Tools, not a plan, and a CRM nobody could target off.
HubSpot was in place, there was an appetite for AI, and no safe plan to connect the two. Claude had been tried a year earlier and shelved, because it couldn't do the job then. Marketing was one person, the lead gen seat was sitting empty, and reporting was rebuilt by hand every week out of the same spreadsheets.
Around 3% of the 4,024 companies in the CRM had ever reached a deal, and the other 3,891 looked identical on the fields you'd filter by. Targeting was guesswork wearing a filter. When anyone asked a generic AI tool about pipeline, the answer came back confidently wrong, because it had never heard of IXT's deal stages.
That isn't an AI problem. It's that nothing they'd bought knew how the company works.
Prove it by hand. Automate it last.
Every skill built, self-critiqued, tested on the live portal and signed off before the next one started.
Official connector, scopes reviewed object by object, smoke tests on the real portal, and a HubSpot audit across 13 areas that scored 75 out of 100, with every remediation task owned by a named person at IXT.
Five skills: tone of voice, ICP, internal definitions, product detail and CRM structure. Each one versioned in a changelog the client signs. Live testing caught a factual error in one, which is exactly what that step is for.
A 45 prompt playbook, a 45 minute live session and the recordings, plus a written go or no-go with a recommendation on what to build next.
Each automation proved by hand with a human check at every step, then scheduled, then handed over. Department by department, in the order that keeps people believing it.
What changed in five months.
running inside marketing, built and proved before the hire even started.
live: prospecting, CRM hygiene, campaign build, content and reporting. None existed in April.
deal review, call prep and a board summary in minutes, instead of rebuilt by hand each week.
marketing live, the CEO and CFO using it, sales and development onboarding through the rollout.
“Now that it's so connected to our HubSpot, it knows exactly what to do. It's all factually correct. It's kind of spot on.”
Thirty minutes of Monday prep, in seconds.
One prompt, typed in plain English. The ICP, Definitions and CRM Structure skills load above the answer without anyone asking for them, because the task matched. That's why the answer comes back in IXT's language, about IXT's pipeline, citing IXT's data.
Then she started building things we never sold her.
In August the CMO took on lead gen herself. A few weeks later all of this was running, and none of it was on our scope.
Claude found around 500 genuinely unique companies in it, scored every one against the ICP and ranked them. Now it hands over 10 a day, creates the records in HubSpot, sets the owner and drafts the email. A list that had been sitting there dead was booking meetings inside three days.
Hundreds of contacts qualified by hand at some point, then left to go stale. There's now a daily batch: scored against the ICP, given an owner, a link to the contact card and an outreach email waiting. Live opportunities recovered from a part of the database everyone had written off.
Scoring kept returning almost nothing new, so she asked it why: 800 companies had no industry recorded. She got a spot-check sample, then a spreadsheet of over 300 confident fixes, and approved the lot. Fixed in an afternoon.
She wrote the campaign down roughly and asked for a plan. It redesigned the invite, checked the copy and landing page, then built the emails, workflows and properties one step at a time, with her approving each one. It picked the right properties because it can read the portal.
Blog posts written against the gaps their AI search tool flags, on a schedule, and a slide deck built in their brand that she used as it came out.
“They're already saving me so much time. And I think almost the challenge is to keep up.”
“I tested Claude a year ago and it didn't feel like all of this was fully possible in any way.”
“I was so limited in thinking what I can give to AI to do.”
Same tool. Same portal. Why did it work this time?
The first attempt was a person trying a tool. The second was a system built around one. Nothing about Claude changed in that year.
Two of the actual deliverables, in their brand.
Not mockups. This is what landed in IXT's hands, and what every department onboards onto.
A phased programme, not a pilot that goes nowhere.
Phase 1 was a fixed-price proof of concept with a written go or no-go. Phase 2, scheduled reporting, was priced per report. Phase 3 was left unpriced until Phase 1 had earned it. They bought Phase 2, and the prospecting agent on the Phase 3 list became the lead gen system that went live in August.
What would this do on your HubSpot?
Tell us what your team still does by hand, and we'll tell you what an AI layer on your CRM would actually change, including if the answer is not much. You'll speak to Rikki, not a salesperson.
