Think Build Implement Repeat
London, UK +44 7367 067226
WhatsApp FOLLOW f in X
  1. Home
  2. Blog
  3. Our Complaints Are Logged as 'Other' or Free Text. How Do We Find Out What Is Actually Going Wrong?
Problems We Solve

Our Complaints Are Logged as 'Other' or Free Text. How Do We Find Out What Is Actually Going Wrong?

When complaints are logged as free text or 'Other', patterns stay hidden. We use machine learning to categorise complaints at volume and trace them to causes.

Updated 3 min readBy SpiderHunts Technologies

Free estimateNo obligation

Get a free estimate

Tell us what you need. A senior engineer reads every enquiry.

Takes under a minute. We never share your details.

  • Free consultation
  • No commitment
  • NDA on request

Prefer to talk? Book a free 30-minute call →

Quick answer — TL;DR

If complaints are recorded as free text or dumped into a catch-all category, nobody can see which problems are growing or where they come from. A machine learning classifier trained on a sample your team has labelled can categorise every complaint by issue and likely cause, link it to the product, supplier, site or process involved, and show the trends that are invisible one complaint at a time.

Hundreds of complaints, no pattern

The complaints log is full. Each one was handled: someone apologised, sent a replacement, issued a credit. But when the operations meeting asks what customers are complaining about most, the answer is a shrug. The category field says "Other" for a large share of them, "Delivery" for a lot more, and the real detail sits in free-text notes nobody has time to read in bulk.

So the same problems keep coming back. A packaging fault on one product line, a courier that keeps leaving parcels in the wrong place, a website instruction that confuses people. Each complaint is dealt with on its own, and the cause behind the pile is never fixed.

Why the categories do not help

Complaint categories are usually a short dropdown set up years ago. They describe the symptom at a high level, such as delivery, quality or billing, and say nothing about the cause. Staff under pressure pick the first plausible option or "Other", because choosing carefully takes time and nobody seems to use the field anyway.

The useful information is in the words: what the customer said, what the agent found, what fixed it. Reading a few complaints tells you a lot. Reading thousands is impossible by hand, so the text is effectively lost as a source of insight.

What hidden complaint patterns cost

Hidden patternWhat it costs
Repeated product faultReturns, replacements and credits that keep coming
Supplier or courier problemComplaints you absorb for someone else's failure
Confusing process or instructionsContacts that should never have happened
Slow rise in one issueSpotted only when it becomes a crisis
No evidence for decisionsArguments about priorities based on anecdote

There is also a regulatory side in some sectors, where complaint handling and root cause analysis are expected to be demonstrable. A log full of "Other" is hard to defend.

How we categorise complaints at volume

  1. We export complaint history from wherever it lives, such as Zendesk, Freshdesk, a CRM like HubSpot or Salesforce, a shared inbox or a spreadsheet, together with order, product and site details where they can be linked.
  2. We work with your team to define a category structure that is actually useful: the issue, the likely cause, and the part of the business responsible, at a level of detail people will act on.
  3. Your team labels a sample of complaints against that structure. We use the sample to train a classification model, and where labelled data is thin, we use a language model to suggest labels for people to confirm.
  4. The model then categorises the full history and every new complaint as it arrives, with a confidence level, sending uncertain ones to a short review queue.
  5. We link categorised complaints to product, supplier, courier, site and time, and build a dashboard that shows volumes and trends by cause, with alerts when an issue rises faster than normal.
  6. Corrections from the review queue feed back into training, and the category structure is revisited when new kinds of problem appear.

Agents do not need to change how they log complaints to begin with. The model works from what they already write, and a better dropdown can come later once you know which categories matter.

What the business sees

A clear view of what customers complain about, why, and where it comes from, updated as complaints arrive. The operations meeting starts with the issues that are growing, the products or suppliers behind them, and example complaints to read.

Fixing causes becomes possible. A supplier conversation can come with a count of complaints linked to their deliveries. A product change can be judged by whether the related complaints fall afterwards.

The customer service team benefits too. When a problem is spotted early and fixed at source, the calls and emails it would have generated never arrive, and agents spend less of their day apologising for the same thing.

Is this your situation?

  • A large share of complaints are logged as "Other" or a vague category.
  • The detail is in free-text notes that nobody analyses.
  • The same problems seem to come back again and again.
  • Nobody can say which complaint types are growing.
  • You handle enough complaints that reading them all is not realistic.

FAQ

Frequently asked questions

The questions readers ask us after this guide.

Still have a question?

Ask us directly — a senior engineer will get back to you.

Ask about your project

How many complaints do we need for this to work?

Enough that patterns are hard to see by hand. For the model itself, a few hundred labelled examples across the main categories is often a workable start.

Why not just use a language model on every complaint?

It can work well, especially at first. A trained classifier is often cheaper and more consistent at high volume, and we choose based on your volume and categories.

Can it read complaints that arrive by phone?

If calls are transcribed or summarised into your system, yes. The classifier works on the text, whatever the channel.

What drives the cost?

The number of sources, how easily complaints can be linked to orders and products, and how detailed the category structure needs to be.

What do you need from us?

An export of complaint history, access to order and product data, and a few people who can label a sample and agree the categories.

Keep reading

More on Problems We Solve

Start here

Tell us what you are guessing today

Describe the decision, the data you keep and how it is made now. We will tell you honestly whether machine learning would help, what it would need, and if a simpler fix would do the job, we will say so.

  1. You tell us what you needTwo minutes on the form, or a message on WhatsApp.
  2. A senior engineer reviews itAnd comes back with questions, a realistic range and an honest view on fit.
  3. Free 30-minute scoping callWe talk through scope, options and a realistic estimate — with no obligation.
Free estimateNo obligation

Talk to someone who builds this

Send a short brief and we will come back with an honest view and a realistic range.

Takes under a minute. We never share your details.

  • Free consultation
  • No commitment
  • NDA on request

Prefer to talk? Book a free 30-minute call →