and parking constraints and reporting yield instantly — it is firm-priced, aimed at developers and feasibility teams. Autodesk Forma (formerly Spacemaker) brings optioneering to early massing and feasibility, with performance analysis baked in — sun, wind, noise — so you can compare options on environment, not just area.
The shape of the work flips. You stop being the person who draws the answer and become the person who frames the question well enough that the machine can search for good answers.
The machine can rank options. It cannot decide which ranking matters. That gap is where you earn your fee.
Garbage constraints, garbage options — and only you weigh the Pareto front
Two jobs stay stubbornly human. First, setting the constraints. The tool optimises exactly what you tell it to, and nothing else. Ask only for maximum yield and you'll get a daylight-starved slab that technically wins. Forget to constrain open space, fire access or NBC setbacks and the 'optimal' option may be unbuildable. Framing the goals — and which ones are hard limits versus soft preferences — is design thinking, not data entry.
Second, judging the trade-offs. Optimisation rarely gives one winner; it gives a Pareto front — a curve where more of one good (units) costs you another (daylight, open space, cost). The software can draw that curve and rank points on a single axis. It cannot tell you that this client, on this site, in this market, should trade two units for a better-lit courtyard. That weighing — across goals that don't share a unit — is judgement. The landscape is clear that early adopters see timelines and costs fall up to ~20% with these workflows; the saving comes from searching faster, not from outsourcing the decision.
Optioneering is most powerful at feasibility and concept on constrained urban sites. Use TestFit or Forma to explore the yield-vs-quality trade-off honestly _before_ you fall in love with a scheme — it makes you a sharper negotiator with developers because you can show the cost of every demand in real numbers. But own the constraint set: encode NBC setbacks, fire access and open-space rules as hard limits, and treat anything the tool calls 'optimal' as optimal-for-the-goals-you-named, nothing more. The judgement on the Pareto front is the design.
Optioneering is rarer in pure interiors, but the mindset transfers directly. When you lay out a restaurant, a clinic or an open-plan office, name your goals explicitly — covers/seats, circulation width, daylight reach, served-vs-back-of-house ratio — and generate variations against them rather than defending your first sketch. Even a floor-plan generator becomes an optioneering tool the moment you start scoring its outputs against a written goal list. The discipline is yours; the tool just makes it fast.
TestFit and Hypar-class platforms are firm-priced and aimed at developers, so as a solo you may not run them daily — but understanding optioneering lets you punch up. When a developer waves a 'maximise units' brief, you're the one who can say 'maximise units _subject to_ daylight and open space, and here's what each unit costs the courtyard'. That reframing is high-value consulting a one-person studio can sell. Practise the discipline on free floor-plan tools by always writing your goals down and scoring against them.
TestFit
Multi-family & site feasibility, real-time
Generates layouts in real time from zoning, unit-mix and parking constraints and reports yield instantly; firm-priced, developer-facing. Brilliant for the yield trade-off; it optimises what you constrain, so your constraint set is the design.
Autodesk Forma
Early massing, optioneering + performance
Was Spacemaker. Compares massing options with sun, wind and noise analysis built in, so you optimise on environment, not just area. Early-stage feasibility tool — its analysis informs, it does not certify code compliance.
Finch3D
Graph-based option optimization
Optimises early-stage layouts with instant feedback as constraints change — fast for generating and comparing arrangement options. Ranks arrangement, not legality or cost; pair with your own checks.
Hypar
Computational building generation
Generates buildings from rules and goals for repeatable, parametric optioneering; firm-priced. Powerful for teams that can author the logic; the rules you write are the ceiling on what it explores.
“Optimization gives you the single best design — the software finds the optimal answer, so you just build the winner.”
Optimisation finds the best option for the exact goals you encoded, and usually returns a trade-off curve, not one winner. Push for maximum yield and you'll get a daylight-starved, open-space-poor slab that 'wins' on the one metric you named. The genuinely best design lives in weighing competing goods that share no common unit — units against daylight against cost against delight — and that weighing is human judgement the machine can inform but never make.
Workshop — run an honest optioneering study against three goals
You'll take one constrained site, set three competing goals, generate options, and build a tiny trade-off table that forces the real design conversation. The point isn't the winning number — it's seeing the Pareto curve and choosing on it.
Better: TestFit or Autodesk Forma (trial/firm seat). Free fallback: any floor-plan generator + a spreadsheet to score by hand.
Write your GOALS and LIMITS before you generate anything: SITE: [plot area] sq m, [zone], FAR [x], height cap [y] m HARD LIMITS (never break): - setbacks per local bye-law - NBC fire access + open space minimum SOFT GOALS (to rank, with a weight 1-5): - max saleable units weight: __ - daylight to living spaces weight: __ - usable open / courtyard space weight: __ - structure cost (lower better) weight: __ GENERATE: 20+ options; record units / daylight / open space / cost
- 1Encode the hard limits first — setbacks, fire access, open-space minimum. If the tool can't take them, note them as a manual filter you'll apply to every option.
- 2Generate at least 20 options and record the four soft-goal numbers for each in a simple table.
- 3Sort by each goal in turn. Notice the top option for 'units' is rarely the top for 'daylight' — that disagreement is the Pareto front made visible.
- 4Apply your weights to score each option, but then do the human move: pick three that score well and look at them as drawings, not rows.
- 5Write the trade-off sentence you'd say to the client: 'Option B gives two fewer units than A but a real courtyard and lower cost — I recommend B because _.' That sentence is the deliverable optimisation can't produce.
- 6Note one constraint you missed the first time (almost everyone forgets open space or fire access) and add it — proof that framing, not generating, is the hard part.
You’ll walk away with
A trade-off table across three competing goals plus a one-sentence, defensible recommendation — the artefact that shows you set the constraints and judged the curve, which is the part of optioneering only you can do.
A quick reframing exercise, if you have five minutes.
- 01Take a brief you've worked on and rewrite 'design the best layout' as an explicit goal list with hard limits and weighted soft goals. Notice how much design thinking was hiding in the word 'best'.
- 02Generate two option sets — one with 'max units' as the only goal, one with daylight and open space added. Compare the 'winners'. The difference is the cost of a lazy constraint set.
Optioneering flips the work: instead of drawing one scheme, you generate and rank many against explicit goals. The machine supplies speed and scoring; you supply the two things it can't — a constraint set that captures what actually matters, and the judgement to choose a point on a trade-off curve that has no single right answer.
Generate many options, score them against measurable goals (yield, daylight, cost), read the Pareto trade-off. TestFit and Forma do the generating and ranking; you frame the constraints and weigh the goods that share no unit. Optimal means optimal-for-what-you-asked, never simply best.
What is design optioneering in architecture?
Optioneering is generating many design options and ranking them against measurable goals — cost, unit yield, daylight, open space, parking — rather than developing a single scheme. Tools like TestFit and Autodesk Forma make it fast: they produce dozens of layouts and score them in real time. The value isn't a single 'best' answer but a clear view of the trade-offs, so the designer can choose with eyes open instead of defending the first idea.
Does AI optimization tell you the best building design?
No. Optimisation finds the best option for the exact goals you encode, and usually hands back a trade-off curve, not one winner. Ask only for maximum units and it returns a daylight-starved slab that 'wins' the metric you named. Choosing the right point on the curve — weighing units against daylight against cost against delight, none of which share a unit — is human judgement the software can inform but never make for you.
Can a small Indian studio use optioneering tools?
TestFit and Hypar are firm-priced and developer-facing, so a solo may not run them daily — but the discipline scales down. You can practise optioneering on free floor-plan generators by always writing an explicit goal list with hard limits and weighted soft goals, then scoring outputs against it. Even understanding the method makes you a sharper consultant: you become the person who reframes a vague 'best layout' brief into a defensible trade-off.
_Optioneering rides on rules and goals you encode — so the next question is how those explicit rules (parametric design) relate to learned, generative ML, and when each one wins._