HubSpot Breeze and AI implementation.
Breeze, pointed at a number it moves. AI features in a CRM are only as good as the data underneath them. We scope each use case to a measurable outcome, fix the data it depends on, and put an evaluation in place so you can tell whether it worked.
Switched on. Nobody measured anything.
The AI features were enabled because they were included. Some people use them, some do not, nobody agreed what success looked like, and the honest answer to whether it helped is that nobody can say.
Enabled. Never evaluated.
AI — the unmeasured rolloutAI with a number attached.
Pick a use case with a metric
Time to first response, summarisation accuracy, meeting preparation time. If nobody can name the number it moves, it is not a use case, it is an experiment.
Fix the data first
AI reading a portal with duplicate records and undefined lifecycle stages produces confident, wrong output. Data readiness is most of the work on most AI projects.
Agents scoped narrowly
Prospecting and support agents given a defined job, a defined dataset and a defined escalation to a human. Broad autonomy produces work somebody has to check anyway.
Guardrails and review
What the AI may send unreviewed, what a human approves, and what it must never touch. Written down before the rollout, not after an incident.
Evaluation you can run again
A test set and a scoring method, so a model or prompt change can be compared rather than argued about on impressions.
Where HubSpot AI is not the answer
Sometimes the job needs a purpose-built system against your own data. We build those too, so we have no reason to pretend otherwise.
Audit, architect, build, hand over.
Audit
The portal read end to end — objects, properties, automation, permissions, integrations and the reports leadership actually opens.
Architect
The model written down before it is built: objects, lifecycle, ownership and the definitions every report has to agree on.
Build
Configuration, automation and integration built to that model, in a sequence that leaves the team working throughout.
Hand over
Documentation, enablement for the people who run it daily, and a period operating alongside your team until it is genuinely theirs.
AI on top of an unfixed model.
It is the fastest way to produce plausible nonsense at scale. A summariser reading a record with three conflicting lifecycle properties writes a confident summary of the wrong thing. Fix the model, then add the AI, and it works the first time instead of the third.
What people ask before they commit.
Is Breeze worth using?
For summarisation, drafting and research inside the CRM, genuinely yes. As a substitute for a data model nobody defined, no, and that is what it is most often asked to be.
What data does it use?
Your portal data, within HubSpot's own boundaries. For regulated industries the right question is what should be in the portal at all, and that is a design decision we take early.
Can you build custom AI against our HubSpot data?
Yes. AI and data is one of our five capabilities. Retrieval over your own content, scoring models and agents built against your data, with evaluation rather than vibes.
How do we measure whether it worked?
A baseline before rollout and the same measurement after. It sounds obvious and it is skipped almost every time, which is why so few AI rollouts can be defended.
How long does it take?
Two to four weeks for a scoped Breeze rollout with evaluation. Custom AI work is a different conversation and a longer one.
Where people go from here.
Point it at a number.
Tell us what you are running
What the system does today, where it breaks, and when it has to work. An engineer reads it — you get an answer inside one business day, not a sequence.