Blog

What we heard at HubSpot GROW Europe 2026

Written by Marie Roberts | Nov 19, 2026, 12:21:29 PM

Integrations, setup and AI: three themes from a day of conversations on the Centralise stand, and the one thing they have in common.

This year some of the Centralise team spent the day on our stand at HubSpot GROW Europe. Between sessions we spoke to dozens of HubSpot customers: finance leads, RevOps owners, marketing teams and founders.

Some had been on HubSpot for years and others were still choosing an onboarding partner.

The conversations varied, but the same three topics kept coming up.

1. HubSpot rarely works on its own

Integrations came up more than anything else. NetSuite was mentioned over and over again. Others told us about finance systems, a custom ticketing tool that sits outside HubSpot, Power BI reporting, and in one case three quoting systems that don't talk to each other. Another business had built its own quotation tool in-house and now needed it to share data with the CRM.

The pain was the same each time. When the data about a customer lives in four places, nobody has the full picture. Sales can't see what finance knows, and the handover from sales to service relies on someone remembering to pass things on.

One visitor had already been quoted for a custom NetSuite build and wasn't convinced and that raises a fair question: should you build custom, use a native connector, or buy something off the shelf?

Our view is that the method comes last. Before choosing how to connect two systems, agree on three things:

  • What needs to move, and in which direction.
  • Which system is the source of truth for each piece of data.
  • Who owns it when the two disagree.

A custom build on top of an undefined process fixes that confusion in place and makes it more expensive to change later. Once those three answers are clear, choosing how to connect the systems is usually the easy part.

2. New to HubSpot or years in, the question is the same

The second group of conversations came from both ends of the HubSpot journey.

At one end were teams moving Sales and Service off spreadsheets and comparing partners to onboard with. At the other were teams two or more years in who know they're only using part of what they pay for. In between were a RevOps lead running several legal entities in HubSpot with no implementation partner, and a regulated business running several portals and struggling with duplicates, inconsistent rules and processes that work differently in each one.

They were all really asking one question: is HubSpot set up to give us what we need?

The structural decisions you make early on are the ones that are hardest to change later. These include how many portals you run, how entities and business units are modelled, how deal stages are defined and who owns which properties. Get them right and the portal grows with you. Get them wrong and every new team, workflow or report adds another workaround.

For teams just starting out, that means making those decisions deliberately in the first few months instead of copying the spreadsheet into the CRM. For teams already a few years in, it means reviewing what's there before adding anything new.

3. AI is only as good as the data underneath it

AI and Smart CRM were the most-ticked interests on the badges we scanned, and AI came up in most conversations in one form or another.

One business that has scaled quickly and now runs almost everything in HubSpot wanted to know where AI fits, starting with customer service, where their current chatbot isn't pulling its weight. Another visitor wanted to bring their data together in one place first, as a step towards a "headless CRM". And as for the regulated business mentioned above, the question was what to hold back. They have to restrict how AI and Breeze are used across their portals, so for them governance matters as much as capability.

What these conversations had in common was the order of work: get the data into one place, then automate. An agent treats your lifecycle stages, deal stages and record owners as fact. If those are inconsistent, the agent acts on the inconsistency faster than any person could, and nobody notices until a customer does.

Before switching anything on, we'd suggest teams decide:

  • What the agent is allowed to treat as fact, and agree those definitions across marketing, sales and service.
  • Which processes already work, because automating a broken handover just makes it break faster.
  • Who owns the output, so every agent has a named person accountable for what it does to a record.

From the sessions

Three sessions on the agenda covered the same ground as the conversations on the stand.

Anthony Rowell, Tradeprint: proving what AI actually does to revenue

Tradeprint is a Dundee print manufacturer that runs sales and customer success on HubSpot. Since January it has run HubSpot's Breeze Prospecting Agent across 16 plays. Its first read of the revenue linked to the agent was roughly 30 times too high.

Anthony explained how that happened. Revenue on any contact the agent had touched was counted, whatever had actually driven it. Existing customers who would have bought anyway were included. Orders placed before enrolment were credited, repeat orders were added on top, and there was no control group to compare against.

The fix used data HubSpot already held, the date each contact was enrolled in a play, plus a set of written rules. Only genuinely new customers count, only their first order, and only an order placed after the agent made contact. On that basis the agent has driven more than £250,000 in first-order revenue from over 2,000 new-customer orders. His line, "A smaller number the board believes beats a big one nobody trusts." made us cheer internally, because it's something we've been saying too. 

His five-step test works on any AI revenue figure: define the win before launch, baseline from zero, check the sequence of events, hold out a control group, and write the rules down so every figure is labelled actual, estimate or directional.

Adam Cuzzort, HubSpot: agents need context to be useful

Adam's starting point was that getting one agent running is easy and managing many is hard. Customers experience the handoffs between tasks more than the tasks themselves. An agent that lacks customer context leaves a person to gather that context, check its output and update the CRM afterwards, which adds steps instead of removing them.

Agents need three things: customer context, a defined business process and somewhere to run. HubSpot's answer is Agent Hub to manage agents, Agent Builder to create them around how a business works, and Breeze Assistant to bring them into the team's day. He reframed the question too. "Where can we add agents?" matters less than which work agents should own, where human judgement matters most, and how to make every handoff better.

Séan Reid, Supermetrics: from health scores to renewal probability

Séan's argument was that a health score tells you how a customer feels today, not whether they'll renew. A yellow score is vague, leans heavily on usage data and gives sales or finance little to act on.

Supermetrics replaced it with a Renewal Probability Score. It's a single percentage built from product usage, billing, support, call sentiment and CSM notes. Because it's a number, finance can model revenue risk with it and leadership can see a renewal forecast next to the pipeline. His advice for getting started: map every system that holds customer health data, define what actually makes your customers renew, and make CSM note-taking part of the workflow, because the model is only as good as what gets logged.

The three speakers covered different topics, but each point depended on data that was already in good shape. Tradeprint could correct its figure because the enrolment dates were already in HubSpot. Adam's agents only help when the customer context is there to draw on. Séan's score only works once signals from several systems land in one place.

Announced on the day: HubSpot goes headless

On the day of GROW, HubSpot published HubSpot is headless. Here's how we'll price it. It sets out how AI tools outside HubSpot can work with the CRM. Everyday use through HubSpot's connectors for Claude, ChatGPT, Gemini and Copilot, its MCP server and its Agent CLI is included in the seats customers already pay for. Heavier users can top up with HubSpot Credits. Third-party agents running on HubSpot's APIs count against the plan's API allowance and don't use credits.

In the same week, HubSpot gave its Claude connector a full-screen view you can pin to Claude's sidebar. It shows the contacts waiting on a reply next to your open tasks, and Claude drafts the follow-up from your HubSpot context.

That fits with the stand conversation about moving towards a "headless CRM". For many teams, a growing share of HubSpot work will now happen in Claude or ChatGPT instead of in HubSpot itself. The lesson from the stand and the sessions still holds, though: Claude can only work with what's in the CRM. If the deal stages, ownership and records underneath are inconsistent, a faster interface just surfaces the inconsistency faster.

Centralise is a Claude partner. If your team is starting to use Claude with HubSpot, our Claude × HubSpot workflow recipe pack is a practical place to start.

The common thread: the foundations

Integrations, setup and AI sound like three separate projects. On the stand and in the sessions, they kept coming back to the same thing: the data model, process and ownership underneath HubSpot. When those are solid, connecting a new system, onboarding a new team or deploying an agent goes much more smoothly. When they aren't, each of those projects takes longer, costs more and delivers less than it should.

That's why every Centralise engagement starts with Assess, a structured look at how your revenue system is actually working before anything gets built or changed. It came up unprompted in two of the strongest conversations of the day, from teams who had reached the same conclusion on their own: look at the whole system first.

See how Assess works →

Thank you to everyone who stopped by the stand, and to HubSpot for a brilliant day. If we didn't get the chance to speak, we'd still like to hear what you took away from GROW.