SkillsSkill 1 of 5
Test the thesis, not the summary
Rebuild any headline metric from the data, starting with its denominator, before it goes in the thesis.
Worked example
The AI summary says 108% retention. Do you believe it?
A summary is only as good as its denominator. Rebuild the headline number before it carries the thesis.
The question
Quillfield, a fictional provider of claims software and outsourced document review to insurers, is for sale. FY25 revenue was $240m. Your deal team’s AI tool summarized the CIM, and the partner wants to know whether retention supports the thesis.
Quillfield retains and grows its customers: net revenue retention was 108% in FY25, reflecting broad-based expansion across the carrier base.
| FY24 | FY25 | |
|---|---|---|
| Top 2 customers | 50 | 69 |
| Other 138 continuing customers | 150 | 147 |
| 12 customers that left in FY25 | 20 | 0 |
| All FY24 customers | 220 | 216 |
Invented data. FY24 revenue was $220m, all from these customers. FY25 revenue was $240m: $216m from FY24 customers plus $24m from new ones. 216 ÷ 200 = 108% counts only customers who stayed; 216 ÷ 220 = 98.2% (rounded) counts everyone. The top 2 grew by $19m, $14m of it from a one-off claims-backlog project.
“The AI summary says Quillfield’s net revenue retention was 108%, from broad-based expansion. Using the customer table, would you put that number in the investment thesis?”
Your turn
Say your answer out loud, or jot notes below. Both work. Or skip to the worked answer.
Not saved by this site. Your browser may restore text when you navigate back.
A strong answer, step by step
Open each step in turn. This is one strong answer, not the only one.
Step 1: Rebuild the number
What would you do here? Open the move you’d make.
Take 108% as given: the CIM and the AI summary agree, and it’s a standard metric.
Tempting, but it misses something. Two sources agreeing isn’t a check. 108% is 216 ÷ 200, which drops the $20m the 12 leavers paid in FY24; the summary copied the CIM’s denominator.
Compare 108% with peer benchmarks to see whether it’s plausible.
Tempting, but it misses something. A benchmark can’t show how this number was built. The table lets you compute it directly, and 216 ÷ 200 leaves out every customer who left.
Rebuild it: 216 ÷ 200 counts only customers who stayed. Put the 12 leavers back and the base is 220, so retention is 98.2%.
The strong move. It checks the denominator before the meaning. Adding back $20m of lost revenue turns expansion into slight contraction: 216 ÷ 220 = 98.2%.
See how a strong candidate says it
“108% is 216 over 200: it divides by the customers who stayed. Put back the 12 who left and FY24’s base is 220, so retention is 98.2%, not 108%.”
Why it works: It checks what the summary computed before arguing about what it means.
Step 2: Test “broad-based”
What would you do here? Open the move you’d make.
Split it: the top 2 grew from 50 to 69, the other 138 fell from 150 to 147, and with the leavers that group kept 147 of 170.
The strong move. It shows where the growth sits. Two accounts added $19m, while everyone else, counting leavers, kept about 86.5%. “Broad-based” is false.
Read the average: 108% across 140 continuing customers means most of them expanded.
Tempting, but it misses something. An average says nothing about spread. The other 138 continuing customers shrank as a group, from 150 to 147; two accounts carry the whole increase.
Note that the top 2 are nearly 29% of FY25 revenue and flag concentration as a risk.
Tempting, but it misses something. The concentration is real, 69 of 240, but a flag isn’t a test. It never checks the claim that expansion was broad; the other 138 fell from 150 to 147.
See how a strong candidate says it
“The top 2 went from 50 to 69. The other 138 continuing customers fell from 150 to 147, and with the leavers that group retained 147 of 170, about 86.5%. The expansion sits in two accounts.”
Why it works: Concentration changes the thesis from ‘sticky base’ to ‘two relationships’.
Step 3: Separate recurring from one-off
What would you do here? Open the move you’d make.
Count the top 2’s $19m increase as expansion: they’re the best customers, so it should repeat.
Tempting, but it misses something. $14m of the $19m is a one-off backlog project. Only $5m is evidence that these customers keep buying more.
Strip out the one-off: $14m of the $19m is a backlog project. Without it, FY24 customers paid $202m, 91.8% retention.
The strong move. A thesis rests on revenue that repeats. Without the project the top 2 grew $5m, and retention is 202 ÷ 220 = 91.8%.
Treat the backlog project as run-rate, since the customers chose to buy it.
Tempting, but it misses something. A project that clears a backlog ends when the backlog does. Counting $14m as recurring adds 6.4 points to retention, 98.2% against 91.8%.
See how a strong candidate says it
“$14m of the top 2’s $19m increase is a one-off backlog project. Without it they grew $5m, and FY24 customers would have paid $202m in FY25, 91.8% retention.”
Why it works: A thesis rests on revenue that repeats, not on a project that ends.
Step 4: Decide what goes in the thesis
What would you do here? Open the move you’d make.
Keep 108% in the memo and add a footnote on how it was calculated.
Tempting, but it misses something. Committees read headlines, not footnotes. A thesis on 108% still rests on leaving out $20m of lost revenue and counting a one-off project.
Leave retention out of the thesis until the seller sends more data.
Tempting, but it misses something. That avoids the question. The table already supports a range from 91.8% to 98.2%, and you can name the data that would narrow it.
Underwrite retention in the low 90s at best, and ask for customer-level revenue, the top 2’s contracts and why the 12 left.
The strong move. It turns a doubtful headline into a working range, 91.8% to 98.2%, plus three specific requests that would move it.
See how a strong candidate says it
“I wouldn’t write 108%. I’d underwrite retention in the low 90s at best, and ask for customer-level revenue by product, the top 2’s contracts and why the 12 left.”
Why it works: It turns a doubtful headline into specific diligence requests.
The answer, in short
- Decision
- Don’t put 108% in the thesis. Underwrite retention nearer 92% to 98% until customer-level data says otherwise.
- Because
- The 108% leaves out $20m of lost revenue, and the growth that remains comes from two customers, mostly a one-off project.
- What would change it
- Signed multi-year expansions from the top 2, or proof that the 12 leavers were small accounts Quillfield chose to drop.
Common traps
- Accepting a summary’s headline without its denominator.
- Reading an average as broad-based.
- Counting a one-off project as recurring revenue.
- Asking for ‘more data’ without naming which data.
Check your answer
- Did I recompute the retention figure myself?
- Did I include the customers who left?
- Did I check how concentrated the growth is?
- Did I name the diligence requests that would settle it?
Want feedback on how you’d answer? Practice “Test the thesis, not the summary” with an AI interviewer on Socratify.
Last reviewed Nov 9, 2026