The Three Words Nobody Is Allowed to Say

"Is this one waterproof?"

It's a reasonable question about a jacket, asked by someone who is going to buy it either way, and the associate has worked here two years.

She doesn't know. The label says water-resistant. There's a symbol she's not sure about. Nobody has ever told her the difference and the man is waiting.

"Yeah, should be fine."

Nothing dramatic happens. He buys it, he wears it in October, it isn't, and he comes back annoyed — at which point the store refunds a worn jacket, and a two-year associate who is good at her job has cost more than the sale was worth, because of three words she didn't feel able to say.

The smallest scenario in the catalogue and the most cross-cutting

This one isn't a difficult conversation. There's no conflict, no complaint, no pressure. It's a single sentence, and it turns up inside every other scenario in all three series: the return that gets refused for an invented reason, the compliance question answered approximately, the delivery date guessed at.

Three things make saying I don't know in customer service genuinely hard, and none of them are about knowledge.

It reads as incompetence to the person saying it. Not to the customer — to the associate. The felt experience of "I don't know" is failing at your job in front of someone, which is why the guess arrives so fast.

Service culture codes it as bad service. Decades of training language — be the expert, be confident, never say no — have made the honest answer feel like the prohibited one. Mystery shopping frequently scores it down.

And the guess is invisible at the point it's made. Nobody in the building sees the sentence. The consequence arrives weeks later as a return, a complaint or a dispute, attached to nothing. There is no feedback loop at all, which is why the behaviour never self-corrects.

The replacement is three parts, not one

The problem with "I don't know" as advice is that on its own it's a dead end, and people correctly sense that. The usable version has three components and takes four seconds.

What you don't know, specifically. "I don't know whether that's fully waterproof or just water-resistant — they're different things and I don't want to guess." Precision here reads as expertise rather than ignorance; she has just told him something useful about the category.

How you'll find out. "The spec's on the tag inside or I can check the system." Naming a method converts a gap into a process.

And when. "Two minutes" — or "I'll ring you before five", or "the manager's in tomorrow at nine." The commitment is the part that makes the honesty acceptable, and it's the part most often dropped.

Three components. Almost everything that goes wrong is one of them missing.

The two kinds of not knowing

Worth separating, because they need different handling and people conflate them.

"I don't know the answer" — a fact exists and the associate doesn't have it. Go and find it. Low stakes, fully fixable, and often two minutes of effort.

"I don't know if we can do that" — a decision hasn't been made, or the authority sits elsewhere. This one cannot be resolved by looking harder, and the correct move is to say so and escalate rather than to keep searching for an answer that doesn't exist yet.

Associates who don't distinguish these spend a long time hunting for a policy that was never written, and then invent one. And when a question turns out to sit in the second category repeatedly, that's worth reporting — the gap is organisational, not personal.

And the things that make it worse

"I think" — the most dangerous two words on a shop floor. It sounds like a hedge and is heard as an answer; the customer repeats it later without the qualifier.

Answering a slightly different question, which happens when someone doesn't want to say they don't know but does want to say something.

Deferring without a time. "I'll find out" with no clock attached is functionally the same as not answering, and it's the most common version.

And going quiet. Five seconds of silence while someone decides whether to guess is read as evasion.

Four ways it goes wrong

The confident guesser, who answers fluently and wrongly. The costly one, because nobody can tell it happened until the consequence arrives.

The hedger, whose "I think" and "should be" are heard as yes.

The open-ended deferrer, who promises to find out and attaches no time, so nothing happens.

The over-discloser, who turns an honest gap into a monologue about how nobody tells them anything — accurate, unhelpful, and it undermines the customer's confidence in the store rather than building it.

Why this isn't trained

Product training rewards recall. Tests, quizzes, knowledge scores. Nothing in that structure teaches what to do at the edge of what you know, which is where the job actually happens.

Nobody has written the replacement sentence. Associates are told not to guess. They are almost never given the three-part alternative, and in the absence of a script under pressure, the guess is the default.

Scorecards can penalise honesty. Where mystery shopping or NPS treats uncertainty as poor service, the operation has purchased confident misinformation and called it a service standard.

And peer role play never asks the question. A colleague running a practice conversation asks what they know the associate can answer — it's the natural, cooperative thing to do. The whole scenario requires a counterpart willing to ask something genuinely unanswerable, which a helpful colleague will not do.

What accuracy-under-uncertainty training can rehearse

A simulation can ask the question the associate can't answer — repeatedly, in different registers — and score whether the gap was named, a method offered and a time committed. It's the only practical way to rehearse a behaviour that by definition can't be scripted in advance. Foretell AI supplies the counterparty configuration, transcripts and rubric-based scoring; the product knowledge, escalation routes and information sources stay with the retailer.

Four to build:

Designing the module

Pass one — the gap. Score whether the associate answered, guessed, hedged with "I think," or named the gap specifically.

Pass two — the follow-through. Score whether a method and a time were both given.

Pass three — the pressure. Score the response to "just give me a rough idea," which is where most modules find the real behaviour.

Rubric on observable behavior: Was an answer given without certainty? Was "I think" or "should be" used? Was the gap named specifically? Was a method of finding out offered? Was a time committed? Was the promise kept within it?

Hedge-word frequency is trivially countable in a transcript and it's the best available proxy for guessing. Most operators have never counted it and could start with recordings they already hold.

The operator case

Guesses become obligations. An invented answer is a representation made by your business, which you then either honour at a cost or refuse at a complaint. Both are more expensive than two minutes of checking.

The questions people can't answer are free product research. A simple log of what associates got asked and couldn't answer is the cheapest content, training and FAQ input available, and almost nobody keeps one. The gaps repeat.

Your scorecard may be buying the wrong behaviour. If confidence is measured and accuracy isn't, you have incentivised the guess. That's a measurement decision, and it's reversible.

And the standard you set for automated channels should apply to people. Operators scrutinise whether an automated system might answer confidently and wrongly, at length, with governance attached. The same failure on the shop floor is unmeasured, unlogged and roughly as common — a useful thing to point out in a meeting where one of those is being discussed and the other isn't.

For customer experience programmes, this is a good illustration that trust is built by calibration rather than confidence: customers do not require staff to know everything, and they do require the answers they get to be reliable.

Frequently asked questions

Is it okay to tell a customer "I don't know"? Yes, when it comes with the other two parts — what specifically you don't know, how you'll find out, and by when. On its own it's a dead end; with those, it reads as care rather than ignorance.

How do you avoid guessing in customer service? By having a replacement ready. People guess under pressure because nothing else is available in the moment. A rehearsed three-part sentence removes the vacuum that the guess fills.

What's wrong with saying "I think" to a customer? It's heard as an answer, not a hedge. The customer repeats it later without the qualifier, and the store is held to it.

How can retailers reduce inaccurate answers from staff? Log the questions associates couldn't answer, check whether the scorecard penalises uncertainty, and rehearse the replacement sentence. The guess is a structural response to a missing script, not a character flaw.

The short version

She didn't know whether the jacket was waterproof, and the store never told her that not knowing was allowed — so she said "should be fine," and it cost a refund on a worn coat and a customer who now checks.

The fix isn't knowledge. It's four seconds: this specifically is what I don't know, here's how I'll find out, here's when.

And someone senior has to say out loud that using it is the right answer, because every instinct on the floor says otherwise.

Foretell AI lets retailers build conversational simulations — including knowledge-gap, accuracy and escalation conversations like the one above — with configurable counterparties, transcripts, recordings, and rubric-based evaluation. If nobody is logging the questions your associates couldn't answer, we're happy to walk through how other operators have structured it.