A shopper checks your app while standing in the aisle. It says the item is in stock at this store. They came in for it. It is not on the shelf, and it is not out the back.
No one lied to them. The system did.
This happens quietly thousands of times a day in retail. In 2024 it happened loudly, and in court. Air Canada was ordered to compensate a customer after its website chatbot described a refund policy that did not exist. The tribunal found the airline “did not take reasonable care to ensure its chatbot was accurate,” and held it responsible for what its AI had said.
The lesson is not that AI is risky. It is that an AI states everything with the same confidence, true or not, and the customer cannot tell the difference.
The confident wrong answer
A customer-facing AI feels authoritative. It answers instantly, in full sentences, without hedging. That is exactly what makes a wrong answer so damaging.
When a person is unsure, they say they will check. When an AI is wrong, it usually sounds just as certain as when it is right. New York City learned this with its official business chatbot, which told owners they could do things that were actually illegal, such as taking a share of workers’ tips. The bot was not broken. It was confidently presenting bad information as fact.
For a retailer, the stakes are quieter but constant. Every time an AI answers a question about stock, price, or delivery, it is making a promise on your behalf.
Most AI is failing, and it is not the AI’s fault
The rush to add retail AI is hitting a wall. MIT’s NANDA research found that about 95% of generative AI pilots fail to deliver measurable business impact. That number is not a story about weak models. It is a story about what the models are fed.
Gartner makes the cause explicit. It has predicted that organisations will abandon 60% of AI projects through 2026 if they are not supported by AI-ready data, meaning data that is accurate, current, and connected. AI does not fix a data problem. It amplifies one. Point good AI at unreliable stock and price data, and it will give unreliable answers faster, more confidently, and in front of more customers.
Retail’s data rarely agrees with itself
This is where retail is especially exposed. The information needed to answer one simple availability question is usually scattered across systems that do not fully agree.
The point of sale holds what has sold. The order management layer holds what is committed and on the way. Inventory sits across both, while adjustments, transfers, and returns move through their own paths. Salesforce found that only 17% of store associates have a unified view, and that new associates now have to learn an average of 16 different systems, up from 12 in 2023.
Picture an associate with a customer waiting. The screen says the item is available, so they say yes. The shelf says otherwise, and the stockroom agrees with the shelf. The associate was not careless. They read the one screen they had, and the screen was wrong.
An AI is in exactly that position, at scale. It reads whichever partial, lagging source it is wired to and answers as if that were the whole truth.
What a wrong answer costs
A confident wrong answer is expensive in three ways.
It loses the sale in front of you. Baymard Institute puts online cart abandonment at around 70%, and a wrong price or a sudden “no longer available” is exactly the friction that tips a shopper into leaving.
It loses the next sale too. PwC found that 32% of customers would stop doing business with a brand they love after a single bad experience. Mislead someone once and you damage a relationship you paid to build.
And it fails at the thing customers care about most. Zebra found that product availability is the number one consideration for shoppers when they decide where to buy. An AI that gets availability wrong is failing at the exact moment that decides where people shop.
The guardrail is better data, not a better script
The fix for an AI that says the wrong thing is not smarter wording. It is a more reliable source underneath it.
Krisp Systems helps retailers connect POS, orders, inventory, and fulfilment into one operational view, so the stock and price an AI or an associate draws on reflect what is actually happening across the business. Given one trusted, current source, the AI has something solid to stand on. It can tell a customer an item is in stock because it is, and quote a price because it is still the price.
Connecting that data is not a step you take after deploying AI. It is the step that decides whether the AI helps customers or quietly misleads them.
The practical takeaway
Before adding customer-facing AI, ask one plain question. Can the business already give a single accurate answer about what is in stock and what it costs?
If that answer lives in several systems that disagree, AI will not settle the argument. It will broadcast it, confidently, to your customers. AI will happily tell your customers anything. Connected data is what makes sure it is true.
FAQs
Why does retail AI give customers wrong information?
Usually because the data it draws on is fragmented or out of date. If stock and price sit in several systems that do not agree, the AI answers from a partial view and presents it as fact.
What does “AI-ready data” mean?
Data that is accurate, current, and connected enough for AI to rely on. Gartner has warned that organisations will abandon most AI projects that are not supported by AI-ready data.
How does bad AI advice hurt a retailer?
It loses the immediate sale, damages trust, and can create liability. In one case a tribunal held an airline responsible for a refund policy its chatbot invented.
Can AI fix inaccurate inventory data?
No. AI amplifies the quality of the data beneath it. If inventory records are unreliable, AI makes those errors faster and more visible rather than correcting them.
What should retailers do before deploying customer-facing AI?
Connect the underlying data first. When POS, orders, inventory, and fulfilment share one accurate view, AI has a reliable basis for the answers it gives customers.
Want your customer-facing tools, AI included, to answer from one accurate view of stock and price? Talk to Krisp Systems about connecting POS, inventory, orders, and fulfilment into one operational view.
