Agents Are Arriving in Your Service Stack. They Inherit Whatever Is Already There.
· CX Pulse
HubSpot has put Agent Builder into every Professional and Enterprise account. Agents inherit your current routing and SLAs, so measure them first.
Agents Are Arriving in Your Service Stack. They Inherit Whatever Is Already There.
Last week HubSpot moved Agent Hub and Agent Builder into public beta for every Professional and Enterprise customer. Agent Hub is a single console for building, monitoring and coordinating agents across marketing, sales and service, including live status for every active agent and one-click activation for dormant ones. Agent Builder assembles custom agents on a canvas using plain language, running on the CRM context already sitting in the account.
That context list is worth reading twice if you run a service team: deal history, contact records, call transcripts, buying signals. The agent does not start from a clean brief. It starts from your data, your routing, and your working definition of a good response.
Automation inherits the process, including the parts that are broken
Support automation tends to fail for an unremarkable reason. Teams automate the response before they fix the routing. The ticket still reaches the wrong person. It just gets there faster, in a more confident tone, and now it is harder to see that it went wrong at all.
Turning on an agent does not repair a queue. It industrialises whatever the queue already does. If intent classification is guessing, the agent guesses at scale. If your first-response target is aspirational rather than measured, the agent has no standard to hold. If escalation paths are tribal knowledge held by three senior reps, the agent has no path to escalate to.
None of that is an argument against agents. We build them. It is an argument about sequence.
What to know before you switch anything on
Most service organisations cannot answer these questions with a number, which is the problem worth solving first:
- What is your actual first-response time by channel, measured rather than targeted?
- What percentage of tickets are routed correctly on the first attempt?
- Which intents drive the most volume, and which of those are self-inflicted?
- What does your published SLA say, and what do your logs say?
- Where does a customer currently have to repeat themselves?
Answer those and agent deployment becomes a scoping exercise instead of a hope. You know which workflows are stable enough to automate, which are too messy to hand over yet, and what improvement to expect. Skip them and you find out after the agent is already talking to customers.
The credit model makes measurement a budget question too
Custom agents built in Agent Builder consume HubSpot Credits when they take configured actions, with an allowance included at Professional and Enterprise and the option to buy more. That turns process waste into a line item. Every unnecessary action an agent takes because your routing sent it down the wrong path now has a unit cost attached to it. Inefficiency used to cost agent minutes. Now it also costs credits.
Measure first, then automate
The teams that get value out of this generation of tooling will be the ones who knew their baseline before they started. Not because measurement is virtuous, but because you cannot tell whether an agent improved anything if you never established what it was like beforehand.
This is what our CX audit does. We contact your business the way your customers do, across email, chat and phone. We time every reply, grade it against the standard you say you hold, and hand back a ranked plan for what to fix first. It takes two weeks and you keep the findings whether or not you work with us again.
Agents are going to land in your service stack whether you plan for it or not. Knowing what they will inherit is the cheapest work you will do all quarter.