Capacity intelligence for architecture firms

Know what your firm can absorb.
Before you say yes.

Prizma is a live model of your architecture firm: every commitment recalculates capacity, fees and risk across the whole practice.

Deterministic core · Agentic layer · Data you already have

Mara VidalStaff stretch medium
available for a new project
4.2 h/wk
support only
Connected projects33.0 h/wk
Harbor Bathsdue 2026-05-25DD18.7 h
Lumen Paviliondue 2027-04-30SD14.3 h
Lumen PavilionCultural · SD 3.02x
labor budget · fixed
$2,818 /wk @ 2.2x
planned burn$2,056 · 73%
room left$762/wk
Harbor BathsCultural · DD 1.30x
labor budget · fixed
$1,288 /wk @ 2.6x
planned burn$2,579 · 200%
room left$-1,291/wk
CulturalTeam · Scope 2.24x
team multiplier · on DPE
2.24x
stability margin4.2 h/wk
tightest staffMara V.
Atelier SurStudio · Firm
net multiplier · on DPE
2.23x
utilization83%
survives losing1
Ask Prizma

Pull a slider · ask the sticky. Every figure recomputes through the model.

Deterministic core, agentic layer Hours are derived, never entered Every number traceable Data your firm already has

( 01 ) · The five questions

Five answers a principal needs. Live, on one canvas.

Not dashboards about last month. The state of the firm right now, recalculated on every connection you make.

01
Can we take this on?
Stability margin

Drop a potential project on the canvas as a dashed node. See what every person and team gains and loses before anything is signed.

02
Does it pay?
Net multiplier

Fee measured against the burdened cost of the exact people connected to the work. Never a blended rate, never an average of averages.

03
Can the project afford this team?
Room left · Planned burn

The budget is fixed in money. The hours ceiling moves with the real rates of who you assign, so a senior-heavy team burns the room faster and you see it instantly.

04
Where are we rigid?
Rigid / liquid margin · Silo gap

How much of your capacity can actually move between teams, and how much is locked inside one silo. The difference is where growth stalls.

05
Who is critical?
Fragility · Single point of failure

Who orphans a project if they leave tomorrow, and how many people the firm can lose before delivery breaks.

( 02 ) · Roadmap

Deterministic first. The rest earns its way in.

Ordered by engineering risk, not by topic. The phase boundary is a wall, not a suggestion.

Prizma X

Deterministic

Arithmetic on data the firm already has. No learning, no prediction, no assumptions. Every connection propagates through the whole model, so it cannot hallucinate.

yield · stability margin · fragility · room left · utilization · break even

Prizma Y

Forecasting

Effort curves across each project, overlaid to pick a start date by peaks, revenue over stages with and without the potential project. Built with the pilot firms, because they decide the level of detail.

effort curves · start dates · staged revenue · reassignment editor

Prizma Z

Intelligence

The system learns from accumulated history: what hospitality actually yields per hour, where fee assumptions are wrong, what to take on next. Agentic from the ground up; the benchmark data compounds here.

type intelligence · fee benchmarks · recommendations · simulation

( 03 ) · Why not a PM tool

Management software records what happened. Prizma computes what happens next.

42% of firms cannot report their own net margin 50% assign staff on intuition alone 68% run firm finances on spreadsheets Factor A/E benchmark 2026 · RIBA poll 2026
They record the past
  • Timesheets, after the week is gone
  • Invoices, after the money moved
  • Utilization reports, four weeks late
  • Post-mortems on projects that lost money
Prizma answers what's next
  • Can we absorb this project
  • Does it pay with these exact people
  • Who breaks if one person leaves
  • What can move between teams, and what can't
Net multiplier
2.31x
Break-even rate
$58/h
Utilization
76%
Stability margin
8.6 h/wk
Survives losing
2 people
One connected model · every figure derived, none entered

( 04 ) · The instrument

The model is the workspace, not a report.

Capacity Model

Every person, project and team is a node. Select one and the panel shows its multiplier, computed once over sums, never averaged.

The Prizma canvas: staff, project and team nodes connected by edges, with the Hospitality team panel open showing a 4.42x multiplier

Effort EQ

Slide one person’s hours between projects. Burn, room and fees recascade through every team it touches.

The Effort EQ panel rebalancing one person's hours across three projects, with burn and room recomputing on each project card

( 05 ) · The math

Deterministic underneath. Agentic on top.

The engine is plain arithmetic on data your firm already has: who works where, what each hour truly costs, what each project pays. Agents sit on top of that model to run scenarios, flag risk and prep decisions; they operate the arithmetic, never invent it. Every number traces back to a connection you can point at.

Hours are derived, never entered.

You connect a person to a project. The model derives their hours from capacity, focus and the team around them.

Ratios are never averaged.

Numerators and denominators aggregate across the firm and divide once, at the end. No average of averages, ever.

Constants are defaults, not truth.

Every threshold and penalty is a configurable assumption, shown and explained. You calibrate them to your practice.

room_left = fee_budget − Σ ( hoursp × burdened_ratep )  // per person connected · no blended rates

( 06 ) · Questions

Before you ask.

Then the math will feel familiar, because a spreadsheet is deterministic too. What it cannot do is propagate: change one assignment and nothing downstream recalculates. Prizma keeps the assumption-first arithmetic and adds the connections, so one change updates capacity, fees and risk across the whole firm, for everyone, at once.

Your team list, salaries, project fees and who works on what. No integrations required to start: a firm of fifteen people is modeled in an afternoon, and the model stays useful from day one.

Neither do most firms: 42% cannot report their own net margin. Prizma does not need calibrated truth to start. You give rough figures, the model makes them explicit and live, and they sharpen as real numbers replace guesses. Assumptions you can see beat certainty you do not have.

No. The model starts from what the firm already knows: who works here, what each hour truly costs, what each project pays. Tools built on timesheets need months of history before a forecast means anything; Prizma answers on day one. If you keep timesheets, they calibrate the model over time. They are never required.

PM and accounting tools are systems of record: they tell you what already happened. Prizma is a decision layer: it tells you whether the next commitment fits, whether it pays, and what it costs the rest of the firm. Keep whatever you invoice with; Prizma sits beside it, not instead of it.

Salary data stays inside your firm. Pilots run on a private instance, nothing is used to train anything, and the benchmark layer (Prizma Z) only ever works with aggregated, anonymized figures from firms that opt in.

Both, in the right order. The engine is deterministic: plain arithmetic on the connections you draw, with nothing predicted behind your back. An agentic layer sits on top of that model to run scenarios, surface risks and prep decisions. Agents work with the numbers; they never make them up.

Four weeks with your real data. We model the firm together in the first session, then you run your next go/no-go decisions through the canvas and judge the answers against your own instinct.

Draw your firm on the canvas.

Pilots for firms of 5–100 people · Four weeks · Your real data · No timesheets