How do I find what customers keep calling about?

In this guide

• Start with a week you can account for
• Make a question ledger
• Group by the answer the customer needs
• Count calls, not repeated sentences
• Turn one pattern into a small change
• Where Call Nerd fits

THE SHORT ANSWER

Start with one complete week of calls. Write down the customer’s actual question, group questions that need the same answer, and count how many calls contain each one. Then review the evidence behind the largest group before deciding what to change.

You remember three difficult delivery calls. Your colleague remembers a run of refund questions. Neither memory tells you what the whole week looked like.

You do not need an elaborate dashboard to get a first answer. You need a consistent record, a sensible grouping rule, and permission to leave uncertain calls unclassified.

Start with a week you can account for

Choose a complete Monday–Sunday period and one clear scope, such as the customer-service phone. Record the total answered calls, how many have a usable transcript or note, and how many you could not review. Keep test calls separate.

If you reviewed 24 of 80 calls, write “24 reviewed calls,” not “what our customers asked this week.” A convenience sample may miss particular shifts, languages or call types. Review everything when the volume is manageable; otherwise use the same sampling rule each week and describe it.

Use records your organisation is authorised to review. Keep customer names, phone numbers and order details out of the working summary when they are unnecessary.

Make a question ledger

One row should describe one customer question. A single call can have several rows. Start with these columns:

Column	What to record
Call reference	An internal ID that lets an authorised reviewer find the evidence
Customer question	The wording in the transcript, or a clearly labelled paraphrase
Shared question	A short name for questions needing the same answer
Evidence	Transcript timestamp, excerpt, or note reference
Outcome	Answered, follow-up promised, unresolved, or unknown
Coverage note	Unclear audio, incomplete note, or uncertain translation

Intercom’s conversation topics documentation describes a related approach: categorising conversations and reviewing results by topic. Your first version can be a spreadsheet. The value is in the grouping and source evidence.

Group by the answer the customer needs

“Delivery” is a useful broad topic, but often too broad to tell you what to fix. These fictional examples belong to different questions:

• “When will it arrive?” → expected delivery time.
• “Can I send it to my office instead?” → changing the delivery address.
• “It says delivered, but nothing is here.” → missing delivery investigation.

Different wording can describe the same need. The same word can describe different needs. Ask: would the same answer and next action help both customers? If the answer is no, keep the groups separate.

Have a second person review a few ambiguous examples. Write down the distinction you agreed on, then revisit earlier rows affected by that decision. Keep an “unclear” group instead of forcing a confident label.

Count calls, not repeated sentences

For each shared question, count distinct call references. If a customer asks when their order will arrive four times during one conversation, that is one call containing the question.

Report a count and its denominator: “Delivery-time questions appeared in 9 of 30 reviewed calls.” Because one call can contain several questions, percentages across groups may add up to more than 100%.

Do not confuse a repeated topic across customers with a repeat contact from the same customer about the same unresolved issue. Measuring the second requires a reliable case link and an agreed time window. Similar wording alone does not establish it.

Turn one pattern into a small change

Take the largest useful group and read several underlying calls. Check whether the likely problem is missing information, a broken process, an unclear promise, or something you still cannot explain.

For a delivery-time question, the next step might be checking whether the confirmation message gives a meaningful date. For address changes, it might be explaining the cut-off before dispatch. These are hypotheses to investigate, not conclusions that a topic count proves.

Assign one owner and a review date. After the change, compare a similar complete period using the same coverage and counting rules. Also check whether calls moved to another channel or became harder to make. A smaller count alone does not prove a better experience.

Where Call Nerd fits

Call Nerd is built for teams handling ordinary cellular calls on company-controlled Android phones. It can turn usable recordings into transcripts, recurring questions and management reports in a private deployment.

Capture quality depends on the device, carrier and microphone conditions; a missing customer voice cannot be repaired by a better question label. Manual recording and audio import are fallbacks. Start by verifying the capture workflow, then use the ledger above to judge whether the resulting analysis is useful.

Once the questions are grouped, turn them into a short weekly brief. Keep the source references attached so the next discussion starts with evidence.