Most teams feed their AI agents signals that even sales reps can't act on. Janet and Clara share how they built the signal foundation before turning on any AI agent: what they tested, what they cut, and the 6 to 8x lift in meetings.

San Francisco, CA - UserGems, the AI command center for outbound and ABM, released the full on-stage session from Humans of GTM 2026, "Building the Foundation for AI Agents: Why HubSpot Started With Contact-Level Signals," featuring Clara Borges, Demand Generation at HubSpot, and Janet Wood, GTM Strategy at HubSpot.

Watch the full session.

"AI is only as good, only as smart, as the signals feeding it. We got very mixed results early on. Automation misfired, we still had generic messaging going out, and the reason for that is we didn't have the right setup of our data. Bad data in, bad data out." - Clara Borges, Demand Generation, HubSpot

The third consumer of your data

Borges opens the session with a shift she says most teams haven't caught up to yet. HubSpot built its signal engine for years with two consumers in mind: marketing and sales, both human. The job was to get data in front of a person and hope they figured out what to do with it. Now there's a third consumer: AI. And AI needs feeding just as much as a rep or a marketer does. It needs signals, it needs context, and it needs prioritization. If the data layer underneath it is messy or incomplete, the AI sitting on top of it isn't going to work right.

From chasing volume to chasing quality

For years, Borges says, the internal conversation at HubSpot was how to get more signals and which new vendors to add. That conversation has changed. Most intent data has become a commodity: competitors buy the same vendors, track the same keyword topics, see the same timing. The signal stops being useful the moment everyone has access to it. So the team narrowed in on the signals that actually moved deals, rather than adding more volume on top of a pile that was already too noisy to act on.

What they tested, and what they cut

Wood walks through the multi-year build behind that foundation:

• 2023, company-level intent. With inbound demand down, HubSpot brought in 6sense, Demandbase, and G2 page views. It worked better than cold outbound and surfaced about 90 new deals a month, but came with three problems: the volume needed a scoring model to make it usable, there were no contacts attached to the intent data so reps couldn't be automated against it, and the workflow reps received was, in Wood's words, an ambiguous number of people at an ambiguous company researching ambiguous topics an ambiguous number of times.
• Q4 2024, job change signals. Reps had been asking for this directly, and some were already tracking job changes manually on LinkedIn for their best accounts. HubSpot brought on UserGems, scaled its best rep's own sequences globally instead of writing new ones from scratch, and paired it with an automation rule: a signal fires, reps and BDRs get seven days to work it, then automation takes over so no high-value lead goes untouched.
• New-hire signals, two plays. HubSpot tested new hires without a prior relationship to the company across high-fit personas already in the database, and separately, cross-sell hires at existing customers, for example a new VP of Customer Success at a Marketing Hub account being a fit for the Service Hub product line. Both plays together produced 2x the deals at half the signal volume of the earlier company-level approach.
• Early 2026, contact-level intent. As soon as UserGems released contact-level web research intent, HubSpot tested it directly against the company-level intent it had been running since 2023. The result: 6x more deals than the company-level approach alone, and a 6 to 8x lift in meetings, without needing a scoring model to make the volume usable.

Feeding the AEO engine in a month

When HubSpot's product team moved up the launch of its AEO product from fall to spring, Borges and Wood had about a month to build demand audiences for it. Because the signal engine already existed, they moved fast: adding AEO, GEO, and AI search topics to what they were already tracking, and using LLMs to generate bespoke signals, like scraping a target account's website to check whether it had a blog or CTA buttons, as a proxy for inbound investment and product fit. They also used AI for the first time to synthesize signal data itself, feeding first-party, third-party, and LLM-generated signals into a shared table that summarized what was happening at an account and recommended outreach, since reps could no longer parse 50-plus signal properties on their own.

The takeaway: foundation first

"Data foundation is not the most glamorous topic, but it is what gets things done. It's the foundation that everything is built on. It needs to be accurate enough that reps can use it and trust it, and complete enough that AI can use it with the full context." - Clara Borges, Demand Generation, HubSpot

Borges and Wood close on two points: the data foundation has to come before any AI agent gets turned on, not after, and contact-level signals were the unlock that made feeding a unified data layer to both reps and AI actually work.

About Humans of GTM

Humans of GTM is UserGems' curated leadership event for B2B GTM leaders. The inaugural event brought 150+ vetted revenue leaders to San Francisco for a day of product announcements, customer playbooks, and roundtables, including sessions from Docebo, Mimecast, HubSpot, CaptivateIQ, and Frontify. Full session recordings are being released on an ongoing basis. Watch the sessions or subscribe to our newsletter for updates on future events.

About UserGems

UserGems is the AI Command Center for B2B revenue teams, the brain that sits between your CRM and execution tools. We unify signals, prioritize buyers using transparent AI scoring, and deploy AI agents (Gem-E) to automate outbound and ABM execution. Customers include Mimecast, Workday, Salesloft, Sendoso. Learn more at www.usergems.com.

‍

‍