The back-test
Run the engine on deals that actually happened and see whether it would have priced them. Four of five landed inside the band; here is the one that didn’t.
Would it have priced real deals?
A valuation method should be able to look at deals that actually happened and land near the prices actually paid. The back-test does exactly that: it runs each disclosed deal in the industry research through the engine, with only that company’s revenue and kind of business, and asks whether the market lens that applies would have bracketed the price.
- A residential run-off book, 2018paid $90.50M · band $72.40M–$108.60M · inside
- An enterprise UCaaS platform, 2022paid $250.00M · band $184.20M–$263.20M · inside
- An own-platform provider, 2022paid $68.00M · band $70.00M–$100.00M · 3% below
- A hosted book sold per user, 2022paid $9.75M · band $7.63M–$17.60M · inside
- An own-platform provider, 2025paid $50.50M · band $35.40M–$50.50M · inside
How to read it. The blue bar is the range the engine’s lens gives for a business of that size and kind. The dot is what the buyer actually paid. A dot on the bar is a hit; the orange dot is the miss.
Deals are described by kind and year here; the report names them, and the buyers.
What we did about the miss
One own-platform provider sold 3% below the band’s floor. With two of the three own-platform deals inside, the size adjustment on that lens is left as it is under policy 1.2. A second miss in the same direction would trigger a refit, and the policy log would record it.
Each deal runs on its own kind of business with every other input at that kind’s median. Books sold at run-off prices are tested on the per-seat lens, because that is the lens a buyer of a book uses.
Six ways to validate, honestly scored
The back-test is one of six layers. The report scores each one, including the ones that aren’t done yet.
| Layer | The question | Status |
|---|---|---|
| Arithmetic | Do the numbers reconcile? | Done, on every build |
| Conceptual | Are the formulas economically right? | Written down for review |
| Directional | Does the value move the right way when an input moves? | Done: 20 checks on every build |
| Back-test | Would the market lenses have bracketed real deals? | Run: 4 of 5 inside |
| Benchmark | Are the bands reasonable against an outside source? | Labelled; not yet second-sourced |
| Out of sample | Does it work on a business it wasn’t tuned on? | Not yet |
Out-of-sample testing is the honest gap: the engine’s tests run eight sample businesses across its kinds, but no second real company has been run end to end. The report says so on its validation page.
- Method. Each deal is run through the engine on its own archetype with its revenue, and its user count where a book was sold per user; every other input at the archetype median
- Hit. The lens band at that deal’s size brackets the price paid
- Result. Run-off book 1 of 1; platform 2 of 3 (1 below); book 1 of 1
- Refit rule. A second miss in the same direction re-fits the size adjustment and increments the policy
See it on a real report, or see it on yours.Download a sample reportSchedule demo