HubSpot's Customer Agent Hit 70 Percent Resolution at 9,000 Customers: The New AI-Budget Audit for Sales Ops

The number to write down is seventy.

In its Q1 2026 earnings call on May 7, HubSpot disclosed that Customer Agent now serves 9,000 customers with an average resolution rate of 70%. Top performers cross 90%. According to CX Today's reporting on the call, the rate climbed five percentage points from the prior quarter. CEO Yamini Rangan singled out Customer Agent as the clearest example of AI hitting real traction across the platform.

A second number matters just as much. Customer Agent now consumes 53% of all AI credits across HubSpot's agent suite. Prospecting Agent sits at 17%. Data Agent at 16%. That split tells you where the budget is actually flowing inside customer-facing AI work.

For Sales Ops, this is the first at-scale benchmark for an AI customer agent in market. It is also a budget question your CRO will ask within the next two quarters.

What the 70 Percent Actually Says About AI-CX at Scale

Until this disclosure, AI customer agent vendor benchmarks were either pilot data (small sample, friendly customers) or single-logo case studies (impressive but unrepresentative). HubSpot's 9,000-customer number is the first time a major customer relationship management (CRM) vendor has put a portfolio-wide resolution rate on the record.

Key Facts

• 70%: HubSpot Customer Agent average resolution rate across 9,000+ customers (HubSpot Q1 2026 earnings call, May 7 2026)
• 53%: Share of all HubSpot AI credits consumed by Customer Agent (vs Prospecting 17%, Data 16%)
• 12 months: Time from launch to reach this scale and rate (HubSpot product timeline)

What that 70% covers: tickets where the agent resolved the customer's issue end-to-end, with no human handoff. The remaining 30% routed to a support agent for the cases the model couldn't close. Top performers (90%+) tend to share three traits: clean and current knowledge base content, narrow ticket scope (returns, password resets, order status, plan changes), and a clear escalation path.

That 30% gap is not a failure. It is the new floor for what your human team will see. Tickets that reach a human now skew harder, more emotional, and more revenue-sensitive. If your support headcount model assumed a flat distribution of ticket complexity, that model is out of date.

The 53 Percent Credit Skew Is the Real Sales Ops Story

The resolution rate gets the headlines. The credit-allocation number is what should land on a Sales Ops dashboard.

When 53% of customer-AI spend goes to a post-sale agent and only 17% to prospecting, the AI budget is back-end heavy. That is not automatically wrong. Customer-side intent signals are clean (the customer has an explicit ticket, in their own words). Prospecting signals are noisy (you are inferring intent from behavior). Resolution rates on cleaner signals will always run higher, so it is rational for credits to flow there first.

But "rational" is not the same as "intentional." Most Sales Ops teams did not pick this allocation. It emerged because Customer Agent shipped earlier, scaled faster, and produced visible numbers (deflected tickets, reduced response times) that justified the next budget cycle's credit purchase.

If your AI budget allocation is being set by which agent ships first, you are running an unmanaged portfolio. The fix is to put a deliberate split in front of leadership: how much credit do we want flowing to post-sale (CX), to in-funnel (sales engagement), and to pre-funnel (prospecting and data enrichment)? Then route credits accordingly.

A useful default for a mid-market sales-led company: 35% post-sale, 35% in-funnel sales workflows, 20% prospecting, 10% data quality and enrichment. Adjust by go-to-market motion. Product-led companies can run lighter on prospecting and heavier on CX. Pure outbound shops invert it.

The New Benchmark Framework: The Three-Number Sales Ops Audit

Use the HubSpot numbers as a forcing function for your own AI audit. Three numbers, three weeks, one decision document.

We call this the Three-Number Audit. Each number maps to a budget decision.

Number 1: Your resolution rate. Pull your AI customer agent (or AI chatbot) resolution rate for the trailing 90 days. If it is below 50%, your knowledge base is the bottleneck, not the model. Ship a content audit before you ship a model upgrade. If it is between 50% and 70%, you are mid-pack. The fastest gains come from scope tightening (ban the agent from billing disputes if it is failing them) and intent classification improvements. If you are at 70%+, your next dollar goes to expanding scope, not improving resolution.

Number 2: Your credit allocation by funnel stage. Run the math: what percentage of your AI credits is going to post-sale agents, in-funnel sales tools, and prospecting tools? Compare to your strategic goals. If your CEO wants pipeline growth and 60% of credits are going to support, your AI investment is misaligned with the strategy.

Number 3: Your human-handoff cost. What does a ticket that escalates to a human now cost you, fully loaded, vs. 18 months ago? If complexity has shifted but cost-per-ticket has held flat, your support model is leaking margin. Either tier-1 hiring needs to drop (replace with mid-skill problem-solvers) or escalation routing needs to skip generalist queues entirely.

A 90-rep B2B SaaS team running AI agents across prospecting, sales engagement, and CX should be able to produce these three numbers in three weeks of light effort. If they cannot, your AI deployment is opaque to the finance team and the budget will get cut next cycle.

Why HubSpot's Number Is the Right One to Anchor Against

Some sales tech buyers will (reasonably) push back. "Sierra has 90%+ for enterprise CX. Salesforce Agentforce has its own internal numbers. Why anchor against HubSpot specifically?"

Two reasons.

First, scale. Sierra's high resolution rates come from heavily-tuned deployments at large enterprises with deep services budgets. Salesforce's Agentforce numbers (where they exist publicly) reflect a similar enterprise-tuning skew. HubSpot's 70% across 9,000 customers averages over deployments from 5-person startups to mid-market enterprises, mostly self-served, mostly with default configurations. That is a closer proxy to what your own deployment will produce in year one.

Second, portability of the comparison. A 70%-at-scale number is comparable across vendors because the operational definition (no human handoff) is consistent. You can put your own number next to it on the same axis without normalizing for company size or industry. That is a rare property in vendor benchmarks.

The right read on Sierra's and Agentforce's higher numbers is not "those vendors are better." It is "those numbers represent your year-three potential, with services budget. HubSpot's 70% represents your year-one floor, without one." Both data points are useful. They serve different planning conversations.

For forecasting discipline, the year-one floor is what matters. It is what you can plan against with confidence.

What to Do This Week

The 70% number creates a board-conversation risk if you cannot answer "what is ours?" within two weeks. Three actions:

Pull your resolution rate by Friday. If you cannot, that is the finding. Make it visible. The lack of a number is the problem.

Map credit allocation to funnel stage. Get one finance partner and one revops analyst in a room for 90 minutes. Pull the credit invoice, tag each line by funnel stage, total it up. If allocation does not match strategy, document the gap and bring it to the next leadership review.

Brief your CRO on the 30% question. The 30% of tickets that reach a human now skew harder. Your support team's mix of skills should follow. If you are still hiring tier-1 generalists at the same rate as 18 months ago, you are over-staffing the layer that AI absorbs and under-staffing the layer that needs senior judgment.

Pipeline hygiene as a cultural practice and RevOps maturity model layers both already include AI-credit governance in their mature stages. If your current state document does not, this quarter is when to add it.

The Sales Ops teams that come out of 2026 ahead are not the ones running the most AI agents. They are the ones who can show their CFO a credit-per-funnel-stage chart with their numbers next to the published benchmarks. That chart is the new budget defense.

---

Learn More

• AI agents in the sales pipeline
• Measuring AI ROI without rebuilding the whole stack
• AI copilots vs AI agents: the operational difference
• The governance gap leaders get wrong about AI at work

---

FAQ

Is 70% actually a good resolution rate, or is it the floor?

For a generalist AI customer agent serving thousands of self-served deployments, 70% is the new at-scale floor. For tightly-scoped enterprise deployments with dedicated services teams (Sierra, large Salesforce Agentforce rollouts), 85% to 95% is achievable. The right comparison depends on which posture you are running. If your agent serves multiple ticket types across multiple business units with default configuration, anchor against 70%. If it serves a single business unit with curated knowledge and tight scope, plan for 85%+.

Should Sales Ops own the AI credit budget or should Finance?

Sales Ops should own the allocation logic by funnel stage. Finance should own the dollar total and the variance reporting. The split matters because credit allocation is an operational decision (what does your funnel need most?) and the dollar total is a financial decision (how much can we afford?). Mixing them produces either a budget that does not match strategy or a strategy that does not get funded. Get both functions in the same monthly review and assign clear ownership of each lever.

How does this change our prospecting AI investment?

It depends on your strategy. If your top constraint is pipeline coverage, the 17% prospecting credit share is probably too low and you should rebalance. If your top constraint is conversion rate (you have plenty of leads, not enough deals), the back-end heavy allocation may be correct. The HubSpot number is not prescriptive for your stack. It is a forcing function to make the allocation explicit instead of accidental.