You’ve turned on AI features in HubSpot. The results aren’t what you expected. Before assuming the tool isn’t working, look at the data underneath it. The prospecting agent, customer agent, Breeze Assistant — teams are eager to put them to work. But there’s a pattern we keep seeing: people turn on AI features, get underwhelming results, and assume the tool isn’t working. More often than not, the tool is fine. The data underneath it isn’t.
When AI surfaces the wrong contacts, generates incomplete summaries, or produces insights that feel off — the tool is usually working exactly as intended. It’s working with bad data. Before you put more time into AI features, invest in the foundation first.
Here’s what we covered in a recent Data Management Excellence Small Group Coaching session — specifically the areas that matter most for getting your data AI-ready.
What’s covered in this post:
Auditing your portal — where to start and what to look for
Cleaning up duplicates — why they’re an AI problem, not just a reporting one
Fixing inconsistent properties — the free-text fields quietly undermining your AI
Enriching your data — filling gaps with HubSpot’s built-in tools
Going deeper live — free Small Group Coaching sessions to work through it in your own portal
__________________________________________________________________
Start With an Honest Audit
Most people don’t know the Data Model tool exists, let alone that it has a built-in health audit. Go to Data Management > Data Model (admin access required), then click the Health tab.
Knowledge Base: Use the data model health check
It’s one of the fastest ways to see where your portal actually stands. It flags low fill-rate properties, duplicate property names, association gaps, and pipeline or lifecycle issues — all in one view. If you’ve inherited a portal that’s been through a migration or two, expect a long list.
Once you’ve looked at the Health tab, run this prompt in Breeze Assistant while you’re in the Data Model:
“Review my HubSpot portal and identify: any custom properties with low or zero fill rates that could be archived, duplicate or near-duplicate properties that could be consolidated, any active workflows that are unnamed or missing descriptions, and any lifecycle stage or pipeline gaps. Give me a prioritized list of what to fix first.”
It cross-references what it sees on screen and gives you a cleanup roadmap. Review the output before acting — treat it as a starting list, not a final diagnosis.
While you’re in the Data Model, the Intro tab is also worth visiting. This is where you can prompt Breeze Assistant to review your entire data model setup based on your actual business — not just flag issues, but recommend the right objects, properties, and associations from the ground up:
“I run a [B2B/B2C] [industry] business. We sell [products/services] to [customer type, e.g. enterprise hotels, SMB retailers, individual consumers]. We use HubSpot for [sales/marketing/service/all of the above] and our team is primarily focused on [e.g. outbound prospecting, inbound lead nurture, renewals]. Recommend the right objects, properties, and associations for our data model.”
This is worth running even if your portal has been live for a while. You can also rename objects here to match how your team actually talks — a B2B team might rename Contacts to Buyers, a B2C team might call them Members or Guests.
Before you start making changes: get a backup in place first. Go to Settings > Data Management > Backup & Restore and run a manual backup before any bulk imports, workflow changes, or integrations. Starter and Pro users can run manual backups anytime; Enterprise can schedule automated ones. Files are downloadable for 14 days.
Knowledge Base: Back up CRM data
Duplicates Are an AI Problem, Not Just a Reporting Problem
This is the big one. AI features in HubSpot pull from your records — and if you have the same contact or company stored three different ways, the AI sees three different entities. That means the AI works with fragmented, incomplete context — and the outreach it generates reflects that.
HubSpot deduplicates contacts on email address and companies on domain/URL.
To find them: go to Contacts or Companies, click ⋯ (three dots), then select Manage Duplicates.
Similarity Score gives each duplicate pair a 0–100 score based on how alike the records actually are. Scroll right in the Manage Duplicates view to see it. A score of 70–89 is typically worth reviewing; 90+ is usually a safe merge.
Data Hub Professional users can bulk review and bulk merge, which is useful when reviewing large volumes of records. If you don’t have Data Hub, two third-party tools worth knowing:
- Dedupely — deduplicates across contacts, companies, deals, and more
- Kickbox — validates and cleans email addresses before they become bad data in the first place
Inconsistent Properties Limit What AI Can Do
Free-text fields undermine AI performance. When reps type whatever they want into a single-line text field, you end up with hundreds of variations of the same value — and AI can’t reliably group, segment, or summarize across them.
Tip: Switch to dropdowns wherever you can.
Recommended Breeze Prompt: “Find my most-used single-line text properties and recommend which should become dropdowns.”
Quick video: How to create a custom property
Tip: Use conditional logic to surface the right fields at the right time.
A “Close Lost Reason” field that only appears when a deal moves to Closed Lost means reps actually fill it out — which means AI actually has something to work with when it’s analyzing lost deals.
Tip: Make the important stuff required.
If a field matters for AI context, reporting, or segmentation — require it. Reps fill out what they can’t skip.
Enrich What You Have, Fill In What You’re Missing
Even with clean data, there are gaps. AI features surface them fast. HubSpot has two tools to help.
Data Enrichment (available on Starter and above) fills in missing contact and company information using HubSpot’s commercial dataset. Check your field mapping settings before running — you control whether it fills empty values only or overwrites existing ones.
Knowledge Base: Enrich your contact and company data
Smart properties via Data Agent go further, covering questions standard enrichment can’t answer. You prompt it in plain language and it builds the property for you. Worth exploring if you have specific data gaps that standard enrichment doesn’t cover.
Data Agent uses HubSpot credits per record, so most teams run it on target accounts rather than the whole database. Check your account settings for current credit costs.
Already have messy data from a migration? If you have Data Hub Professional or Enterprise, the Format Data workflow action can standardize values at scale — capitalize names consistently, trim spaces, replace strings.
Format your data with workflows (Requires Data Hub Professional or Enterprise)
New admin taking over a messy portal?
If you’re on Enterprise and inherited a portal with years of unused forms, lists, workflows, and properties clogging things up, the Account Cleanup tool is worth knowing about. It surfaces unused assets across your account so you can remove them in bulk rather than hunting them down one by one.
Knowledge Base: Remove unused assets from your account with account cleanup (Requires Enterprise)
Want to Work Through This Live or Explore Additional Tips/Tricks?
To go deeper on any of this with a HubSpot professor and a small group of peers, join Small Group Coaching. These are free, live sessions — usually 60 minutes, capped at a small number of participants — so there’s actually time to get into your specific situation.
The sessions most relevant to getting your data AI-ready:
Data Management Excellence — the full deep-dive on everything covered in this post
Mastering Analytics: Your Guide to Better Reporting — cleaner data leads to better reports; this session shows you how to build them
Browse all Small Group Coaching sessions
Working through any of this in your own portal right now? Drop a comment — always good to hear what’s actually working (or not) for different teams.