Your CRM Has Four Versions of One Client

Search a client name in your CRM and four records come back. One with the trading name, one with the legal entity, one created by a form submission with a personal email address, and one somebody made last March because they could not find the others.

Everybody knows about this. It gets treated as housekeeping, something to tidy up when there is time, and there is never time. It is not housekeeping. It is the foundation problem sitting underneath most of the other complaints in the business.

What it actually breaks

Duplicates are visible and mildly annoying. The consequences are neither.

Every number is wrong. Customer counts, average value, retention, revenue per client. All computed over a set where some clients are counted four times and some are counted zero times because their work was booked under a different entity.

Nobody can see the relationship. The salesperson looking at one record sees one enquiry from 2023, not eleven years of work across three systems. So they treat a major client like a stranger, and the client notices.

Automation fails silently. This is the expensive one. Anything that routes, matches or updates depends on knowing which record to act on. When there are four, it picks one, writes to it, and the person looking at a different record sees nothing. The system appears not to work, and the conclusion is that automation does not work here.

The cause is upstream of the CRM

The records are duplicated because nobody ever decided what a customer is.

That sounds like a philosophical question and it is an entirely practical one. Is the customer the legal entity or the trading name. If a client has three sites, is that one customer or three. When a person moves firms, does the relationship follow the person or stay with the company. When a group acquires one of your clients, do the records merge.

Those questions have no universally right answer. They have a right answer for your business, and in most businesses nobody has ever written it down. So every system, and every person entering data, resolves it differently, and the disagreement compounds daily.

Write the definition first

Before any cleanup, get three people in a room for an hour: whoever sells, whoever delivers and whoever invoices. Ask them to describe a customer. They will disagree within four minutes, and the disagreement is the whole finding.

Come out with one page: what a customer is, what a site or location is, what a contact is, and which of those the money attaches to. That page is worth more than any software you buy, and it is the specification for everything that follows.

Then decide which system is the master. One system owns the truth about customers, the others receive it. Two systems that both create customers will diverge, always, whatever the integration between them promises.

Why the cleanup never happened

Because matching records requires judgement, and judgement meant a person.

Rules based deduplication catches exact matches and almost nothing else. It cannot tell you that Groupe Tremblay Construction and Tremblay Const. Inc. at a slightly different address are the same business, or that they are deliberately separate entities, because that is a question about the world rather than about strings. So the tooling flagged a hundred obvious pairs, somebody merged them, and the real work was left.

What AI changed here

It makes the judgement, at volume, with its reasoning visible.

Given two records it can weigh the name, the address, the domain, the contacts, the invoice history and the job history together, the way an experienced person would, and say these are the same business and here is why, or these look similar and are not, and here is why. Not a similarity score. An argument you can check.

That turns an impossible project into a reviewable list. It proposes the merges, ranks them by confidence, and a person spends an afternoon confirming the top layer and adjudicating the middle. What was a six month project nobody would fund becomes a week.

Then it stays on. New records get checked against the book on creation, so the form submission from a personal address gets attached to the right company on the way in, rather than becoming duplicate number five.

The three merges to be careful with

Not everything that looks like a duplicate is one, and merging wrongly is harder to undo than leaving it.

Deliberately separate entities. Groups frequently run separate companies for tax or liability reasons and want them invoiced separately. Merging those creates a billing problem and an awkward conversation.

Franchises and branches. The same brand at two addresses may be two independent businesses with different owners. The name is identical and they have nothing to do with each other.

The person who moved. A contact who has changed firms is not a duplicate, it is two relationships, and the history belongs in both places for different reasons.

All three of those are exactly why an AI step that shows its reasoning beats one that returns a score. You can read the argument and see immediately that it has not accounted for something only you know.

Do it before anything else

If you are planning to automate anything, or to build reporting you intend to trust, this comes first. Not because it is satisfying, because everything downstream inherits it.

An automation built on ambiguous records produces confident wrong actions, which is worse than no automation. A dashboard built on them produces numbers people quietly stop believing. Both failures get blamed on the new system, and the actual cause was there before it arrived.

The half day version

• Export your customer list and sort by name. The scale of it is usually visible in ninety seconds.

• Count your ten largest clients and check how many records each has. This is the version of the number that gets a decision made.

• Write the one page definition with the three people who disagree.

• Name the master system and switch off record creation everywhere else.

The first two take a morning and will tell you whether this is a tidy up or a foundation problem. In most businesses over about twenty people, it is a foundation problem.

AI Optimize resolves duplicate records with the reasoning attached so you can check the calls, then keeps the book clean as new records arrive. That work sits under Custom CRM.