Every week someone asks me which AI model they should use for their first agent. It's the wrong first question. The model is the engine. Your CRM is the road. If the road is a goat track, a better engine just gets you lost faster.
This guide is for the business owner or ops lead who is about to build an agent on HubSpot and wants to understand three words we use constantly: objects, pipelines and associations. No other jargon. Once these three click, you'll understand why some agents work on day one and others never do.
An object is a type of thing your business keeps records about. HubSpot ships with a handful as standard:
Beyond those, HubSpot has objects built for particular kinds of business, such as Services, Listings, Courses and Appointments, and on Enterprise you can create custom objects for almost anything else. We wrote about the newer objects HubSpot added when they first arrived.
Each record has properties, which are just its fields. A contact has an email, a phone number and a lifecycle stage. A deal has an amount, a close date and an owner.
Why this matters to an agent: an agent reads objects the way a new staff member reads a filing system. If a "job" is sometimes a deal, sometimes a ticket and sometimes a note on a contact, the agent has no idea where to look. If a job is always a deal in the Jobs pipeline, it knows exactly where to look, every time.
Write down the five things your business handles every day. Customers, quotes, jobs, invoices, bookings, whatever they are for you. Next to each, write the HubSpot object that holds it. If you can't fill in a line, or you write two objects on one line, that's your first fix, well before any agent.
A pipeline is a set of stages a record moves through. Deals and tickets have pipelines out of the box, and so do several of the newer objects. A sales pipeline might run: New enquiry, Qualified, Quote sent, Negotiation, Won, Lost. A delivery pipeline might run: Kick-off, Waiting on client, In build, Review, Complete.
For a person, a pipeline is a picture of where things are up to. For an agent, a pipeline is a set of instructions. Each stage tells it what's true about the record right now and what should happen next. "Quote sent" plus "no reply in five days" is a clear trigger for a follow-up draft. "Waiting on client" plus "nothing uploaded in three days" is a clear trigger for a polite chase.
Three things make a pipeline agent-ready:
Associations are where most of the value hides, and where most CRMs fall down. An association is a link between two records. This contact works at this company. This deal belongs to this contact. This ticket is about this customer's project.
Associations let an agent answer the question every good staff member asks before they act: "What else do I need to know?" When a support ticket comes in, an agent with good associations can see the contact, their company, the open deal, the last few tickets and the service the customer pays for. An agent without them sees an email address and a complaint.
HubSpot also lets you label associations, such as "decision maker" or "billing contact", which helps an agent pick the right person to write to. And you can copy key values across associated records, which we covered in our post on sync properties.
Put the three together and you get something an agent can follow like a map. Here's an illustrative example for a training provider:
An agent picking up that ticket doesn't have to guess. It can see who the student is, what they enrolled in, when, and what's already been said. That's the structure we lean on in our builds for training providers.
Compare that with the version we see more often: enrolments live in a spreadsheet, the course name is typed into a free-text field three different ways, and questions land in a shared inbox. A person can muddle through with memory. An agent can't, and it won't tell you it's confused. It will give a confident wrong answer instead. That's the heart of why most AI agents fail on the data, not the model.
This is the Context step in our method: check the data, pick one job, agree what done means, and measure what that job costs you today. For the data check, I'd work through these in order:
None of this needs AI. It's plain CRM housekeeping, and it's the part most agent projects rush. If you're early in your HubSpot setup, our HubSpot implementation guide for Australian SMBs covers the structure in more depth. Once the map is clean, an agent built on HubSpot's own agents, or a custom agent connected to HubSpot, has something solid to follow.
Not sure whether your objects, pipelines and associations are ready to carry an agent? Book an AI Readiness Audit and we'll check them against the one job you want to hand over first, then tell you plainly what to fix. You can also read how we work to see the full method from Context through to Result.