Feedback Triage Scoring

Feedback Triage & Scoring

Why a blended framework

Any single lens on feedback misleads you in a predictable way:

• ARR-only ranking lets one loud, high-value account dictate roadmap even when the ask is niche to them.
• Frequency-only ranking under-weights a small number of very high-value accounts asking for something that would prevent churn.
• RICE-only ranking (Reach × Impact × Confidence ÷ Effort) is a good general-purpose tool, but on its own it treats "10 free-tier users mentioned this" the same as "10 enterprise accounts mentioned this," and it has no natural way to represent revenue risk.

The point of blending them is that each one catches what the others miss. Don't average them into one number and stop thinking — the goal is to run all three lenses and then make a judgment call informed by all three, not to mechanically produce a single score and rank by it blindly.

Process

1. Normalize the raw feedback

Before scoring, make sure every item has:

• Account(s) attached (if known) and their ARR
• Count of independent instances — separate mentions from separate people/accounts, not the same person repeating themselves
• Source (support ticket, sales call, review, community, internal)

If an item can't be tied to an account or a count, it can still be scored — it just relies more heavily on the RICE lens (below).

2. Run the three lenses

Lens 1 — ARR exposure Sum the ARR of every account that raised this specific ask (dedup accounts, don't double count if one account mentioned it twice). This produces a dollar figure per item — directly comparable across items without any weighting scheme needed.

Lens 2 — Frequency Count independent instances across all sources. This is your signal for "quiet many" issues — things that don't come from your biggest accounts but show up constantly and quietly erode retention or word-of-mouth.

Lens 3 — RICE (for everything, but this is where "everything else" gets caught)

• Reach — how many users/accounts would this affect if shipped (estimate, not just who's asked)
• Impact — how much does it move the needle for those it reaches (use a simple scale: 3 = massive, 2 = high, 1 = medium, 0.5 = low, 0.25 = minimal)
• Confidence — how sure are you about Reach and Impact (100% / 80% / 50%)
• Effort — person-months to build
• RICE score = (Reach × Impact × Confidence) ÷ Effort

RICE is the common denominator — it's the only one of the three that accounts for cost to build, and it's the fallback for items that don't have strong ARR or frequency signal (e.g., a research-driven insight nobody has explicitly asked for yet).

3. Look for outlier patterns before ranking

Before combining anything, scan for two specific patterns — they change how you should read the data, not just the score:

• The loud few: high ARR concentrated in 1–2 accounts, low frequency elsewhere. Ask: is this genuinely strategic, or are you letting your biggest account set the roadmap? Worth a gut-check conversation with sales/CS before committing.
• The quiet many: high frequency, low ARR per account. This is often where churn risk or word-of-mouth damage hides — individually each account looks small, but the aggregate signal is real.

4. Combine into a working priority order

A workable approach that avoids false precision:

1. Flag anything in the top ~10% of ARR exposure OR top ~10% of frequency as a must-evaluate — these don't get ranked away by a lower RICE score without an explicit decision to deprioritize.
2. For everything else, rank by RICE score.
3. Merge the two lists, with must-evaluate items pulled to the top regardless of RICE, but keep RICE visible next to them — a low RICE score on a high-ARR item is a useful flag that it might be an expensive one-off, not a reason to bury it silently.

Don't hide the individual lens numbers behind one composite score in the output — show ARR, frequency, and RICE side by side so anyone reading the list can see why something ranked where it did, and can disagree with the weighting on an individual item without disagreeing with the whole system.

5. Output format

| Item | ARR exposure | Frequency | RICE | Flag | Notes |
|---|---|---|---|---|---|
| [feedback item] | $X | N accounts/mentions | R×I×C/E = score | must-evaluate / quiet-many / standard | [1 line: why it ranked here, any risk of loud-few bias] |

Sort by the combined judgment from step 4, not by any single column.

Anti-patterns to avoid

• Don't let this become a black-box score nobody can question — the whole point of showing all three lenses is auditability.
• Don't recompute RICE confidence upward just because a deadline is close — Confidence should reflect actual evidence quality, not urgency.
• Don't treat a single whale account's request as automatically top priority without checking whether it's genuinely reach-expanding or genuinely account-specific — that's what the "loud few" check in step 3 is for.
• Re-run this on a regular cadence (e.g., before each planning cycle), not just once — frequency and ARR exposure shift as new feedback comes in, and last quarter's ranking will quietly go stale.

Example (illustrative numbers, placeholder accounts)