Prove AI on one real workflow before you commit to more
An AI pilot puts AI to work on one job, for one team, for a set number of weeks. We agree on the measure up front, and a person on your team checks the AI's work. At the end, the numbers tell you whether to keep it, widen it or switch it off.
Free, 30 minutes, and you'll leave with at least one clear next step.
What goes into an AI pilot?
Every pilot has the same six parts. Keeping each one small makes the result easy to read.
- One workflow. A single job with a clear start and finish, like turning an emailed request into a HubSpot deal.
- One team. The people who do that job today, so the comparison is fair.
- One measure, agreed up front. Hours a week, time to first reply or the share of drafts approved without edits. We record the starting number before the AI touches anything.
- One named reviewer. A person on your team who checks the AI's work on a set schedule and can pause it at any time.
- A fixed length. We set the number of weeks before we start and stick to it.
- A decision at the end. Keep it, widen it or stop. Stopping is a fine result if the numbers say so.
What does a pilot include, and what does it leave out?
In scope
- Mapping the job as it runs today, with the people who do it
- Cleaning the records the AI will read for that one job
- Building the AI step and connecting it to HubSpot
- Testing on a batch of your past records before go-live
- Running live in assist mode with your reviewer
- A readout with the before-and-after numbers
Out of scope
- A second workflow, even a related one
- Rolling out to other teams or locations
- Letting the AI send, price, delete or merge anything without your reviewer
- A full CRM cleanup beyond what the job needs
- New software you'd have to switch to
- Big changes to the process around the AI
How does a pilot run week by week?
- Before day one
Pick and measure
We choose the job together, write down what goes in and what comes out, and pull the starting numbers from the last few months.
- Weeks 1 to 2
Build and test
We build the AI step, connect it to HubSpot and run it on a batch of past records. We compare each answer with what your team did.
- Middle weeks
Run in assist mode
The AI works on live records. Every output goes to your reviewer, who approves, edits or rejects it. Each week we read the rejections and fix the instructions.
- Final week
Readout and decision
We show the measure against the baseline, what went wrong and what it would take to widen. You make the call.
Workflows that make good pilots
The best pilots happen often, take up hours of someone's week and are cheap to get wrong. Each one below comes with the number we'd track.
New inquiry intake
AI reads each emailed or web request and turns it into a deal for the right rep. Measure: hours from request to first reply, compared with the last three months.
Custom AI workflowsCommon customer questions
AI drafts answers to the questions your team gets all day, like order status, booking changes or account details. Measure: replies sent per hour and the share of drafts sent without edits.
Customer AgentLead sorting
AI sorts incoming leads so reps only open the ones that fit what you sell and where you sell it. Measure: hours spent sorting each week.
AI sales agentRepeat customer alerts
HubSpot flags customers who are late on their usual reorder or renewal, based on past history. Measure: lapsed customers caught, and how many buy again within 60 days.
NetSuite integrationCall notes into HubSpot
After a sales call, AI fills in fields like competitor, timeline and next step for the rep to check. Measure: share of deals with complete notes.
What happens to your data in a pilot?
The AI gets only the fields the one job needs, through a business account with the AI provider whose terms we've read with you. Every action is logged, so your reviewer can trace any answer back to the record it came from.
If you stop the pilot, we switch off the AI step and remove its access the same week. Your HubSpot data stays exactly where it was, and nothing about your normal process has changed.
What does this project have to do with AI?
There was no AI in it. We moved one client from Salesforce to HubSpot and connected its sales process to its back-office system. The team now gets hours back every week. Pilots depend on that kind of groundwork. An AI that reads new requests only helps if the deal it creates lands in a CRM that's tied to the rest of the business.
When is a pilot the wrong move?
If you can't name the job yet, start with an AI readiness assessment. If no one has time to review the AI's work each week, wait until someone does, because the review is the pilot. If the job happens a few times a month, the payoff won't cover the effort, and a checklist will serve you better. And if one of HubSpot's AI agents already does the job, we may just switch it on and watch it. That's smaller than a pilot.
Examples by industry
Good first pilots change from one business to the next. These pages go deeper on a few.
- Manufacturing: turning incoming requests into deals faster.
- Construction: sorting bid invites so estimators see the right ones.
- Distribution: catching buyers who are late on a reorder.
AI pilot program questions
How long does an AI pilot last?
We fix the length before we start, based on how often the job happens. It has to be long enough to get past the rough first weeks and short enough that you decide quickly. A daily job needs fewer weeks than a weekly one.
What happens if the pilot doesn't work?
You stop, and you know why. The readout shows what the AI got wrong and whether it's fixable. Sometimes the answer is cleaner data, sometimes a different job, and sometimes AI isn't the right tool for it.
Who on our team needs to be involved?
One person who knows the job well, one named reviewer who checks the AI's work (often the same person) and a manager who makes the final decision. Expect a few hours a week from the reviewer.
Can a pilot use HubSpot's built-in agents instead of custom work?
Yes. If HubSpot's Customer Agent or Prospecting Agent fits the job, we pilot that. The shape stays the same, with one team, one measure and a person reviewing the output.
What happens after a pilot that works?
You choose. We can widen it to more teams or let it act without review on the cases it handles well. We can also start a second pilot, or hand it to your team with written notes on how it works.
How do you pick the measure for a pilot?
It should be something your team already cares about and that we can record before the AI starts. Time to first reply on new inquiries is a good one. Vague measures like "productivity" aren't.
Related
More in AI services
Pick the one job you'd test first
Bring the workflow you'd most like off your team's plate. On the call we'll tell you whether it makes a good pilot, how we'd measure it and what would need to be true before it starts.