Every HubSpot we inherit looks fine on the surface. There are contacts. There are deals. The pipeline has numbers in it. But the moment you start asking specific questions — who owns these records? when was the last note logged? which of these lifecycle stages actually means something? — the cracks appear fast.

The traditional approach is to spend several days clicking through views, exporting CSVs, and building a findings deck in PowerPoint. The client gets a document a week later that diagnoses the problem but doesn’t quite get to the fix. We wanted to do this differently.

Over the past few months we’ve been developing an audit process that uses Claude — Anthropic’s AI — connected directly to a client’s HubSpot via the MCP integration. Here’s how it works, what it surfaces, and why it produces better output faster than manual methods.

Before explaining the method, it’s worth naming what makes CRM audits hard in the first place. The data issues are rarely mysterious — missing fields, orphan records, stale close dates — but they exist in volume. A database of 4,000 contacts with structural problems across multiple dimensions requires either a lot of manual filtering or a lot of export-and-pivot work to quantify properly.

More importantly, a list of problems is not an action plan. A senior leader looking at “886 companies have no associated contacts” doesn’t know whether that’s a critical issue or a minor annoyance, who should fix it, or what the right sequence of fixes is. The gap between audit and roadmap is where most engagements stall.

The process has three phases that happen in rapid sequence, usually within a single working session.

To give this some texture, here’s a snapshot of what we found on a recent engagement with a life sciences client running a two-team HubSpot for their commercial and partnerships functions.

These numbers matter because they’re not just untidy — they actively break things. When 93% of contacts show as “unworked,” the CRM can’t tell you who’s been engaged and who hasn’t. When nearly a fifth of your contact database sits in stages with no defined meaning, your funnel reports are unreliable. When a quarter of your companies have no associated contacts, your account-level views are fiction.

The audit also surfaced a more interesting finding: the problem wasn’t that the team didn’t know how to use HubSpot. One rep was logging exemplary notes — detailed, structured, capturing context, decisions, and next steps in a form anyone could scan in thirty seconds. The rest of the team simply had no shared standard to follow. The process existed. It just hadn’t been written down or socialized.

The output we give clients isn’t a list of problems — it’s a phased plan with clear ownership and timing. For this client, that meant three phases.

Phase one was structural cleanup: archive orphan companies, assign owners to unowned records, investigate and relabel a mystery lifecycle stage that had been created during an import and never properly mapped. These are one-time fixes that a HubSpot admin can work through in a week. They don’t require a change in how anyone uses the CRM — they just remove the noise that’s making everything else harder to read.

Phase two was behavioral: introducing a note-logging standard, backfilling current-state notes on all open deals, starting to log calls (the team had logged zero calls in the system, ever), and enforcing close-date discipline. This required an all-hands session and a change in what the weekly pipeline review demanded of each rep.

Phase three was maintenance: monthly hygiene checks, a KPI dashboard visible to the whole team, and a quarterly review of the standard itself. This is the part most consultants skip — but without it, the CRM drifts back to its previous state within six months.

It would be easy to overstate this. Claude is not doing anything a skilled RevOps consultant couldn’t do — the analytical logic is the same. What it changes is the economics of doing it thoroughly.

Running 15+ structured queries against a live HubSpot, synthesizing the results against a client’s internal documentation, and producing a structured Word document with metric tables and a prioritized roadmap would previously have taken two to three days of consulting time. We now do it in a single session. That changes what’s viable to offer — and to whom.

It also changes the quality of the output. Because Claude can hold the full audit data in context while drafting the roadmap, the recommendations are grounded in the actual numbers rather than being written from notes after the fact. When the document says “the 37 deals with no associated contacts should be linked or closed-lost,” that’s not a general recommendation — it’s a specific action tied to a specific count from the live database.

This process works well for structural and completeness issues. It’s less good at things that require subjective judgment — whether a particular deal should be closed-lost, whether a lifecycle stage definition makes sense for a specific business model, whether the team culture will actually support a new logging standard. Those conversations still happen with humans, in a room.

It also requires that the client’s HubSpot is reasonably well-organised at the top level. If the pipeline structure itself is wrong — wrong stages, wrong deal types, pipelines being used for the wrong purposes — the audit can identify symptoms but the fix requires a deeper conversation about commercial process design.

The honest answer is: if you haven’t run a structured audit in the last year, it probably does. The signs are usually the same: a pipeline that looks healthy but doesn’t forecast accurately, reps who know more about their accounts than the CRM does, and a nagging sense that you’re making decisions based on data you don’t quite trust.

If any of that sounds familiar, we’re happy to run a no-obligation discovery session. We can usually give you a clear picture of where things stand — and what the highest-leverage fixes are — in a single conversation.