Build the Benchmark from Your Best Customers Playbook 2026

One row per customer who has come up for renewal after at least 12 months paid, or who left in the last 24. Four things per row:

• Tenure & stage: start date, and the three windows you'll measure: Trial, New (first 90 days paid) and Established (the 12 weeks ending 90 days before the scheduled renewal, the same window for every account, so leavers aren't measured while they wind down).
• Usage by feature, per window: weekly active users, seats, and weeks each key feature was used.
• ARR & plan history: ARR now (or at leaving), ARR 12 months earlier, plan changes.
• Outcome, at the renewal: Grew (renewed with ARR up 10% or more, or upgraded plans with ARR at least unchanged), Renewed flat (renewed at the same or higher ARR, short of Grew), Shrank (renewed at lower ARR), Left (canceled). Defaults to tune. Customers who renewed at the same or higher revenue (Grew and Renewed flat) are the benchmark cohort; Shrank and Left stay separate.

A user is active in a week when they take one of your key actions – the ones that deliver the value, not a login. Under about a hundred accounts, do all of this by hand in the spreadsheet below.

Before comparing anything, the CS lead strikes customers who would skew it and writes the reason in Fair match note: internal, partner or test accounts; deep discounts or custom contracts; accounts migrated from another product; tracking gaps of two weeks or more; seasonal customers measured out of season. Keep the struck list – someone will ask.

Split by what changes what good looks like: plan tier or pricing model first, then size (seats or ARR band). Start with no more than four segments. Within each segment and stage, take the median of each measure for each outcome group, with its count beside it – medians, because one large account drags an average.

The median of customers who renewed at the same or higher revenue is the benchmark. The grew median sits beside it, with its own count, as the growth line for expansion plays; the shrank and left medians show what a gap looks like. Use the benchmark as a reference, not a risk threshold: about half of its own cohort sits below it.

Use matched peers for descriptive comparisons. Before a measure triggers a risk flag, check its threshold against later renewal outcomes; for expansion plays, check it against later expansions.

Usage and revenue go together. Say established mid-market customers who renewed at the same or higher revenue have 62% of seats active weekly, use four key features and pay $24,000; those who left had 21%, one feature and $15,000. The $9,000 between them is a comparison, not a forecast. (The numbers are illustrative.)

Rule, default: a segment and stage needs at least 10 fair-match customers who renewed at the same or higher revenue. Fewer, and it isn't benchmarked yet: a median from a handful of customers isn't a benchmark. Show the counts beside every benchmark. Few or no cancellations doesn't block a descriptive comparison, but it limits how far you can test a risk threshold.

For each account in a segment below that count, set Recommended move to No move, with the reason in Move reason and a Review date at the next quarterly refresh. The weekly check skips them rather than guessing. If two small segments look alike, merge them and recount.

Default: 20 or more who renewed at the same or higher revenue, with 12 months of usage tracking behind them, is Provisional. Fewer, or less tracking history (say you changed analytics tools in the spring), is Low confidence. A benchmark becomes Checked only after the frozen version is tested on later renewals it wasn't built from, with its misses, false alarms and uncertainty written down.

One row per segment, stage and measure that passed. We keep what we saw apart from what we think might be true and still need to test: What we saw (customers who renewed at the same or higher revenue used reports four weeks in four) goes in the benchmark; What we think (reports cause the growth) goes on its own tab with how you'll test it.

When there is little history, we trust the benchmark less, and we say so. Set Benchmark confidence to Low; every move the weekly check drafts in that segment carries the low-confidence note below.

The CS lead, one CSM or AM, and someone from sales or finance for the revenue side, on the agenda below. Expect an argument about at least one measure – that's the benchmark earning trust. Nothing reaches an account record until the CS lead signs it off.

Set Benchmark segment, Customer stage and Benchmark confidence on each company in HubSpot. From here, the weekly check compares each account with its own segment and stage.

Before you change anything, test the frozen benchmark on the renewals that came due last quarter: did the gaps show up on the accounts that went on to leave or shrink, and not on the ones that renewed at the same or higher revenue? Count the misses (leavers with no gap) and the false alarms (renewers with one), and write down how sure you can be with that many renewals. Record the answer in the Refresh log: Held up (keep it; after its first pass, mark it Checked), Measure dropped (one missed – drop it), or Rebuilt (it didn't separate – rebuild the segment). Only then re-label outcomes and re-run the medians. A "what we think" row that keeps showing up is still an association, not proof that the feature caused growth; it stays on the To test tab until you've tested it. Fold in what the eight-week review found about which moves worked.