What-If Machines

Most expensive debates are unshared assumptions wearing the costume of strategy. A what-if machine makes scope, time, cost, and demand into toggles people can stress-test together. Build the sim before you escalate the thread.

When I wrote about opportunity AI, the pattern under half the ports was the same: externalise trade-offs that used to live only in someone's head. Shapeable proposals. Decision simulators. Explorable demos. Those are what-if machines.

Efficiency AI drafts the email that continues the argument. Opportunity AI ships a small interactive object that ends half of it.

Why meetings fail the what-if test

In a slide deck, "what if demand is half" is a verbal side quest. Someone says it. Someone else says "we'll watch leading indicators." Nobody updates the numbers in shared view. The room leaves with different movies in their heads.

In a what-if machine, demand is a slider. Margin is a formula. Staffing is a constraint. The pessimistic case is a button, not a personality trait.

That shift matters from SMB pitches to enterprise investment committees. The artefact becomes the meeting.

Three machines worth building

1. Shapeable proposals

Instead of a PDF with a single price, ship a small page:

• Scope tiers the client can toggle
• Time as a control (rush vs standard)
• Cost that recalculates live
• Explicit exclusions that appear when a toggle is off

I have used this framing for builder work around RivaFlow and photography packages. The buyer navigates by their constraint, not my slide order. Reply-all shrinks because the trade-off was felt before the legal thread started.

2. Decision sims for operators

Pick one recurring decision that burns hours: hiring pace, inventory buffer, rollout order, training load for a BJJ program, publish cadence for a media series.

Encode:

• Assumptions (labelled as assumptions)
• Levers
• Outcomes you care about
• A "break the plan" case

Run it with the team once. Then leave the sim in the repo. Next time the debate returns, open the same URL. Update assumptions in public. That is institutional memory without another strategy doc.

3. Explorable demos

Buyers ask about security, setup cost, and failure modes in an order you cannot predict. An explorable demo lets them navigate by question. Agents make these cheaper to compose: scripted paths, generated UI stubs, recorded proof clips. The point is not polish. The point is shared inspection.

How I build an ugly v0 in four hours

1. Name the fight. One sentence: what people keep arguing about.
2. List five assumptions. Write them as numbers or booleans.
3. Pick three levers. Only three. More is theatre.
4. Choose one metric that decides the call (cash runway, cycle time, injury risk, NPS proxy, publish rate).
5. Ship a single HTML page or notebook with sliders and a results panel. No auth. No brand system.
6. Sit with two humans and watch where they get confused. Fix labels, not graphics.
7. Park it next to the proposal with a one-line BLUF: "Open this before we debate."

Stack does not matter. I have done this with a static page, a spreadsheet with form controls, and an agent-generated React stub. The machine is the product. The framework is disposable.

SMB to enterprise

SMB: One proposal sim per offer. One ops sim for the bottleneck that wakes you at 2am. Measure whether sales cycles shorten or internal rework drops.

Enterprise: Treat what-if machines as decision hygiene. Require investment memos to link a live assumption panel. Version the assumptions with the decision record. Prohibit "we'll model that later" when the model is a half-day of work.

Guardrails still apply. Do not put confidential data in a public CodePen. Use synthetic numbers for external shares. For internal sims, access control and audit still matter. The demo is not the system, but the sim can be a first-class artefact in the system.

Failure modes

• Too many levers (a flight simulator nobody can fly)
• Hidden formulas (trust dies)
• Treating the sim as prediction instead of shared assumption hygiene
• Building for a committee of fifty before testing with two
• Polishing visuals before the metric is honest

Closing

Arguments are expensive because they are low bandwidth. What-if machines raise the bandwidth: they force numbers onto the table and make disagreement inspectable.

If your next thread is already writing itself in your head, stop drafting the email. Spend four hours on the machine. Then invite people into the what-if, not into your certainty.