How to prioritize customer feedback against impact, effort and strategy

Why does customer feedback need its own prioritization step?

Feature frameworks assume someone already decided what the candidate features are. Customer feedback arrives as hundreds of requests in customers' own words, from support, sales, surveys and reviews, with very different weights behind each one. Running RICE directly on that pile scores the wording of requests rather than the problems behind them.

Sean McBride's RICE method at Intercom is still the most useful base, because it forces four separate judgements: "Reach × Impact × Confidence ÷ Effort". It scores impact on a fixed scale, "3 for 'massive impact', 2 for 'high', 1 for 'medium', 0.5 for 'low', and finally 0.25 for 'minimal'", and confidence as "100% is 'high confidence', 80% is 'medium', 50% is 'low'". What feedback adds is a way to fill in reach and confidence from evidence instead of estimates, and a rule for strategy, which RICE leaves out.

The procedure below has three passes: evidence, score, decision.

Which evidence should each feedback theme carry?

Four numbers per theme, gathered before anyone argues about impact.

Evidence field	What it measures	Where it comes from
Distinct customers	How many separate accounts raised it this quarter	The feedback log, deduplicated by account
Revenue at risk	Annual revenue of requesting accounts that renew within two quarters and have named this theme as a concern	CRM plus the account owner's notes
Cost to serve	Monthly support cost of the theme's tickets	Handle time multiplied by loaded agent cost
Independent sources	How many different channels show the theme	Tickets, survey comments, call notes, reviews, usage data

Distinct customers is the field that changes the most decisions. A theme with 300 mentions can turn out to be 12 accounts, and a theme with 40 mentions can be 40 accounts that each wrote once. The cost-to-serve method is the one from how to turn support tickets into product insights, applied per theme instead of per ticket topic.

Independent sources is the field that keeps the score honest. John Cutler's warning at Amplitude is that "the feedback gathered will often say little about how customers are actually using a product or service and what might cause them to return." A theme that shows up in tickets, in call notes and in usage drop-off is a problem. A theme that shows up only in a survey vote is an opinion.

How do you turn evidence into a score?

Map the evidence onto RICE, then add strategic fit as a separate column.

• Reach is distinct customers this quarter, not mentions.
• Impact uses McBride's scale against one named outcome, such as retention or weekly active use. Write the outcome at the top of the sheet so everyone scores against the same one.
• Confidence comes from the source count: 100 percent for three or more independent sources where usage data agrees, 80 percent for two, 50 percent for one.
• Effort is person-months from engineering, as a range collapsed to its midpoint.
• Strategic fit is 2 (core to this year's strategy), 1 (neutral) or 0 (works against it).

Fit is not multiplied in. A theme with fit 0 is declined unless leadership explicitly chooses to change the strategy, because no amount of reach makes it right to build something that pulls the product away from where it is going. That is the one decision a score should never make on its own.

Revenue at risk is not multiplied in either. Set a threshold, for example 5 percent of total annual revenue. Any theme above it gets an explicit decision from whoever owns the revenue, whatever its score. That way a large account's request is always looked at, and never silently outranks a pattern that touches dozens of smaller customers.

How do you weigh one loud account against a quiet pattern?

Put both through the same sheet and compare reach and revenue at risk side by side. The loud account usually has reach of 1, a high revenue figure and a senior person sending emails. The quiet pattern has reach in the dozens, modest revenue per account and nobody escalating, often because the people affected stopped writing in.

Des Traynor's line at Intercom names the risk on the loud side: "No customer can be more important than a good product. The road to consulting-ware is signposted just this once for just this customer." Each special case looks reasonable alone; ten of them make a product that serves ten customers well and the rest badly.

The quiet side has its own risk, which is assuming that quiet means low value. Rob Markey and Fred Reichheld describe an American Express analysis in the Bain Closing the Loop piece: analysts reviewing the processes that created the most detractors found that the highest-value customers needed card replacements more often than average, and that their Net Promoter scores after a replacement "were almost 25 points lower than the average among other customers." The pattern only showed when the feedback was cut by customer value. That case is one card issuer and a service process rather than a product feature, but the method carries: before you dismiss a quiet theme, check which customers sit behind it.

What does a filled scoring sheet look like?

Here is a quarter's sheet for a fictional expense-management app with 600 paying accounts and €3.6 million in annual revenue. The numbers are illustrative. The outcome everyone scores impact against is 12 month account retention, and the at-risk threshold is 5 percent of revenue, €180,000.

Pass 1: evidence

Theme	Distinct accounts	Revenue at risk	Support cost per month	Sources
A. Bank feed sync fails for two banks	85	€210,000	€3,800	3: tickets, survey comments, churn reasons
B. Accounting export totals do not match	60	€140,000	€4,200	3: tickets, CSM notes, survey comments
C. Mobile receipt capture is slow	140	€0	€600	2: app reviews, survey
D. Dark mode	210	€0	€0	1: survey vote
E. Single sign-on with SAML	1	€96,000	€0	1: sales
F. Custom report builder	25	€190,000	€300	2: sales, CSM notes

Pass 2: score and decision

Theme	Reach	Impact	Confidence	Effort (person-months)	Score	Fit	Decision and one-line rationale
B	60	2	100%	1	120	2	Build now. Highest score, cheapest fix, and it removes €4,200 a month of support cost.
A	85	2	100%	1.5	113	2	Build next. Over the at-risk threshold, so it is confirmed with leadership; first test whether a status message and retry solve most of it, since the failures sit with two bank partners.
D	210	0.25	50%	0.5	52.5	0	Decline. The score beats C only because a survey vote inflates reach; one source, minimal impact on retention, and nothing to do with the strategy.
C	140	1	80%	4	28	1	Plan for next quarter, after a usage check. Confirm how many accounts capture receipts on mobile weekly before committing four person-months.
F	25	1	80%	6	3.3	0	Decline, by leadership. Over the at-risk threshold, so leadership made the call: the strategy is staying simple for small finance teams; offer the scheduled export and a report template instead.
E	1	1	50%	2	0.25	0	Decline for now, with a revisit condition. One account and two prospects; revisit only if leadership decides to sell to larger companies, which is a strategy decision, not a backlog one.

Three things in the sheet are worth more than the ranking. Dark mode outscores mobile capture, and the rationale column says why the score is wrong there: reach from a one-click survey vote is not the same kind of reach as 140 accounts writing reviews. Theme A gets built and still goes to leadership, because the threshold exists to make large revenue visible, not only to block things. And themes E and F are declined with a condition or an alternative attached, which is what makes the next step, telling the customers, possible.

When should you decide explicitly not to build?

Decide not to build when any of these is true, and write the reason in the sheet:

• The theme has strategic fit 0 and leadership has not chosen to change the strategy.
• The underlying problem has a cheaper answer: documentation, an existing feature the customers have not found, a workaround or a partner tool.
• The effort is above a few person-months and the evidence comes from a single source.

An explicit "will not do" is a decision with an owner and a date. A theme left at the bottom of the backlog is a decision nobody made, and it comes back in every quarterly review with a slightly bigger count. Once the decision is made, the customers who asked need to hear it, in the channel they used; how to close the loop on customer feedback has the decline template and a worked example for theme F.

Is a scoring sheet false precision?

Partly, and McBride says so himself: "RICE scores shouldn't be used as a hard and fast rule. There are many reasons why you might work on a project with a lower score first." An impact of 2 against 1 is a judgement, an effort estimate of 1.5 person-months is a guess, and a score of 113 against 120 is a tie.

The sheet earns its place by making the disagreement specific. When sales argues for single sign-on, the conversation moves from "our biggest prospect needs it" to "reach is 1, fit is 0, so this is a question about selling to larger companies". When support argues for the export fix, the €4,200 a month is on the table. How to use AI for prioritization makes the same point about AI-generated scores: treat the number as an argument to test, not an answer. For the general framework without the feedback inputs, see the RICE prioritization framework.

What keeps the sheet trustworthy from quarter to quarter?

The reach column is only as good as the theme definitions behind it. If "export problems" quietly grows to include formatting complaints, the count rises and the score rises with it, with no change in customers' experience. A fixed, versioned codebook, as described in AI for customer feedback analysis, keeps a theme meaning the same thing in March as it did in December.

Voice of the Customer is a 1 week Builders Camp bootcamp with 10 self-paced microlessons. Its public syllabus covers prioritisation with VoC, combining customer evidence with impact, effort and strategy to pick what to build next, alongside taxonomy and tagging, synthesis into insights and closing the loop.

See the Voice of the Customer bootcamp

Before your next review, recount the top three themes by distinct accounts instead of mentions. If the order changes, the old list was measuring how often people wrote, not how many people had the problem.