Service Hub

Service Hub's 2026 story is what happens when AI starts reading your Help Desk ticket queue — scoring it, learning from it, and suggesting what to do with it next.

LatestEnhanced Meeting & Call Transcription for NotetakerSep 4, 2026

What we think

The show's throughline take is that HubSpot has quietly made your ticket history the best AI training ground in the whole product — and every feature built to read it ships without anyone yet in charge of checking what it hands back.

Look across what actually shipped in Help Desk this year and the same shape holds no matter who on the show is talking about it. Chris watched HubSpot start scoring every closed ticket automatically with Customer Experience Score (CXS), and his caution was specific: the tool's own signal list runs a couple of positive factors against roughly a dozen negative ones, so an AI reading tone against a list built that way will miss the good calls and over-index on the bad ones the same way a person would — put your own business context into the portal before you let the number drive a decision. Zach's take on Quality Assurance Score (QAS), scored off that same ticket data for tone, clarity and empathy, landed on the identical warning from a different door: any score that's gameable gets gamed, and nobody has shown the line yet from a tone score to an actual business result. Then the show turned to what those tickets can build instead of just measure. The knowledge base agent drafts articles from ticket and email history, and customer agent goes a step further, generating FAQ-style knowledge straight from resolved tickets — Chris called both a real shortcut for a portal sitting on years of support history and no knowledge base, and both times he landed on the same unfinished piece: every one of these agents ships with a human-approval step, and almost nobody has decided who that human is. Automatic similar ticket suggestions closes the loop from the other side, surfacing resolved tickets next to the one an agent is working, and carries the caveat one more time in Chris's own words: this only works as well as the Help Desk data already sitting there. That same shape keeps recurring: Help Desk is where AI is doing its most immediately useful work in Service Hub right now, and the show's standing take is that the value is real but conditional on somebody actually doing the review step HubSpot shipped as optional.

Why this is hard

Every one of these tools does real work by reading Help Desk tickets nobody wrote with an AI audience in mind — and that only holds up if a specific person is checking what comes out the other end.

CXS and QAS both score off a fixed or lopsided signal list — CXS runs a couple of positive factors against roughly a dozen negative ones, and QAS's tone-and-empathy categories aren't customizable yet — so reading the number as truth means trusting the shape of a checklist, not your customers.
The knowledge base agent and customer agent's ticket-to-knowledge feature both ship with a human-approval step, and on Chris's read almost nobody has decided who that human is — the org that never decides ends up with whatever the agent guessed, saved as if it were law.
Automatic similar ticket suggestions and every scoring tool alike are only as good as the data already sitting in the Help Desk — a ticket queue nobody has been keeping clean trains a confident answer instead of a right one.

Where are you coming from?