AI everywhere: when it hurts the customer experience

Adding AI to a bad process does not make it good. It makes it faster, and its errors more convincing.

AI has made real progress. It summarises a meeting, drafts an email, spots the next action, surfaces information without a sales rep having to type everything. These are real gains, and we set them up every day. But for two years we have also seen the opposite: AI put everywhere, because you have to “have some”, and a customer who pays the price.

When AI works on a bad foundation

If nobody knows what a “qualified” opportunity means, if the pipeline stages match nothing precise, if the starting data is wrong, AI changes nothing. It works on a bad foundation, and it produces summaries, scores and forecasts that look serious.

That is the real danger: a human error is often visible, an error well written by an AI passes for analysis. A qualification score calculated on half-empty records is still wrong, even shown in a nice dashboard.

When the customer pays the bill

The worst case is when misplaced AI reaches the customer directly. The examples are easy to recognise:

• the chatbot going in circles that never lets you talk to a human, when the question is off its script;
• the automatic follow-up that misses the point, sent to a customer who just signed or just complained;
• the wrong but confident answer on a price, a lead time or a warranty, that nobody reviewed;
• the “personalised” email that sounds generic, and that the customer spots straight away.

Each time, the company thinks it is saving time. In reality it loses a customer’s trust, which costs far more than an hour of data entry. We also cover this in our article on AI for customer service: AI is useful there for sorting and preparing, much less for replacing the person who answers.

What to fix before you automate

Before adding AI to a process, three questions are enough:

1. Why does this process exist? What result is expected, for the customer or for the company?
2. Who owns it? A process without an owner degrades, automated or not.
3. What is expected at each step? A clear, written definition the tool can follow.

It is not the most spectacular part of an AI project. It is often the most useful, and it is where we start every engagement, with a process audit before touching the tools. Very often, that audit shows a simple rule is enough, with no AI at all: that is the subject of our article on automation without AI.

Where AI really belongs

AI belongs where it removes useless work from a process that is already clear: data entry after a meeting, the first draft of an email, sorting incoming requests. It does not belong where it serves to hide an organisation that does not work, or where it keeps the customer away from a human at the moment they need one.

So the right question is not “where can we put AI?”, but “what useless work can we remove without the customer noticing?”. A poorly kept CRM stays poorly kept with AI inside: the real reason is often elsewhere, as we explain in why sales reps do not fill in the CRM.

Our view

Putting AI everywhere has become a sales pitch. For the end customer, it is rarely progress. The companies that really benefit from AI are the ones that first put their house in order: clear processes, named owners, reliable data. Only then does AI save time without damaging anything.