Why vote counts and mention counts mislead
Most teams rank requests by how often they hear them. That rewards whoever is loudest. One customer who writes in thirty times looks bigger than thirty customers who each mention a problem once. A single request from a large account can outweigh a pattern across your whole segment, or be missed because it arrived as one support ticket.
It matters because most product strategy is still shaped by whoever asks. In ProductPlan’s 2023 survey of more than 1,500 product managers, about half said their strategy is mainly influenced by requests from senior executives or direct customer feedback (ProductPlan, 2023). Revenue gives that feedback a weight you can defend.
Step 1: group requests into problems
Customers ask for solutions (“add a CSV export”). Group the requests by the problem underneath (“can’t get data into our reporting tool”). Five different feature asks often turn out to be one problem, and the problem is what you prioritize.
Step 2: count people and accounts, not messages
For each problem, count the distinct people who raised it and the distinct accounts they belong to. Messages are a measure of noise. People and accounts are a measure of reach.
Step 3: attach revenue from your CRM
Link each person to their company account, by email domain or by the account id your support tool already holds, and read the account’s deal or contract value from the CRM. Be strict: never link a free-mail address like gmail.com to a company, and count each account once per problem, however many of its people wrote in.
Now each problem has three numbers: people, accounts and revenue.
Step 4: weigh revenue, do not let it decide alone
Revenue is a weight, not the answer. Ten accounts asking for the same thing are usually stronger evidence than one account worth the same combined amount, because the ten describe a market and the one describes a customer. Upvoty’s guide makes the same point and suggests capping the revenue weight so one unusually large account cannot override strategy (Upvoty).
A simple way to read the numbers:
| Pattern | What it usually means |
|---|---|
| Many accounts, high revenue | A strong candidate. Check usage and fit with strategy. |
| Many accounts, low revenue | A real problem in a smaller segment. Decide if that segment matters to you. |
| One account, high revenue | A retention or sales question. Handle it as one, not as roadmap evidence. |
| Few people, no accounts linked | Too early to tell. Watch it. |
Step 5: add usage and reliability evidence
Check what your analytics show for the affected area (how many users touch it, whether usage is falling) and whether errors cluster there. A complaint backed by a drop in usage is a different priority from one that is not.
Step 6: decide, and write down why
Record the call (build, defer or reject), the reason, the numbers you looked at, and what would make you look again. Pendo found that 80% of features in the average product are rarely or never used (Pendo, 2019); the record is how you learn which of your bets were worth it. See how to keep a product decision log.
Doing this in purpal
purpal reads feedback from Intercom, Slack, Zoom, Google Meet and uploaded files, groups it into themes, and ranks themes mainly by the number of people affected, weighted with message volume, urgency and growth. It links each person to their account, and with HubSpot connected each theme shows the accounts and deal value behind it, for example “40 customers (Intercom), 3 accounts worth $240k (HubSpot)”. Usage from your connected analytics tool (PostHog, Mixpanel or Google Analytics) is there when you ask purpal about a theme. When you decide, the decision and your reason are recorded on the theme.
Questions
Should we use ARR or deal value?
Use whatever your CRM holds reliably. Current ARR measures what is at risk; open deal value measures what a problem is costing you in sales. Say which one you used.
What about RICE or weighted scoring?
They still work. Accounts and revenue make the “reach” and “impact” inputs real numbers instead of guesses.
What if most feedback has no account attached?
Count it by person and say so. A theme with 200 people and no linked accounts is still evidence; it just cannot carry a revenue number yet.