Your HubSpot AI journeyStage 4 of 4

AI runs across your business. Someone needs to own it.

You have agents in marketing, sales and service, probably more than one AI tool, and a business that keeps changing underneath them. The work now is keeping it accurate, joined up and worth what it costs.

HubSpot as the source of truthCustom agents for your processSpend watched monthly
Mark, Emma and Rikki working around the boardroom table
Quick self-check

This sounds like you if…

Tick what's true today. Nothing is sent anywhere. It's just a quick way to check you're reading the right page.

0 of 6 ticked
What's really going on

The risk at this stage is quiet drift.

Nothing breaks on a single day. Context goes stale when an offer changes. An agent keeps quoting the old process. Credits creep up because nobody turned off the experiment from March. Two AI tools give two different answers because they read from different places. Trust drains away and people go back to doing it by hand.

The setup that holds up is simple to describe. HubSpot holds the customer record and the company context, securely. Claude does the deeper analysis, writing and orchestration across your other systems. HubSpot's agents do the last mile inside the CRM. A person confirms anything that matters.

The gate that holds it all together is the human who confirms.

HubSpot at this stage

The tools for running AI at scale.

Checked against HubSpot's knowledge base, November 2026.

Agent builder
What it doesCustom agents with your own instructions, actions, knowledge and inputs. They can read and write CRM records, browse the web, and call other systems through MCP connections.
PlansStarter and above on most hubs, Smart CRM Professional and Enterprise. HubSpot Credits required.
CreditsCharged as the agent thinks and acts, so cost varies with the task. Testing in the builder is free, and you can set monthly run limits.
Our tip Write the instructions like a job description: role, goal, approach and what good output looks like.
Agent Hub
What it doesOne home for every agent: the builder, the agent inbox for reviewing their work, the marketplace, and the Context tab.
PlansStarter and above, with HubSpot Credits.
CreditsEach agent uses credits as above.
Our tip Use the agent inbox as the weekly review, not a dashboard nobody opens.
Agents in workflows and records
What it doesPublished agents run inside workflows on Professional and Enterprise, from side panel automations on Starter, or from a card on any record.
PlansWorkflow builder Professional and Enterprise. Automations on Starter.
CreditsAgent credits plus the workflow action.
Our tip Put the human check in the workflow, not in somebody's memory.
Connectors and MCP
What it doesHubSpot connects to Claude and other AI tools, and Breeze Assistant connects to apps like Gmail and Slack, so every tool works from the same customer record.
PlansClaude connector on all plans. Connected apps need setting up first.
CreditsNo HubSpot credits for the Claude connector.
Our tip Decide which system is the source of truth for each fact, and write it down.
HubSpot Credits

Keeping spend tied to value.

At this stage credits are a budget line, so they get managed like one. HubSpot lets you set an account limit, a limit per feature and a limit per agent action, and shows usage by feature each month.

Our monthly check-in puts usage by feature next to the outcome each agent was built for. Anything not earning its keep gets changed or switched off.

Account limit
Defaults to 50% above your monthly credits once you buy extra. Features pause when it is reached.
Feature limits
A monthly cap per tool, such as the Prospecting agent
Action limits
A monthly cap on Customer agent and Buyer Intent actions
Alerts
Emails to billing contacts at 75%, 85% and 90%, and when the limit is passed
From real portals

What leading looks like in practice.

From our own client work. Names left out on purpose.

Leadership team

Weeks of analysis in an afternoon.

A leadership team wanted a view of the business that would normally take weeks of analyst time and cost £5,000 to £6,000. With HubSpot as the context and Claude doing the analysis, it became a review job instead of a project. It was only as good as the data underneath, which is why the data work came first.

The lesson The value is real when the data is right.
Our own business

£100k+ of pipeline found on the first sweep.

We run our own sales on HubSpot and Claude. The first CRM integrity sweep found 11 opportunities nobody had logged, worth more than £100k. Every closed deal now carries a win or loss reason the system learns from.

The lesson Build it for yourself first. We did.
Founder-led firm

Seven logins replaced by one daily brief.

Seven systems and one founder's head became one connected operating system on Claude and HubSpot. 14 scheduled automations, each one fails loudly, and 15 to 20 hours a month back.

The lesson Fail loud beats fail quiet. Every automation tells someone when it breaks.
How we help

How we run it with you, month to month.

Ongoing, one monthly fee, 30 days' notice.

Monthly
Usage and value

Credits by feature next to the outcome each agent was built for. Anything not earning its keep is changed or switched off.

Monthly
Context review

Offers, prices, processes and people checked against what the AI reads. Stale sources updated or retired.

Quarterly
Outcomes review

What moved, what didn't, and which agents or builds come next, in priority order.

As needed
Build and tune

Custom agents, Claude builds and integrations scoped and delivered as priorities change.

What you leave with

Things you can point at, not a slide deck.

Everything is documented and owned by you. It keeps working whether or not we're still involved.

One source of truth for customer data and company context, with an owner for each part.
Custom agents built around your own processes, each with a human check.
HubSpot and Claude connected so both work from the same record.
Monthly credit reporting tied to outcomes, with limits set.
A quarterly roadmap of what to build, change or retire next.
Proof
15 to 20
hours of founder time back every month

One connected AI operating system on Claude and HubSpot, with a daily brief, a self-checking dashboard and 14 scheduled automations. In its first fortnight it caught a won retainer that was not yet invoicing.

Read the case study →

Emma, Rikki and Mark around a laptop in a booth
Where to start

The services that fit stage 4.

Published prices. Fixed before we start.

Is it working?

Working well when
Every agent has an owner, a number and a review date.
Your AI tools agree with each other because they read from one place.
Credit spend is reviewed next to results every month.
Needs attention if
Agents were switched on in a burst and nobody has looked since.
Different teams run different AI tools with different context.
You find out about spend from the invoice.
Common questions

Stage 4, answered straight.

Why use Claude if HubSpot has its own agents?

They're good at different things. HubSpot's agents are best inside the CRM: chat, prospecting, enrichment. Claude is better at deep analysis, long writing and working across systems HubSpot doesn't hold, like finance or project tools. HubSpot stays the source of truth for both.

Can HubSpot agents use our other systems?

Yes. Agents built in the agent builder can call external systems through MCP connections, and Breeze Assistant can use connected apps. We decide which system owns each fact first, so nothing contradicts.

How do we stop credits creeping up?

Limits at account, feature and action level, a named owner per agent, and a monthly look at usage next to results. Most creep comes from experiments nobody switched off.

Who should own AI internally?

One person accountable, usually in RevOps or operations, with an owner per agent. We can be that function alongside your team through the Growth Partnership until you're ready to run it yourselves.

Is it safe to let agents write to the CRM?

Yes, with guardrails. Scoped permissions, approval on anything that changes records in bulk or reaches a customer, and an audit trail. We've seen what an unchecked filter can do, so we never skip the test run.

Keep it accurate, joined up and worth it.

Thirty minutes. Bring your list of agents and tools, and we'll tell you where the drift is likely to be.

Book a 30 minute callNo sales team, no gatekeeper.
Rikki Lear
Rikki LearCommercial Director