CRM with AI for business: what actually changes

Every CRM on the market has added “AI” in the last year, and for most of them it means the same thing: a little chat window in the corner that summarises what you were already looking at. That changes nothing, because the CRM problem was never reading it.

The four things that actually change

1. The record fills itself in

An email, a form or a WhatsApp message arrives and the contact is created with name, company, phone and what it is about — with nobody typing. That is 90% of the value, and it is the most boring part to show in a demo, so it almost never gets shown.

2. Follow-up does not depend on someone remembering

A process watches the open conversations and flags the one that has been dead for twenty days, with the draft reply already written. Most lost sales are not lost on price: they are lost to silence.

3. The conversation comes in where the client lives

Email, WhatsApp and forms in the same inbox, tied to the record. If your team has to leave the CRM to talk to the client, the CRM will always be out of date, with or without AI.

4. The report writes itself

Nobody does the monthly report because it has to be assembled. When the system has the data and can write, the report goes from two hours to a five-minute review — and then it actually happens every month.

What to ask before signing

Five questions that separate a product from a marketing layer. If the salesperson hesitates on the third, you already know.

1. Does the AI write, or only read? If it only reads, you are paying for a summary.
2. What happens when it gets it wrong, and who finds out? If the answer is “it does not get things wrong”, the conversation is over.
3. How much autonomy does it have? Does it propose, or send? Can it change a client’s status without asking? Can it write to a client with nobody seeing it first?
4. Where does my data live and who trains on it? With European client data this is not a technical question, it is a legal one.
5. Can I switch it off module by module? A system where the AI cannot be disabled in parts is a system that has not thought about your team’s initial distrust, which is real and legitimate.

The third question has an answer you can demand in writing. That is why I published the Vergara Agentic Autonomy Scale: six levels, V0 to V6, that turn “it has AI” into “it operates at V2 with human approval for class B actions”. That is a clause, not a brochure.

Migration: do not bring the rubbish with you

The part nobody mentions about changing CRM is that you will copy the previous one’s problems across. Two that always show up:

• Duplicates by phone number. The same mobile written four ways — with a country code, with spaces, with dashes — creates four records for the same client. Normalise the number to digits before importing, not after.
• Contacts that are the same person with two email addresses. If you do not merge them first, the history is split forever and the automatic follow-up will write to the same client twice.

One afternoon of cleaning before importing saves months of dirty data. And if the new CRM has follow-up AI, dirty data is no longer a cosmetic problem: it is duplicate automated emails reaching your client.

What I built, and why

I built a CRM with AI inside because I needed one in my own companies and none of them did the part that matters: feeding the record without anyone typing. Today it is a product, Koble, and the AI inside answers, qualifies and gets ahead of admin work instead of summarising what you can already see.

In events the same problem has a different shape, and that is Hubents. I am not counting them as a pitch: I am counting them because the four things above are exactly what had to be built when the one suffering the badly fed CRM is you.