I'm explaining it to someone else · 6 min read
What to say when the score gets challenged
You're a support leader, a HubSpot admin or a RevOps person carrying Service Hub's 2026 AI features to whoever decides — and you need answers ready for the objections that come back fast.
If you have five minutes
- They're asking three things, usually in this order: will this actually help my team, will it embarrass us if it's wrong in front of a customer, and who's accountable if it is.
- A feature list answers none of those.
1 of 5
Whoever signs off is not asking what the feature does
They're asking three things, usually in this order: will this actually help my team, will it embarrass us if it's wrong in front of a customer, and who's accountable if it is. A feature list answers none of those. The objections below are the ones the show's own takes already surfaced — not hypothetical, not softened.
2 of 5
Say this, not that
- Say: this reads every closed ticket automatically, not just the ones a survey happened to catch.
- A post-ticket survey only hears from people motivated enough to answer — usually the unhappy ones, the same reason online reviews cluster at five stars and one star with nothing in between. Customer Experience Score (CXS) reads the whole population instead of that self-selecting slice.
- Do not say: the score tells you the truth about your team.
- It doesn't, not on its own. Chris's own caution on CXS: "the signal list is lopsided. There are a couple of positive factors against roughly a dozen negative ones." An AI reading tone against a list built that way will miss positives and over-index on negatives — put your own business context and exclusions into the portal before anyone reads the number as fact.
- Say: this is a coaching tool, not a compliance tool — if you set it up that way.
- Zach's warning on Quality Assurance Score (QAS) applies to any score you put in front of people: "Any score that's gameable gets gamed. AI just applies that truth at a scale humans never could." The moment agents learn which words move the needle, you're measuring compliance with the model, not the actual experience.
- Do not say: we have proof this moves the business.
- Nobody does yet, on the record. Zach again: "what's still missing is the connection to outcomes. Right now you can get a tone score and an empathy score, but nobody's shown the line from those numbers to revenue or deal velocity yet." Promise a coaching input, not a proven revenue lever.
- Say: a human still reviews everything the agent writes into your knowledge base.
- Both the knowledge base agent and customer agent's ticket-to-knowledge feature ship with a human-approval step before anything publishes. That's a real control, and it's worth naming out loud in the room.
- Do not say: it just works — plug it in.
- Chris's own line, and the one to lead with instead of hide: "if your data quality and your processes are not in order yet, this is a shiny object: very valuable done right, and you have no chance of doing it right on a cruddy foundation." Same logic applies to automatic similar ticket suggestions — it reads your existing ticket history, and it's only as good as that history is.
3 of 5
The objections, and the honest answers
- "This is just another score my team will learn to game."
- Zach's own answer, put directly to the room: any score that's gameable gets gamed, full stop — that's true of every QA system that has ever existed, and AI just runs it at a scale a person couldn't. The honest framing is that this is a real, known limitation of scoring anything, not a defect unique to this feature.
- "Who reviews what the AI writes into our knowledge base?"
- This is the question Chris says almost nobody has actually answered. His own words: "Every one of these agents now ships with a human approval step, and almost nobody has decided who that human is. It could be a marketing manager, a sales ops person, a RevOps lead, a HubSpot admin, an IT person, or the partner doing the work." If the room doesn't have a name for that person by the end of the meeting, that's the finding to report back, not a detail to paper over.
- "What happens if nobody decides?"
- Chris's answer is specific and worth quoting directly rather than softening: "The ones that never decide end up with whatever the agent guessed, saved in the knowledge base as if it were law — and then AI treats it as law. That is how a one-off exception you made for your biggest customer quietly becomes the policy for everybody."
- "Our ticket data is a mess — will this even work?"
- Say it plainly rather than reassure past it. On automatic similar ticket suggestions: "AI can excel at trend analysis, but not over a help desk nobody has been keeping clean." The honest recommendation is to fix the queue before turning on anything that reads it.
- "Is this available on our tier?"
- Check, don't assume. Zach's take on QAS names Service Hub Enterprise only, at private beta, as of when it aired (2026-08-05 through 2026-09-02). The current live record marks QAS as live rather than private beta — which may mean it's moved, or may mean the record's status field has simply caught up to a later date than the take did. Either way, verify against the live portal before promising a tier you haven't confirmed.
- "Does this replace our existing CSAT survey?"
- No — and don't claim it does. CXS is a different measurement built off a different population (every closed ticket, versus whoever answers a survey). The two can, and probably should, sit side by side rather than one replacing the other; nothing in the record here says HubSpot intends CXS to be a survey replacement.
4 of 5
Terms that will lose the room if you use them loosely
| Term | What it means | What loses the room |
|---|---|---|
| Customer Experience Score (CXS) | A score calculated automatically on every closed ticket, reading tone against a signal list | Presenting the number as ground truth rather than a read that needs your own exclusions and business context set first |
| Quality Assurance Score (QAS) | A tone, clarity and empathy score on tickets, scored the same automated way | Treating it as proven against revenue or deal velocity — nobody has shown that line yet |
| knowledge base agent | Drafts new knowledge base articles from ticket and email history; a suggestion, not a publish | Calling it a knowledge base that writes itself — it stops at a draft, and someone has to say yes |
| customer agent (ticket-to-knowledge) | Reads resolved tickets specifically and drafts FAQ-style knowledge from them | Saying it's "the same thing" as the knowledge base agent — different source (resolved tickets vs. ticket-and-email history) and a different entry point |
| automatic similar ticket suggestions | Surfaces resolved tickets next to the one currently open, for the agent working it | Assuming it clusters open tickets on a shared theme — that's not what shipped; it's in development, not what many teams initially expect |
| deflection | The share of inbound support volume an agent resolves without a human | Quoting a deflection number as this Hub's own finding — the one deflection figure behind this door (roughly 70 percent) is Kyle Jepson's own economics argument on a different show, not a Service Hub feature claim |
5 of 5
The one sentence to hold on to
HubSpot has made your ticket history the raw material for four different AI features this year — two scores, two knowledge generators — and every one of them ships with the same open question: who is checking what comes back.
That's not a reason to wait. It's the one decision that has to be made before any of the four gets turned on, and it's the subject of the next door.
Where to next
What Service Hub is, and what changed here in 2026
The features behind the words, and what the record says about each.
Decide before you configure the Help Desk AI
The decision the room has to make before anything gets turned on.
Every Service Hub clip
The clip to send when a quote lands better than a paragraph.
Where this comes from
- Chris Carolan on "Track Customer Experience Score (CXS)". the lopsided signal list, and the instruction to set business context and exclusions before the number is read as fact
- Zach Hussion on "Track Quality Assurance Score (QAS)". the gameable-score warning, the missing line to a business outcome, and the tier he named on air
- Chris Carolan on "Customer Agent now generates knowledge from your ticket history". the human-approval step, who that human could be, and what happens in the org that never decides
- Chris Carolan on "Knowledge Base Agent". the shiny-object caution about turning this on over a foundation that is not in order
- Chris Carolan on "Automatic similar ticket suggestions in help desk". the answer to give when the room says the ticket data is a mess
- Kyle Jepson on The Road to UNBOUND, 2026-07-22. the roughly 70 percent deflection figure in the terms table — his own economics argument on a different show, named as such rather than borrowed as a Service Hub claim
What we do not know yet
- Nobody has shown the line from a tone or empathy score to a business result. Zach, on the record: "what's still missing is the connection to outcomes. Right now you can get a tone score and an empathy score, but nobody's shown the line from those numbers to revenue or deal velocity yet." Promise a coaching input, not a proven revenue lever.
- Which tier has QAS today is not settled here. Zach's take names Service Hub Enterprise only, at private beta, as of when it aired (2026-08-05 through 2026-09-02); the current live record marks QAS as live rather than private beta — which may mean it has moved, or may mean the record's status field has simply caught up to a later date than the take did. Verify against the live portal before promising a tier you have not confirmed.
- Nothing in the record here says HubSpot intends CXS to be a survey replacement, so the honest answer to "does this replace our CSAT survey" is no — the two can, and probably should, sit side by side.
Turning the Help Desk AI on, or stuck on who reviews what it writes?
Come walk through it on the show and we will work it out on air, weekday mornings.