A real prediction report, annotated
Below is one complete MiroFish report, exactly as the product produced it — nothing rewritten, nothing tidied, nothing cherry-picked. Beside each section is a note on what to trust and what to verify. One section is arithmetically wrong, and we have left it in and explained it, because knowing where these reports fail is more useful than another polished sample.
Annotated by Zinedine · Founder & Developer · Report generated 2026-08-26 on gpt-4o-mini · 750 output tokens
Update — 27 August 2026
Writing this page found a real bug. The probabilities below sum to 120%, which is not possible, so we changed the engine: the simulation prompt now requires the most likely outcome and every alternative to sum to exactly 100%, in a single fixed format, and that constraint is checked automatically against eight scenarios before release. It passes 8/8. Reports generated from today will not have this fault. We have left the original report below untouched rather than quietly regenerating it — it is the evidence for the fix, and hiding it would make every other number on this site worth less.
The scenario submitted
“We're a 6-person B2B SaaS doing $40k MRR. We plan to raise prices 30% for new customers only, grandfathering all 340 existing accounts at their current rate, announced via a blog post and in-app banner. Predict: new-signup conversion impact over 90 days, whether grandfathered customers still churn, competitor response, and how the announcement is received publicly.”
Report output begins
Price Increase on New Customers Will Boost MRR, But Risk Churn for Existing Accounts
Executive Summary
The simulation indicates that the planned 30% price increase for new customers will lead to an increase in monthly recurring revenue (MRR), projecting an additional $12,000 in MRR from new signups over the next 90 days. However, it is forecasted that existing customers may experience an uptick in churn due to perceived value shifts, with potential churn rates rising to 5% among grandfathered accounts. Competitors are likely to respond aggressively, possibly by lowering their own prices or increasing their marketing efforts.
Read with care — The dollar figures are illustrative, not calculated
The $12,000 and 5% are not derived from your books — the model never saw your funnel, your churn history, or your conversion rate. It saw "$40k MRR, 340 accounts" and produced numbers of a plausible shape. Treat the direction as the output and the digits as decoration. If you want a figure you can plan against, the arithmetic has to be yours.
Key Drivers
- ·Customer Perception: The way existing customers view the value of the service relative to the new pricing structure.
- ·Competitor Activity: Competitors’ potential moves to capitalize on dissatisfaction among existing customers.
- ·Market Demand: Overall demand in the B2B SaaS market for your offerings.
- ·Communication Strategy: The effectiveness of the blog post and in-app banner in framing the price increase as a positive change.
- ·Customer Support Engagement: The level of support provided to existing customers post-announcement.
Worth trusting — This is the section to actually read
Driver lists are where this kind of simulation earns its money. Not because the list is surprising, but because it is complete — and the one that catches people out is usually communication strategy, which teams treat as an afterthought and which the model correctly ranks alongside the price itself.
Simulation Dynamics
Throughout the simulation, different agents representing customers, competitors, and internal staff interacted in varied ways. Existing customers expressed concerns about the price hike and its implications for service value. New customers, however, responded positively to perceived premium offerings.
Competitors closely monitored the situation, with many expressing intent to reposition themselves in the market.
"I feel like we’re being punished for being loyal. If my rate goes up, I might consider alternatives." – Existing Customer Agent
"This is our chance to attract users who might be unhappy with their current service." – Competitor Agent
Read with care — Nobody said this
These quotes come from simulated agents, not from your customers or from any real conversation. They are useful as a rehearsal of the objection you will face — "punished for being loyal" is genuinely the sentence grandfathering is designed to prevent and often causes anyway — but they are not evidence that anyone feels this way. Never paste one into a deck as a customer quote.
Most Likely Outcome
The most likely outcome predicts an increase in new signups resulting in an overall revenue boost but also a potential churn rate increase among existing customers. The projected MRR increase from new signups is estimated at $12,000, while churn among grandfathered customers could yield a loss of $6,000 in MRR due to the anticipated 5% churn rate.
~70% likelihood of achieving the projected MRR increase while facing moderate churn.
This is wrong — The arithmetic here is wrong, twice — and we fixed one of them
First: 5% churn on 340 accounts at $40k MRR is roughly $2,000 of lost MRR, not $6,000. Second, and worse: this section claims ~70%, and the three alternatives below claim 25%, 15% and 10% — summing to 120%. A probability distribution cannot exceed 100%. We published this rather than regenerating until it looked clean, because it is the single most important thing to know about reading these reports: the numbers are language-model output and they need checking.
Alternative Scenarios
- ·Scenario A: Grandfathered customers remain satisfied, leading to a churn of only 2%. New signups yield a total MRR increase of $15,000. Probability: 25%.
- ·Scenario B: Competitors react swiftly, causing an increase in churn among existing customers to 10%, which balances the new signup gains to zero. Probability: 15%.
- ·Scenario C: The price increase is well-received, with churn stabilizing at 3% and new signups achieving $10,000 in revenue. Probability: 10%.
Worth trusting — The branches are the useful part, the percentages are not
Ignore the numbers and read the shapes. Scenario B is the one worth planning for: it describes the specific mechanism by which this move nets to zero — a competitor moving fast enough to convert your annoyed grandfathered base before your new pricing compounds. That is a concrete, checkable risk you can put a tripwire on. The 15% attached to it is not.
Risks & Uncertainties
- ·Potential backlash from existing customers leading to higher than expected churn rates.
- ·Competitors launching aggressive promotions or enhanced services in response.
- ·Miscommunication in the announcement leading to dissatisfaction and confusion.
- ·Economic downturns affecting overall market demand for SaaS products.
Read with care — The last one is filler
"Economic downturns affecting demand" is true of every business decision ever made and tells you nothing. This is the most common failure mode in generated analysis: a list where three items are specific to your scenario and the fourth is a generic hedge. Delete the hedge, act on the three.
Recommended Actions
- ·Enhance customer support for existing users to mitigate concerns about the price increase.
- ·Monitor competitor pricing strategies closely and be prepared to adjust your offerings.
- ·Craft a strong narrative in your communications to highlight improvements or added value that justify the price increase.
- ·Establish a feedback loop with existing customers to understand their concerns and experiences post-announcement.
- ·Prepare targeted marketing campaigns to attract new customers and retain existing ones through loyalty programs or exclusive offers.
Read with care — Sound, but generic — and that is the honest limit
Every item here is reasonable and none is specific to a 6-person team at $40k MRR. A simulation predicts reactions; it does not know your roadmap, your team’s capacity, or which of these you already do. Read this section as a checklist to argue with, not a plan to execute.
What this report is good for
Read the drivers and the alternative scenarios; treat everything with a currency symbol or a percentage sign as a prompt to do your own arithmetic. The genuine value in this output is Scenario B — the specific mechanism by which a grandfathered price rise nets to zero, because a competitor moves fast enough to convert your annoyed existing base before the new pricing compounds. That is a concrete risk you can set a tripwire on. The 15% attached to it is not.
It is a structured way to be argued with before the argument is real, by something that has no stake in your decision and no reason to be polite about it. That is a genuinely useful thing to have at 2am the week before a launch. It is not a forecast you can bank, and we would rather say so here than have you discover it later.
For the specific things this engine gets wrong, and why, see our research — including exactly what one of these reports costs us to produce.
Questions
What does an AI prediction report actually look like?
It is a structured document, not a paragraph of prose: an executive summary, the drivers the model considers decisive, a narrative of how simulated stakeholders reacted, a most-likely outcome with a stated probability, alternative scenarios with their own probabilities, risks, and recommended actions. The full unedited example on this page is roughly 750 tokens and took about five minutes to produce.
Are the numbers in an AI prediction report reliable?
Treat them as estimates that need checking. The report on this page projects a $6,000 MRR loss from 5% churn on a $40k base, which is arithmetically wrong — the figure is closer to $2,000 — and its probabilities summed to 120% rather than 100%. We have since fixed the probability fault in the engine and now verify it automatically before release, but the wider point stands: the structure and reasoning are the useful output, and any specific figure should be checked against your own numbers before you plan against it.
Are the customer quotes in the report real?
No. They are produced by simulated agents, not collected from real customers. They are useful as a rehearsal of the objections you are likely to face, and they should never be presented as customer research or quoted in a deck as a real voice.
What is an AI prediction report genuinely good for?
Surfacing the drivers and failure modes you have not thought of yet, and rehearsing an argument before you have it for real. It is a structured way to be argued with. It is not a forecast you can bank, a substitute for talking to customers, or a source of numbers to put in a board pack.
Run your own scenario. Monthly plans from $2.99, cancel anytime.
See pricing