The Number to Get Before You Buy an AI Support Agent
One number decides whether an AI support agent is worth funding: the share of your tickets it could close on its own. Here is how to get it before you sign anything.
Step 3: Read the Three Categories Properly
Each category points to a different action.
Category A, AI could resolve alone. These are your password resets, "where's my invoice," opening hours, basic how-to questions, and status checks. High volume, low judgement, consistent answers. This is the number that justifies (or doesn't justify) an AI support agent. Most teams find this sits somewhere in the 40 to 60% range, though it varies a lot by industry and product complexity, so treat any benchmark as a starting point, not a promise.
Category B - AI could draft for team member approval. These need a human to check the answer, but not to write it from scratch. Think nuanced account questions or anything where the right answer exists but needs a judgement call on tone or specifics. This is where AI-assisted drafting tools earn their keep, even if full automation isn't appropriate.
Category C, requires human judgement. Complaints, sensitive accounts, complex troubleshooting, anything emotionally charged or commercially significant. This work should stay with people, and that's the point, knowing your category C size tells you how much genuinely human support capacity you need to protect.
Step 4: Turn the Number Into a Decision
The category A percentage is the headline, but the top 5 ticket types are where the action is.
If the AI flags "password reset," "invoice request," and "delivery status" as your top automatable types, those are your first candidates for either a knowledge base article, a canned response, or an AI agent scope. The FAQ route is the cheapest of the three, and turning HubSpot support tickets into a ready-to-publish FAQ uses the export you have already done. Start with the highest-volume, lowest-risk types and expand from there.
If you're considering HubSpot's own Breeze Customer Agent, this number is exactly the input you need. Breeze is priced per resolved conversation (as of April 2026, $0.50 per resolved conversation - you only pay when it actually closes the ticket without personal escalation). Your category A percentage, multiplied by your monthly ticket volume, gives you a realistic estimate of how many conversations it could resolve and what that would cost versus the work hours it frees up. That is the number you will be asked to defend when the first invoice arrives, so get it before you sign the contract. Check the pile for tickets that were never questions at all while you are in there, because automating follow ups and closure on stale tickets removes some of the volume without any AI spend.
A Note on the Numbers
Treat the output as directional, not gospel. The AI is categorising based on ticket metadata, not reading the full resolution history, so it's making an informed estimate of what could have been automated, not a guaranteed forecast.
Unsure what your ticket mix would actually score?
Most teams guess high before they read their own tickets.
The real value isn't the exact percentage. It's that you move from "we should probably look at AI for support" to "47% of our volume is automatable, concentrated in these five ticket types, here's the business case." That's a decision you can actually act on.
Run it once a quarter and the trend matters more than any single reading. Pair it with auditing your HubSpot response time in five minutes using AI and you have both halves of the picture, what you could automate and where the current setup is actually slow. As you publish documentation, refine processes, and adjust your product, your automatable percentage shifts. Tracking it tells you whether your support operation is getting more efficient or just busier.
One Check Before You Run It Again
Grading tickets this way means sending real customer support content to a third party model, which counts as a disclosure of personal information under the Privacy Act. The OAIC treats that under Australian Privacy Principle 6, use and disclosure, and under APP 8 if the model processes the data outside Australia, which most do. Two practical consequences. Strip names, emails and account numbers from the export before it leaves your portal, since the categorisation only needs subject lines, ticket types, timestamps and resolution notes. And check your privacy policy actually covers overseas processing before you make this a quarterly habit.
Want Help Building the Automation Itself?
Finding your automation ceiling is the easy part. Scoping an AI agent properly, writing the knowledge base it draws from, setting the escalation rules so the human-judgement tickets always reach a team member , that's the work that makes or breaks an AI support rollout.
Neighbourhood is a Diamond HubSpot Partner. We help Service Hub teams turn "we should automate some of this" into a setup that actually holds up with real customers, which is the work described on our AI Engineering service page. If your ticket data is telling you there's capacity to claw back, get in touch.
Talk to us about your HubSpot setup.
Get More Hacks Like This
Ops Intel is our newsletter for HubSpot admins and support teams - practical hacks and no filler, straight to your inbox every week. Subscribe to Ops Intel.
We are also on Facebook and YouTube breaking down hacks like this one. Come find us there.
Scored your tickets and unsure what the middle pile means?
Most teams park the middle pile and never come back to it.
See you there.