HubSpot's Context Problem Is Your Data Problem: Reading UNBOUND 2026

HubSpot renamed its conference, and the word it chose to build the week around was "context." Every agent announced at UNBOUND in Boston (September 16–18, 2026) was pitched as running off a shared picture of your business: who you sell to, what you sell, how your team works, what's in the CRM. The product that holds that picture is Context Home, which HubSpot says is included in every edition and, on paid tiers, lives inside Agent Hub.

Strip the branding and the message is one we've been delivering to clients for years, usually to a less receptive audience: the agents are only as good as the records. HubSpot is now saying it on the main stage. That's useful. It also means the CRM data problem you've been deferring just got a deadline.

What Context Home holds, and what it doesn't

HubSpot's own page describes Context Home as "where you manage what HubSpot knows about your business." Admins put in portal-level material: brand voice and guidelines, a company profile, products and services, team methodologies, process documentation. Each user can add their own profile, email personality and custom context.

That's the hand-written half. The other half is the CRM itself. The same page frames "high-quality context" as populated CRM fields, logged activity, connected integrations and team settings. No document you upload fixes a pipeline where half the deals have no amount and a third of the contacts have no company.

So before the Content Agent drafts in your voice or the Prospecting Agent scores a company on fit, the question is whether the records those agents read are trustworthy. HubSpot's claims about outcomes for customers with "high-quality context" (3.2x more deals, 2.7x more email replies) are HubSpot's numbers, from HubSpot's data, with no methodology published. We'd treat them as a direction, not a forecast.

The credits angle

Most of the new agents consume HubSpot Credits. HubSpot's credits documentation, updated September 16, 2026, lists Customer Agent, Prospecting Agent, Data Agent and Data Studio syncs as consumers. The rate sheet in HubSpot's Product & Services Catalog puts a Prospecting Agent recommendation at 100 credits per lead and a Data Agent prompt at 10 credits per record. Overage is $0.01 per credit.

Here's why that matters for data quality specifically. An agent that scores 2,000 companies you'd never sell to has spent 200,000 credits (about $2,000 at overage rates) producing confident output about junk. Duplicates get scored twice. Companies with no domain get scored on nothing. The cleanup you skipped doesn't just degrade the agent's answers; you pay per answer.

Standard data enrichment, by contrast, doesn't consume credits on Starter, Professional or Enterprise, according to HubSpot's enrichment documentation. If you were going to fix company fields anyway, the included enrichment is the cheap part.

What we'd do before turning the agents on

This is roughly the order we run when a client asks us to get a HubSpot portal ready for AI features. None of it is new; it's the same work with a better reason attached.

1. Pull the coverage numbers

Export contacts and companies with the fields the agents will read: company domain, industry, employee count, lifecycle stage, deal amount, close date, owner. Count blanks per field. If "industry" is empty on 60% of companies, the Prospecting Agent's fit score is a guess on 60% of your database, and Context Home can't fix that.

2. Dedupe companies first, then contacts

Merging in the other order means re-associating the same contacts twice. We wrote up the merge rules we use, including which fields the older record should win, in HubSpot deduplication without breaking attribution. Snapshot everything before the first merge.

3. Decide what the agents may write

HubSpot says Deal Progression now updates deal records from emails and call transcripts, not only from post-meeting notes. That's helpful when the deal is clean and alarming when it isn't. Before enabling any write-back, agree on which properties an agent can set, which it can only suggest, and who reviews the suggestions. Write it down. The agent won't.

4. Set the credit caps

HubSpot's spend controls let you set an account-level limit and per-feature limits. Set the per-feature limits low for the first month, watch the usage breakdown under Account & Billing → Usage & Limits, and raise them once the output is worth it.

5. Then fill Context Home

Brand voice, ICP, products, process docs. Do this last. It's the easiest step and the one people start with, which is backwards: a perfectly described ICP pointed at a database with no industry field still can't find your customers.

Where this leaves a Pro or Enterprise portal

Our view after reading the announcements: nothing at UNBOUND changed the economics of CRM cleanup, but it changed who's asking for it. For years the person pushing for clean data was ops. Now it's whoever bought the agents and wants them to work.

If your coverage export looks rough, the HubSpot data cleanup page describes what a cleanup engagement includes, and the 300,000-contact case study has the shape of what a full run looks like. For a smaller portal the whole thing is often a few days of work, which is less than most teams spend arguing about whether the agents are any good.