I'm implementing it · 4 min read
Decide before you configure the Help Desk AI
The decision to turn these on is made, or nearly, and you need the order things go in and what each step actually costs before anyone touches a setting.
If you have five minutes
- Every feature in this door reads the same underlying thing — your Help Desk ticket queue — so the decisions below aren't per-feature.
- They're prerequisites that sit underneath CXS, QAS, the knowledge base agent, customer agent's ticket-to-knowledge feature, and automatic similar ticket suggestions all at once.
Decide before you configure
Every feature in this door reads the same underlying thing — your Help Desk ticket queue — so the decisions below aren't per-feature. They're prerequisites that sit underneath CXS, QAS, the knowledge base agent, customer agent's ticket-to-knowledge feature, and automatic similar ticket suggestions all at once. Get them wrong once and every feature built on top inherits the mistake.
Five decisions, named owners, what it costs to get them late
| Decision | Who decides | What it costs to get it late |
|---|---|---|
| Who reviews what an agent drafts into the knowledge base | Someone named — a marketing manager, sales ops, RevOps, a HubSpot admin, IT, or the partner doing the work | Per Chris's own words: "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." A one-off exception for one customer quietly becomes policy for everybody. |
| What business context and exclusions get set on CXS | Whoever owns the Help Desk queue's configuration | The signal list is lopsided by design — "a couple of positive factors against roughly a dozen negative ones" — so an unset score will miss positives and over-index on negatives before anyone notices. |
| Whether your ticket data is clean enough to trust an automated read | Whoever owns Help Desk data hygiene | "AI can excel at trend analysis, but not over a help desk nobody has been keeping clean" — a messy queue produces a confident answer, not a right one, and confidence is the part nobody questions after the fact. |
| The review cadence for agent-generated knowledge | Whoever owns the knowledge base | Chris's own recommendation: "Start monthly, not daily. Twenty suggested knowledge moments, agree on one." Daily review at scale burns the reviewer out before the tool proves itself. |
| Whether other input sources feed the knowledge base beyond tickets | Whoever owns product and support alignment | A knowledge base fed only by resolved tickets and email never covers a feature that shipped before any support ticket about it exists — plan for other sources as the base grows. |
What each step actually costs, from the operators' own words
On business context.
"So do not take the number and run with it. Put your own business context into the portal, the terms and moments that actually signal something for your business, before you let this score drive a decision." — Chris Carolan, on Track Customer Experience Score (CXS)
On review ownership.
"This is where you switch what the humans are responsible for. The person who used to write that answer does not stop being the expert; they stop drafting and start confirming. That only works if confirming is easy. If review means copy-pasting into a Google Doc and trading comments back and forth, you are not going to build knowledge anywhere." — Chris Carolan, on Customer Agent now generates knowledge from your ticket history
On data quality.
"This class of feature is only ever as good as the data already in your system - AI can excel at trend analysis, but not over a help desk nobody has been keeping clean." — Chris Carolan, on Automatic similar ticket suggestions in help desk
The order things go in
- Get your Help Desk ticket data clean and current first.
- It is the raw material every feature below reads, and it is the recurring caveat across every take in this pack — not a nice-to-have step before the "real" work starts.
- Confirm your tier and beta status against your own live portal, not this page.
- CXS and QAS both moved status between when their takes aired and when this record was last read; Zach's QAS take named Service Hub Enterprise only, private beta, at the time it aired. Verify before promising a rollout date to anyone.
- Name the person who reviews agent-drafted knowledge, before either the knowledge base agent or customer agent's ticket-to-knowledge feature is turned on.
- This is the single decision Chris calls out by name as the one most organizations skip.
- Set CXS exclusions and business context in the portal before any team is measured against the score.
- This is configuration, not an afterthought — do it before the first report goes to a manager.
- Start the knowledge-review cadence monthly, not daily, and pick one suggestion to act on rather than trying to clear the whole queue.
- Twenty suggested knowledge moments, agree on one — that's the pace Chris names on the record.
- Only after the above, turn on automatic similar ticket suggestions.
- It's in development as of its first covering (2026-08-31), so plan around it changing rather than around it being final, and it inherits whatever state your ticket data is already in.
Two features with no take, worth naming before you build a rollout plan around them
Not everything in this vein has been discussed on air yet, and this door doesn't invent commentary where none exists. Two examples from the wider Help Desk feature set, both filed under this Hub, both carrying a status but no on-air take: Workforce Management in Help Desk (public beta, no first-covered date recorded) and the New Help Desk Composer Experience (public beta, no first-covered date recorded). If either is part of your rollout, the record gives you a status and nothing more — treat that as license to test it yourself before writing procedure around it, not as a gap to fill with a guess.
Credit: full takes cited above — Chris Carolan, Track Customer Experience Score (CXS).
Chris Carolan, Customer Agent now generates knowledge from your ticket history.
Chris Carolan, Automatic similar ticket suggestions in help desk.
And the show, New Customer Agent Coaching Opportunity: Action Gap.
Where to next
What Service Hub is, and what changed here in 2026
The features these decisions sit underneath, with the status the record carries for each.
What to say when the score gets challenged
The words for the room you have to get these decisions out of.
Every Service Hub clip
The mornings these decisions were argued out on air.
Where this comes from
- Chris Carolan on "Track Customer Experience Score (CXS)". business context and exclusions set in the portal before the score drives a decision
- Chris Carolan on "Customer Agent now generates knowledge from your ticket history". review ownership, what happens to the org that never decides, and the monthly cadence he names
- Chris Carolan on "Automatic similar ticket suggestions in help desk". in development as of 2026-08-31, and the data-quality precondition under everything else
- The show on "New Customer Agent Coaching Opportunity: Action Gap". the caveat that neither host has run Customer Agent in anger, and where they pointed instead
- The Service Hub material pack, built 2026-09-05. the status and first-covered date for Workforce Management in Help Desk and the New Help Desk Composer Experience, from a live read of the HubSpot Updates object
What we do not know yet
- Nobody in the material behind this door has run customer agent "in anger." On New Customer Agent Coaching Opportunity: Action Gap, the show's own caveat was on the record: "neither has run Customer Agent in anger" — Madelyn pointed at Zach as the trusted source for the agents specifically because the hosts had not operated one themselves. That caveat should travel with any implementation plan built from this alone.
- Whether the human-approval step is configurable per role has not been read from a live portal. The takes describe that review happens, not the mechanics of assigning or routing that review inside HubSpot's own permission model.
- No independent, non-AOM operator account of implementing these Help Desk features exists in the archive searched for this material. The three genuine outside-show hits found in a 1,402-episode search were about Customer Success Workspace and Customer Agent economics, not Help Desk implementation — this door's operator quotes are all from the show's own takes, named as such throughout, not supplemented by an outside implementation account that does not exist yet.
- No client engagement and no engagement figures appear here, in keeping with the confidentiality this door is written under — every quote above is public, on-air commentary, not private consulting material.
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.