Why the complaints inbox is a bad thermometer

Most businesses do measure satisfaction, they just do it by accident. They read the complaints that arrive and assume the silence is approval. It is not. Complaining takes effort, and the customer who cannot be bothered does something cheaper instead: they order somewhere else next time and never explain why.

That is the whole reason to measure deliberately. Not to produce a number for a slide, but to hear from people who would otherwise leave without a word. Everything below is in service of that.

Three scores, three different questions

These are the standard ones, and they are not interchangeable. Picking the wrong one is how you end up with a healthy number and an unhealthy business.

If you run only one, run CSAT on the moments that matter and keep the free text field. It is the cheapest to collect and the easiest to act on, because the answer arrives attached to a specific order you can go and look at.

NPS, worked through once

NPS is the one people calculate wrong, because the middle of the scale is thrown away. Answers are split into three groups, and the passives in the middle count towards the total but not towards the score.

The result runs from minus 100 to plus 100, which is why it cannot be read as a percentage. Note what the passives do: 35 people who are quietly fine move the score not at all, yet they are the group most likely to leave for a slightly better price. The score does not see them coming.

Ways to find out that are not a survey

A questionnaire reaches people who are willing to fill in a questionnaire. These reach the rest, and the first one on the list is worth more than any score.

• Lost customer analysis. Take the customers who used to order and stopped, and ask them why. Uncomfortable, cheap, and by far the most informative thing on this list, because these are the people your survey never heard from.
• Complaints and suggestions, treated as data. Not just answered one by one, but counted and grouped. Five complaints about the same thing in a month is a pattern, not five bad days.
• Mystery shopping. Somebody buys from you as an ordinary customer and reports what it was like, including from your competitors for comparison. It shows you what the process feels like rather than what it looks like from inside.
• Focus groups. Eight to twelve customers and a moderator who only steers. Slow and qualitative, useful when you do not yet know which question to ask.
• Asking your own staff. The people answering the phone know exactly what customers complain about, usually months before it shows up in a score. Nobody asks them.

Four ways the number lies

• Only the invested answer. Delighted and furious customers reply, the indifferent middle does not. Your score is therefore a measurement of your extremes, and it usually looks better than reality because the furious have already left.
• Asking too late. A question sent weeks after the order measures memory, not experience. Ask within a day or two of the thing you are asking about.
• One number for everything. A company-wide score averages your best channel with your worst. Split it by what you can actually change: by courier, by product group, by whether the order was late.
• Collecting a score without a reason. A 6 tells you there is a problem, not what it is. One free text box, asking why, is worth more than three extra rating questions.

Making it something you act on

• Tie every answer to the order it came from. A score on its own is a mood. A score next to an order you can open, with its delivery date and its courier, is a cause you can chase.
• Follow up on the low ones individually. Not to defend yourself, but to ask. Some of them come back, and all of them tell you something.
• Watch the trend, not the value. Whether 25 is good depends on your market. Whether it was 32 last quarter does not.

Where the answers come from

Two things make this practical rather than a project. The first is asking automatically at the right moment, which means the trigger has to be the order status and not somebody’s calendar. The second is that the answer has to land next to the order, so that a low score is traceable to a late delivery or a particular courier rather than floating in a separate survey tool. If you want to remove the most common reason for a low score before you even measure it, start with telling customers where their order is.