Customer feedback management
Forty-one pieces of feedback across three systems pointed to the same problem.
FlowDesk had support conversations, sales notes and churn interviews, but nobody had analysed them together. Shorter Loop grouped the evidence into themes and connected the largest cluster to a product opportunity.
Feedback becomes more useful when related comments can be analysed together instead of one item at a time.
Step 01
The feedback had been collected, but not analysed as one body of evidence
FlowDesk sells shift-scheduling software to mid-sized logistics operators. It has 400 customers, mostly in the Benelux.
The company had 1,100 support conversations, notes from 60 sales calls and 14 churn interviews that had been filed in February.
Each source had been captured for a valid reason, but the sources lived in different systems. Nobody had reviewed them together to see whether the same problems appeared across support, sales and research.
Reading feedback one item at a time makes recurring patterns harder to see.
Step 02
Detect repeated needs even when customers use different words
Semantic duplicate detection groups related requests by meaning rather than relying only on shared keywords.
Customers did not ask for “a second seat” using the same language. They talked about screenshots, texting the roster, view-only access and giving the depot team access to schedules.
Individually, these comments looked like small requests. Analysed together, twelve customers were describing the same underlying access problem. A simple keyword search would have missed the connection because the comments shared almost no vocabulary.
Step 03
Group feedback by the problem it describes
The source still matters, but the theme tells the product team what customers are repeatedly struggling with.
Grouping by source shows where feedback arrived. Grouping by theme shows which customer problems recur across those sources.
The three largest FlowDesk clusters contained 41, 18 and 11 items. The largest cluster combined support tickets, sales-call notes and the February interviews, which had never previously been reviewed together.
The second cluster concerned setup and import effort and contained 18 items. That mattered because the CEO's preferred solution assumed setup difficulty was the main cause of churn; the larger cluster pointed elsewhere.
Step 04
Use feedback to test the explanation behind the roadmap
The 41-item cluster changed how the team understood early churn before it committed to a solution.
FlowDesk's Q4 objective assumed onboarding difficulty was driving churn. Cross-referencing the 41 items with the churn cohort challenged that explanation because every account in the cluster had already completed onboarding.
Those customers had imported rosters and published a first schedule. The recurring problem appeared afterwards: one person remained alone in the product and customers kept asking for a way to share schedules with others.
The most useful feedback that quarter changed the team's diagnosis before engineering work began.
Step 05
Turn a supported feedback cluster into a product opportunity
The opportunity keeps the original requests attached so the team can inspect the evidence behind the decision later.
The largest cluster became an opportunity: the second user never arrives. All 41 feedback items remained linked to it in both directions.
Opening the opportunity shows the original customer comments. Opening any of the 41 feedback items shows what decision it contributed to. The connection remains available months later without relying on somebody's memory.
The opportunity could then be scored against the objective, followed by solution options and an experiment to test the most important assumption.
Traceability lets a product decision point back to the customer evidence that influenced it.
Step 06 · 14 November
Tell customers when the problem they raised has been addressed
Because the release remained linked through the opportunity to the original feedback, FlowDesk already had the list of customers who had raised the problem.
All 41 customers were notified. Nine replied and two became reference customers. The feedback record therefore continued beyond collection and prioritisation into customer follow-up.
Ideas and pain points, side by side
Fair questions
What people ask about feedback management
- Where does customer feedback come from?
Feedback can come from support tools, chat, surveys, analytics, the Shorter Loop feedback portal or bulk imports. In the FlowDesk example, the most useful sources were support conversations, sales-call notes and previously completed interviews.
- How does duplicate detection work if people use different words?
It compares meaning rather than relying only on matching words, then presents likely duplicates with a similarity score for review. The twelve related requests in the FlowDesk example used different wording.
- Does every request have to become a feature?
No. A cluster can be promoted to an opportunity, where it still has to be evaluated against an objective and other evidence. Most feedback should not become a feature automatically.
- Can customers see what happened to their request?
Yes. The relationship runs both ways, so a shipped release can show which feedback items it addresses and the team can identify the customers who raised them.
- We only get a handful of requests a month. Is this overkill?
The value comes from reading across sources rather than from volume. The reframe in the demo comes from 41 items, and it had been missed for months because those 41 lived in three different places.
- How long does it take to get anything out of it?
It depends on how fast you can connect what you already have. For a ten-person team that has typically taken about five days.
Bring together the feedback you already have.
Connect support conversations, sales notes, research and other sources so repeated customer problems can be analysed together and traced into product decisions.
FlowDesk is the Shorter Loop demo workspace. You can inspect the feedback clusters, opportunity links and release traceability used in this example.