Here's a pattern we see constantly in portals: a team spends real time configuring Customer Agent, clicks into Batch Testing, and then spends another hour manually typing in one question at a time. Meanwhile, they've got a spreadsheet of 200 support tickets sitting right there in another tab.
That gap is exactly what this update closes. Upload the CSV. Run the tests. See the gaps.
What This Update Actually Is
HubSpot added CSV upload support to the Batch Testing tool inside Customer Agent. That's the core of it.
Before this, every test question had to be entered manually. Now you can upload a CSV with two columns: a required Question column and an optional Expected Answer column. HubSpot converts each row into a test case and runs them all in one batch.
When you include expected answers, they appear side by side with what Customer Agent actually said. No toggling between documents. No copy-pasting. Just a direct comparison in one view.
This feature is available with Customer Agent on Smart CRM, Marketing Hub, Sales Hub, Service Hub, Content Hub, and Data Hub. All at Professional and Enterprise tiers. Customer Agent does consume HubSpot Credits, so keep that in mind as you scale your test sets.
Why HubSpot Shipped This
AI agents don't get better by accident. They get better because someone runs structured tests, finds the gaps, and fixes the knowledge sources or guidelines behind them.
The problem is that testing one question at a time doesn't scale. Most teams already have hundreds of real customer questions sitting in support inboxes, FAQ docs, and old ticket exports. Making those humans re-enter every question by hand is friction that kills the testing habit entirely.
HubSpot's answer: meet the data where it already lives. If the questions exist in a spreadsheet, let that spreadsheet do the work. The optional expected answer column adds a second layer of accountability. Instead of eyeballing whether a response feels right, you now have a written benchmark to compare against.
How to Use It Step by Step
- Open Customer Agent in your HubSpot portal and navigate to the Batch Testing section.
- Click Upload CSV. Download HubSpot's template if you don't already have a file ready. The template has two columns: Question and Expected Answer.
- Pull in your existing question sources. Good candidates: your FAQ page content, Zendesk or Help Scout ticket exports, internal support scripts, and any questions your sales team flags as objections.
- Add expected answers in the second column for any question where accuracy really matters. You don't have to fill every row. Focus on high-stakes questions first: pricing, refund policy, technical specs, compliance details.
- Upload the file. HubSpot turns each row into a test case automatically.
- Run the batch. Review the side-by-side output. Where the agent's response doesn't match your expected answer, trace back to the knowledge source or guideline that's causing the gap and update it.
- Re-upload and re-run after each round of changes. Treat batch testing like a QA sprint, not a one-time task.
What It Touches in Your HubSpot Strategy
This update lives inside Customer Agent, but its ripple effect touches more than just the AI configuration tab.
Service Hub teams will feel the most immediate impact. If you're routing conversations through Customer Agent before human handoff, the accuracy of those first responses shapes the entire ticket volume your reps handle. Better testing means fewer bad escalations.
Key Takeaway
Batch Testing with CSV is a quality control system for your AI agent. Use it on a regular cadence, not just at launch. Every time you update a knowledge source, run a new batch to confirm you didn't break existing responses.
Marketing Hub teams running Customer Agent on landing pages or chat widgets should map their test questions to their content. If your agent is supposed to answer questions about a campaign offer, your test set should include the specific edge-case questions that offer creates.
From a RevOps perspective, this update makes Customer Agent auditable in a way it wasn't before. You can now document what the agent was expected to say, compare it to what it actually said, and create a change log as you refine the configuration. That's the foundation of responsible AI governance inside your CRM.
If you haven't already, pair this with the update that brought the Customer Agent tester to every configuration page. That update lets you spot-test guidelines and handoffs in context. This CSV update lets you stress-test at scale. They're complementary, and you should be using both.
Key Takeaway
Build your CSV test set from real data. Past support tickets, your top 50 FAQ page questions, and any question that caused a customer complaint in the last 90 days are all better sources than questions you invent from scratch.
One thing to watch: Customer Agent runs on HubSpot Credits. A large CSV with hundreds of rows will consume credits each time you run the batch. Know your credit balance before you run a 500-question test set, especially if you're iterating quickly.
Who Should Care Most
This update will matter most to a specific set of roles and company types. Here's who should move on it now:
- Service Hub admins who manage Customer Agent for support deflection. Faster testing means faster improvement cycles.
- RevOps leaders who want an auditable QA process for AI-generated customer responses, not just a vibe check.
- Customer success teams at mid-size companies where the AI agent handles onboarding questions. Incorrect onboarding answers cost real retention. Test before those conversations happen.
- Marketing ops humans who deploy Customer Agent on campaign or product pages and need to verify that the agent reflects the latest offer details.
- Any team in a regulated industry where the agent's responses have compliance implications. The expected answer column gives you a written record of what a correct response looks like.
If your company has fewer than 50 customer questions in total and low chat volume, manual entry is probably fine for now. But if you've got an existing knowledge base of any real size, the CSV upload is the faster path, and you should be using it.
George's Take
What excites me most about this isn't the CSV upload itself. It's what the expected answer column unlocks. For the first time, teams can build a living benchmark for their AI agent, a written record of what good actually looks like. That's the difference between teams that deploy AI and hope for the best and teams that deploy AI and know it's working. When I see humans using this tool on a weekly cadence, treating it like a QA sprint instead of a launch checkbox, that's when Customer Agent starts to really flourish inside a portal.
“The expected answer column isn't a nice-to-have. It's the foundation of AI accountability. If you can't write down what a correct answer looks like, you're not ready to deploy the agent to real customers.”
If you're building out a full AI-assisted service layer, this pairs well with the work HubSpot is doing on Customer Success Space and onboarding plans in the client portal. Better agent testing and better client-facing onboarding tools point in the same direction: more confident customers, fewer support escalations, and service teams that can actually scale.
If you want help building a structured testing process for your Customer Agent or auditing your current AI configuration, that's exactly the kind of work we do at Sidekick Strategies. Book a strategy call and let's look at your portal together.
Frequently Asked Questions
What is CSV Batch Testing for HubSpot Customer Agent?
It's a feature inside HubSpot's Customer Agent Batch Testing tool that lets you upload a CSV file of test questions instead of entering them one at a time. You can include expected answers in a second column, and HubSpot will show your agent's actual responses next to those benchmarks so you can spot gaps quickly.
What format does the CSV file need to be in?
The CSV needs a Question column and an optional Expected Answer column. HubSpot provides a downloadable template inside the Batch Testing tool. You can also use your own CSV as long as it matches that column structure. Each row becomes a separate test case.
Does running batch tests use HubSpot Credits?
Yes. Customer Agent runs on HubSpot Credits, and each test case in a batch consumes credits. If you're running a large CSV with hundreds of questions, check your available credits before starting. Prioritize high-stakes questions in your first test runs to manage credit usage efficiently.
Which HubSpot plans include CSV Batch Testing for Customer Agent?
This feature is available on Professional and Enterprise tiers across Smart CRM, Marketing Hub, Sales Hub, Service Hub, Content Hub, and Data Hub. It's not available on Free or Starter plans.
How often should I run batch tests on my Customer Agent?
Run a batch test any time you update a knowledge source, change a guideline, or add a new use case to your agent. Treat it like a QA sprint, not a one-time setup task. Teams that test on a consistent cadence catch regressions before customers do.
What's the best source of questions to include in my test CSV?
Start with real data: past support tickets, your existing FAQ page, questions your sales team hears on calls, and any question that caused a customer complaint in the last 90 days. Real questions from real humans will expose gaps that invented test questions won't.
