Motor Insurance AI Cross-Sell: Value Measurement
The figures below come from a demo dataset used to show the platform working end-to-end — not from a named or unnamed insurer’s live book.
Most AI ROI numbers are hypothetical, or at best overstated.
The Platform exists to replace that with an identified effect and an honest net figure.
150,000 motor insurance quote sessions. Two equal, balanced groups. Half saw a new AI cross-sell chatbot; half did not. The attach-rate lift was caused by the use-case, not correlation. The gross headline was £270k in extra premium. Net Value, after the costs that usually get left out, was £64,857 — a figure with a confidence interval behind it, not just a headline.
The gap between those two numbers is where most AI ROI claims fall apart. The platform is built to show both.
The setup
In this demonstration, 150,000 motor insurance quote sessions are split into two equal, balanced groups.
- Treatment: sessions that saw a new AI cross-sell chatbot
- Holdout: sessions that did not
Equal, balanced groups are the point. Without a holdout, a lift in attach rate is a story you can tell. With one, it is an effect you can identify.
The result
Cross-sell attach rate rose from 5.8% to 7.6%.
That is +1.8 percentage points, 95% CI 1.55–2.05pp, genuinely caused by the AI use-case, not correlation.
The interval sits above zero. The effect is not an artefact of a noisy week, a seasonal mix, or a friendly dashboard.
Gross vs net
The topline number was £270k in extra premium.
That is the number most AI ROI slides stop at.
After subtracting:
- cancellations
- a small drag on core quote completion
- the AI’s own running costs (build & deploy, compute, support load)
the real net Value was £64,857.
Same use-case. Same RCT. Same sessions. The difference is honest accounting: what the chatbot attached, minus what it displaced, minus what it cost to run.
£270k is a gross headline. £64,857 is a Value figure with a confidence interval behind it. Boards should ask for the second number. Most vendors will only offer the first.
How the platform measures it
No modelled ROI. No spreadsheet multiplier.
The result is read from the data warehouse or lake, the data already there, and observed in the FernAI AI Value Measurement Platform, fully deployed and hosted on Google Cloud.
Causal inference does the identification: an RCT, not a before/after, not a correlation dressed up as impact. The outcome sits on the Value Ledger which is an append-only record of monetised AI outcomes, not a slide that expires when the pilot ends.
What this means for a buyer
If your AI business case is a gross revenue line with no holdout, no cost stack, and no interval, you do not have a Value number. You have a hypothesis.
This demonstration shows the standard the platform is built to hold:
- Identify the effect. Split the traffic. Keep a holdout. Report the lift with a confidence interval.
- Account for the full stack. Extra premium is not Value. Subtract cancellations, operational drag, and the AI’s own running costs.
- Measure from the warehouse. Read the data you already trust. Observe it in the platform. Do not outsource the truth to a vendor dashboard.
- Keep the record. A net figure on the Value Ledger survives board scrutiny. A gross headline does not.
The commercially useful question is not “did attach rate move?”. It is “what was the net Value, and can you prove the AI caused it?”.
Get in touch
If you need that standard on a use-case you are about to fund, or one you have already deployed:
