We Needed a CRM Manager. So We Hired Claude.

Fun fact: everyone at 9x has had to manage a company's CRM at some point in their career.

Salesforce, HubSpot, Pipedrive, Attio, and a couple more.

Between Jan, PYV and myself, we have spent countless hours inside these systems, across very different companies and very different roles.

None of us was ever hired to do it.

I believe it is a rite of passage for operators, and it happens the same way every time: something looks wrong in the CRM, you go in to fix it, and three weeks later you are the person everyone pings when an issue arises.

The Job Nobody Signs Up For

Almost everything that needs to be done in a CRM is relatively straightforward. Of course, you can spice things up with sophisticated attribution or lead scoring rules, but everything else is “easy”:

• Keeping contact and company records current

• Logging calls, emails, and meetings in the right place

• Updating stages, close dates, owners, and next steps

• Fixing duplicates and filling in missing information

• Pulling lists, answering one-off questions, and preparing the numbers for review

None of these tasks is especially difficult. Most of them take a few minutes. The problem is that there are always more of them, and they all depend on the last person having done their part properly.

What makes it difficult, then?

Not the tasks themselves. What makes it difficult is the volume, the repetition, and the discipline of doing it right every time, for years, without it ever being finished. And it is not up to you alone: one person skips a step or breaks a convention, and the whole system quietly degrades. Getting an entire team to respect a shared discipline is the genuinely hard part, and it has nothing to do with the software.

And Yet: It Is the Foundation

A well-managed CRM is what a solid commercial operation sits on. Your forecast is only as honest as your close dates. Your campaigns are only as good as your segments. Your renewals only happen on time if somebody is tracking them somewhere other than in their head. Your pipeline review is either a decision-making meeting or a data-cleaning meeting, and which one it turns out to be depends entirely on the state of the records.

But a CRM will not build and maintain itself. Not yet, at least. To genuinely support your GTM teams and whatever they are trying to do this quarter, it needs constant care and iteration.

Every solid commercial operation we have seen sits on a CRM that someone is actually managing. Every chaotic one sits on a CRM that everybody complains about and nobody owns.

So the work matters.

It has just always been miserable to do.

The Tool Got More Powerful. The Job Got Harder.

The strange thing is that this should have improved by now.

CRMs have become dramatically better over the last twenty years. What used to be a database with contacts in it is now a platform: sequences, enrichment, custom objects, automations, call recording, reporting, scoring, and dozens of integrations writing into it from every direction. Each of those capabilities is genuinely useful. We use most of them.

But every capability you add is also a surface that has to be maintained. More automations means more automations that can break silently. More integrations means more sources creating records with different conventions. More custom objects means more definitions that have to stay coherent with each other. More power in the tool means more complexity for whoever is holding it together.

Translation: CRMs got better, but maintaining them got harder.

You can see this in how companies staff it now. In a lot of the teams we speak to, it is no longer one person managing the CRM. It is a RevOps lead, plus someone in marketing ops who owns the sending side, plus a sales ops person who owns the pipeline, plus an external consultant on retainer for the parts nobody internally understands. Four people, partially, to keep one system honest.

That is the tension. CRMs have never been more capable, and running them has never cost more.

Hiring Claude as Our CRM Manager

At 9x we reached the point where a serious company hires someone to take care of our CRM. We looked at it properly, and after careful consideration, we decided not to.

Not because the work stopped mattering, and not because we found a way to care less about our data. Because the job is already being done, by Claude, and it has genuinely changed how the three of us work with our CRM.

We want to be specific here, because "AI helps with your CRM" is the kind of sentence that means nothing. So here is what it actually does, function by function. We run Attio, and everything below is a workflow we use.

Sales

• Call notes get pulled and written to our standard, every time. Claude reads the call from Attio through MCP and drafts the note against the structure we defined once: what they said about their situation, what they are trying to do, the objections raised, what we committed to. We review and correct. The gap between our best note and our worst note has gone to roughly zero.

• The analysis and the next steps come with it. Claude works through what the call actually means, proposes the next steps, and then executes them, from the follow-up email to the proposal, instead of handing us a summary and leaving the work exactly where it was.

• Deal fields get updated from evidence rather than memory. Stage, next step, close date. Where the activity supports a change it proposes one and says why. Where it does not, it flags the contradiction instead of guessing.

• The pipeline review arrives already prepared. Before the call we get the summary: what moved this week, what has gone quiet, which close dates no longer match any activity, which deals sit in a stage they no longer qualify for. We start at the discussion instead of spending the first fifteen minutes reconciling reality.

• Pre-call preparation happens before anyone thinks to ask for it. Claude pulls the CRM history on the account and enriches the people who will actually be in the room, so we walk in knowing who we are talking to rather than working it out in the first ten minutes.

Marketing

• Lists get pulled in plain language, and land where they are needed. Jan describes the segment and the exclusions in a sentence. Claude finds the matching records in the CRM, applies the rules, and pushes the finished segment straight into Customer.io, all without leaving the chat interface. That request used to sit in someone's queue for two days.

• Overlap checks happen as part of the pull. Who has been contacted in the last thirty days, who was on Tuesday's list, who is in an active sales conversation and should absolutely not receive a marketing email this week.

• Segments got narrower. This is the second-order effect, and it matters more than the time saved. When defining a precise segment was expensive, we sent broader than we should have and called it efficiency. Now the precise version costs the same as the lazy one, so we send the precise one.

• Post-campaign follow-through actually happens. Who engaged, who replied, who should be handed to sales today rather than eventually.

• Event registrants stop falling through the cracks. After each session, registrants and attendees get matched against the CRM, the new ones get created with the right source, and the ones worth a conversation get routed to sales instead of sitting in an export nobody opens.

Account Management

• Account histories get summarised before the conversation instead of reconstructed during it. Claude pulls context from every system and record we have, so we get a clear picture of every interaction that company has had with us since the very beginning, in order, in a page.

• Quiet accounts surface on their own. No customer should go dark for two months without anyone noticing, and previously that only got caught when someone happened to think of it.

• Renewals and check-ins stop depending on memory. What is coming up, what was promised last time, what has not been delivered yet.

• Commitments made on calls become tracked follow-ups. What we promised, by when, and whether it actually happened, taken from the notes rather than from somebody's recollection.

• Expansion signals surface from the conversation history. When a customer mentions a need we could serve, on a call or in an email thread, it gets flagged at the time instead of being noticed six months later by accident.

Data Management

• Deduplication runs weekly instead of quarterly. Claude works through the records, applies our matching rules, and proposes merges with its reasoning attached. We approve. The mess never gets the chance to compound.

• Records that break our own rules get flagged rather than discovered six months later by someone building a report.

• Enrichment gaps get filled where the data is findable, and marked as unknown where it is not, instead of sitting silently empty.

• New records get normalised on the way in. Company names, domains, job titles and sources follow our conventions regardless of which integration happened to create the record.

• One-off data questions get answered without building a report. How many of these do we have, which ones are missing that field, what changed since last month. Asked in plain language, answered against the live data.

None of this replaced the enrichment and automation infrastructure we already had running in the background. It complements it, covering the judgement-shaped parts those tools were never going to reach.

The honest summary: the tedious ninety percent is handled continuously, at a frequency no human was ever going to sustain. Not once a quarter when someone finally has a free afternoon. Every week, in the background.

What We Kept for Ourselves

The team still decides what a stage means, what qualifies as an opportunity, and which fields are mandatory. Those are business decisions. Claude applies them rigorously and does not get to set them.

Nothing is written to the CRM and nothing reaches a human being without one of us approving it first. That is not about distrust. It is the design choice that lets us run these workflows every week rather than cautiously and rarely.

And the cases that do not follow the rules stay with the people who understand the context. That is where the actual selling happens, and it is exactly where the reclaimed time goes.

Where This Leaves You

We are not claiming the CRM manager role should not exist. In a large enough organisation it clearly should.

But most of what made that job miserable was never really a job. It was a backlog of small, recurring, rule-based tasks that got bundled into a role because a human was the only thing available to do them. That stopped being true, and it changed our answer to a hiring question we were about to act on.

If you are the person holding your CRM together right now: write down every recurring task it generates, and mark the ones that follow rules you could state out loud. There’s a good chance a lot of it can be managed with AI nowadays.