◆ PRODUCT DEMO · NO SIGNUP REQUIRED
This is what MarginFuse shows you.
Support Copilot is a fictional AI support product with 66 customers on three paid plans and a free tier, running entirely in dry-run. The figures below illustrate the original seeded month. The live workspace rolls forward with time: its selected period, prorated revenue, and totals may differ.
READ-ONLY · FICTIONAL DATA · NO ACCOUNT NEEDED
WORKSPACE · SUPPORT COPILOT · ○ DRY-RUN
CUSTOMERS
66
COST MORE THAN THEY PAY
▼ 4
PLANS
$29.00 · $99.00 · $299.00
POLICIES
1 dry-run · 1 draft
ILLUSTRATIVE SEEDED MONTH · FICTIONAL COMPANY · USD · MONTHLY SUBSCRIPTIONS
MARGIN PER CUSTOMER
Four of 66 customers are the entire problem.
The workspace ranks all 66. These are the four underwater ones plus the two that frame them - the worst and the healthiest end of the same month.
| CUSTOMER | PLAN | REVENUE | AI COST | MARGIN |
|---|---|---|---|---|
| Zephyr Works | Pro | $99.00 | $626.40 EST | ▼ −532.7% |
| Nova Labs | Basic | $29.00 | $35.67 EST | ▼ −23.0% |
| Alder Point | Pro | $99.00 | $118.60 EST | ▼ −19.8% |
| Bluepeak Systems | Basic | $29.00 | $34.20 EST | ▼ −17.9% |
| Pixel North | Pro | $99.00 | $62.20 EST | ◆ 37.2% |
| Acme Studio | Pro | $99.00 | $18.40 EST | ● 81.4% |
Averages hide this. Blended margin across 66 customers looks acceptable while Zephyr Works burns Pro-plan revenue six times over in one day and Nova Labs loses money quietly, every month, on a plan that looks fine on paper.
ILLUSTRATIVE SEEDED MONTH · FICTIONAL COMPANY · USD · ONE MONTH
COST PER FEATURE
Not all requests cost the same.
Features get cost and cost share, not a margin. A feature only has a gross margin if it earns revenue directly or you tell MarginFuse how to allocate revenue to it - so until you do, the workspace reports what a feature costs and refuses to invent the rest.
image_generation
GPT-IMAGE-1
the expensive one nothing was stopping - 24 requests in a single day cost more than six Pro subscriptions
document_analysis
CLAUDE-SONNET-4-5
heavy tokens, concentrated in a handful of document-first customers
ai_chat
GPT-4.1
the volume feature - every customer, every day
smart_reply
GPT-4.1-MINI
high volume, negligible cost - the shape you want
The single worst day in the month: Zephyr Works ran 24 image-generation requests on Aug 26 for $612.40 EST - against a $99.00 monthly subscription. Nothing in the product was watching that boundary.
ILLUSTRATIVE SEEDED MONTH · FICTIONAL COMPANY · USD · ONE MONTH
WHAT PROTECTION WOULD HAVE DONE
Every decision, simulated before it’s real.
In dry-run, policies watch traffic and record what they would have done. You inspect each decision before anything is allowed to act.
Protect Basic plan customer margin
○ DRY-RUN
Downgrade gpt-4.1 chat requests to gpt-4.1-mini once a Basic-plan customer’s AI cost plus the next request would pass $25.00 in their billing period.
○ Dry-run: 6 Nova Labs requests crossed the boundary and would have run on the cheaper model - $1.95 EST avoided. Production was never touched.
Block runaway image generation
‖ DRAFT
Block image_generation once the feature passes $60.00 of cost in one day.
‖ Draft, written the day after Zephyr Works’s spike. It has never run, so it has no evidence yet - and the report says so rather than crediting it with the spike it was written for.
The workspace separates two things this page keeps separate too: what the recorded decisions prove, and what replaying a policy over the full month suggests. They are different numbers because they measure different traffic, and the savings report says which is which rather than picking the bigger one.
ILLUSTRATIVE SEEDED MONTH · FICTIONAL COMPANY · USD · ONE MONTH
Now run it on your own traffic.
Add one track call and revenue from Stripe, RevenueCat, or declared plans to explore your own data - free, no credit card required.