How do you spot repeat customer complaints across customers?

This method is for business-to-business complaint records on chemical products, where each complaint names a customer, a product and a date. It does not replace a regulated complaint procedure, such as device complaint files under 21 CFR 820 or medicinal product quality defect handling under EU GMP Chapter 8, and it does not decide a recall.

What do you need before you look for repeat customer complaints?

Before you look for repeat customer complaints you need complaint records with five fields, one fixed code list and a written analysis plan dated before any counts are read. The five fields are the customer, the date received, the product, the lot or production period, and the customer's application or industry. Keep the customer's own words in a sixth field, apart from the code.

The plan is one page. It names the period, the scope, the comparison axes and the review rule, and it carries the date it was written. Writing it first matters because axes picked after looking at the data tend to confirm what the reader already suspects.

The five steps, each a single action:

1. Fix the period and the scope, for example the last 12 complete months and one product family (illustrative).

2. Fix 3 to 4 comparison axes: industry, product, production period or lot range, and application.

3. Code each complaint to one symptom code from a list of about 15 to 30 codes (illustrative).

4. Count distinct customers per code, not complaint records, so one vocal customer does not look like a trend.

5. Compare each count with the code's usual rate, then check whether those customers concentrate on one axis value or spread across all of them.

Records that lack the lot or production period can still be counted in step 4, but they cannot be compared in step 5.

Why does handling complaints one at a time hide recurring problems?

Handling complaints one at a time hides recurring problems because each case closes on its own evidence, while a shared cause shows only when several customers' records are read side by side. In an illustrative case, one technical service engineer sees haze in a coolant at a machining plant; another sees a white deposit at a second plant three weeks later. Each checks the certificate of analysis, finds the lot within specification, and closes the case. Nobody sees that both lots came from the same production week.

Quality guidance in regulated industries builds the cross-case review into the investigation itself. EU GMP Chapter 8 says quality defect investigations should include a review of previous quality defect reports "for any indication of specific or recurring problems requiring attention" (European Commission, 2014, section 8.12). The same chapter says that when a defect is found or suspected in a batch, checking other batches, and in some cases other products, should be considered (European Commission, 2014, section 8.11).

That guidance is written for medicinal products, but the logic carries to industrial chemicals: a single complaint answers whether one delivery met its specification, and only a grouped count answers whether the product, a lot range or an application is failing. ISO 10002:2018 likewise lists analysing and evaluating complaints, so the organization can improve its products and services, as part of complaints handling (ISO, 2018).

Which period, scope and comparison axes should you fix before reading complaint records?

Before reading complaint records, fix a period of complete months long enough to give a baseline, a scope that states which products, regions and channels count, and 3 to 4 comparison axes. Fixing them first stops the analysis from drifting toward the first cluster that catches the eye.

Period. Use complete months or quarters, and compare the latest period with several earlier periods of the same length, for example the latest quarter against the four before it (illustrative). A single quarter has no baseline to compare with.

Scope. State which product family, which regions and which complaint channels are included, and whether technical questions without a defect claim count. If the logging channel changed during the period, for example a new form or a new team, write that down: a rise that starts on the day the logging changed may be a recording effect, not a product effect.

Comparison axes. These four axes are the ones this method compares:

• Industry of the customer, such as automotive parts or food packaging.

• Product, at the level of the product code the customer buys.

• Production period or lot range, which ties complaints to a manufacturing window.

• Application, such as the substrate, process or equipment the product is used on.

Write the axes into the plan with the date. Adding an axis later is allowed, but record it as a new question and test it on a fresh period.

How do you code complaints so different customers' wording matches?

To make different customers' wording match, code each complaint by the observed symptom and where it was observed, using one fixed list with written definitions, and never by the suspected cause. "Cloudy", "hazy" and "milky" become one code such as appearance: haze in use. A cause code like contamination would decide the answer before the analysis starts.

Give each code a one-line definition plus one example that belongs and one that does not. Keep the customer's exact words in their own field so a later reader can check the coding.

Split complaints that contain more than one issue. Consumer feedback guides make the same point: breaking long feedback into separate units "ensures each issue is categorized correctly" (Clootrack, 2025). A complaint about both foaming and residue gets two codes, each counted once for that customer.

Test the list before trusting the counts. Have two people code the same 20 records independently (illustrative) and compare. Where they disagree, the definition is unclear; fix the definition rather than the record. Keep an unclear code for records that fit nothing, and review it each period: a growing unclear code often hides a new symptom.

Regulated systems recognize similar complaints too. The US device rule lets a manufacturer skip an investigation already performed for a similar complaint, provided it records the justification (21 CFR 820.35(a), 2026). A clear code list is what makes "similar" defensible.

How do you tell a real repeat complaint pattern from coincidence?

A real repeat complaint pattern has more distinct customers for one code than that code's usual rate explains, and those customers share a value on at least one comparison axis. Coincidence is a count inside the usual range, or customers spread across every product, lot and application.

For the rate, a counts control chart gives a simple rule. Its upper limit is the average count per period plus k times the square root of that average, and many practitioners set k to 3 (NIST/SEMATECH e-Handbook, section 6.3.3.1). If a code averaged 2 customers per quarter, the limit is 2 plus 3 times 1.41, about 6.2, so 7 customers in one quarter is a signal (illustrative). The rule assumes counts behave like a Poisson process, which is shaky when a period holds only one or two records. It also assumes each period covers about the same inspection unit (NIST/SEMATECH e-Handbook, section 6.3.3.1). Here that means about the same number of active customers; if that changed by much, compare customers per 100 active customers instead of raw counts.

The check table below decides whether a cluster of complaints is a recurring pattern or noise from one customer or one event.

The deciding row is concentration: 4 customers below the rate limit who all used one lot range still deserve a lot investigation, while 8 customers spread across every lot point to an application or logging question instead.

What should you do once a repeat complaint pattern shows up?

Once a repeat complaint pattern shows up, open one investigation that covers every affected customer, check the other lots in the same production window, and rerun the same plan next period to see whether the count falls back inside the limit.

Start from what the customers share. If they share a lot range, pull the retained samples and batch records for that range and its neighbors. EU GMP Chapter 8 frames the same step for medicinal products: consider checking other batches, and other batches that may contain portions of the defective batch or components should be investigated (European Commission, 2014, section 8.11). If they share an application, compare the conditions on site, such as substrate, water quality or temperature, before suspecting the product.

Record the corrective action and how its effect will be judged before it starts. EU GMP Chapter 8 says the effectiveness of corrective and preventive actions should be monitored and assessed (European Commission, 2014, section 8.18). Use the same code, the same axes and the same limit, so the after-count is comparable.

Call in the product's technical team, or a statistician, when counts are too small for the rate rule, when two axes overlap, or when a safety or regulatory report may be due. Lubinpla runs this kind of trend analysis on your own complaint records and site history: it fixes the period, scope and comparison axes before reading results, looks for repetition and concentration across customers and periods, and returns one cited document for your review.