Sometimes you ask customers what they think and the answers don't agree. Half say the music is too loud, half love the atmosphere. Service was rushed, says one. Attentive and well-paced, says the next. Three reviews call the food too spicy. Two complain it was bland. You can't act on all of it. You can't really ignore it either.

This is one of the most common reactions people have to running feedback collection for the first time. The data they expected to give them a clear direction turns out to be noisy and conflicting. The natural response is either to start picking the comments that match the conclusion they already had, or to throw their hands up and decide feedback is pointless.

Both are wrong. Reading contradictions is a real skill, and it's mostly about understanding what they actually are.

Why feedback contradicts itself

A few things produce the appearance of contradiction.

The first is that different customers want different things from the same experience. A cafe is a different product to a freelancer with a laptop than it is to a parent meeting a friend with two toddlers in tow. The freelancer wants quiet and reliable wifi. The parent wants high chairs and a tolerant attitude to spilled juice. Both are correct about what a good cafe is, from where they're sitting. The cafe just can't be both at the same time.

The second is that the same person says different things at different moments. A busy Saturday lunch isn't the same business as a quiet Tuesday morning. Service that feels rushed at peak hour is the same service that feels well-paced at 10am. Asking the same question across both will return apparent contradictions that are really two accurate descriptions of two different situations.

Framing matters too. One person's "efficient" is another person's "rushed." One person's "warm" is another's "stuffy." The same underlying reality can produce opposite-sounding answers depending on what the customer happened to think of when they read the question.

And then there's the response filter. The people who feel strongly one way or the other are the ones who fill out the form. Two strong opinions in opposite directions can look like a 50/50 split when the silent middle would actually report something close to "fine."

What to look at instead

The starting move is to stop reading contradictions as contradictions and start reading them as a distribution.

If you have 100 responses and 40 say the music is too loud, 5 say they love the energy, and 55 don't mention the music at all, that's not really a contradiction. That's a clear signal with a small counter-current. The temptation is to weight the 5 positive comments equally with the 40 negative ones because they're rhetorically opposite. They aren't. One is the dominant theme, the other is a sub-cohort.

The pattern to look for is whether the contradiction tracks with a customer segment. Is it new visitors saying one thing and regulars saying another? Morning customers vs evening customers? People who came for a specific event vs walk-ins? Most apparent contradictions resolve cleanly the moment you slice the data by who's saying what.

It also helps to look at what the responses don't disagree on. Feedback that disagrees on the music often agrees that the coffee is good, or that the staff are friendly. The consensus areas are usually where the actionable signal sits. The disagreements are where the trade-off conversation happens.

When both can be true

Some contradictions are real and reflect a genuine trade-off in the business. You cannot run a cafe that's a quiet workspace in the morning and a lively dinner spot in the evening with the same volume and seating plan. You have to pick.

A useful exercise is to take an honest stab at your target customer and weight their feedback more heavily. Not exclusively, more heavily. The cafe that decides it's primarily a weekday workspace and accepts that families with toddlers are not its core audience will have a clearer time interpreting its feedback than one trying to be everything to everyone. The mistake is staying undecided and reading every contradiction as an instruction to do both, which is how you end up with a confused business.

This is also a place where Ask AI features can earn their keep. Querying feedback with prompts like "are new customers and regulars saying different things about service?" or "what do customers under £15 spend say vs over £30?" gives you the cuts that resolve apparent contradictions into clear segment patterns. Qria does this kind of slicing on the responses you've already collected, which makes the conversation about what to actually change a lot less guesswork-driven.

What not to do

A few habits worth dropping:

  • Acting on the loudest comment because it stung. The most strongly worded review is usually not representative of anything, and changing course based on it is a good way to whiplash the rest of your customers.
  • Averaging two opposite comments into a middle position. The "median customer" you'd be optimising for almost never exists.
  • Throwing out the feedback because it's "all over the place." It usually isn't. It's usually a few clear patterns wrapped in some noise.
  • Asking again, and louder, hoping for a clearer answer. Those contradictions are showing you something real about your customer base, not a data quality problem to be fixed.

The post on why people tell you what you want to hear covers a related failure mode, where the data is too clean for the wrong reasons. The two problems sit at opposite ends of the same skill: learning to read what your customers are actually telling you, and what they aren't.