How Much Does an AI Agent Cost in Your CRM? The Real Math Behind HubSpot and Salesforce Credits

An AI agent in your CRM no longer costs a fixed license: it costs one action at a time. As of September 2026, HubSpot and Salesforce both bill their agents on consumption models — credits and conversations — and the question shifted from "which plan do I buy?" to "how many actions will my team execute per month?". The short answer, using current official rates: resolving a conversation with HubSpot's Customer Agent costs $0.50, generating one asset with its Content Agent $10, and one Salesforce Agentforce conversation $2. For a small business, that translates into $50 to $850 per month in credits, depending on volume — and volume is exactly the variable that grows fastest.

In two weeks, the two biggest CRMs changed the rules

On September 16, HubSpot opened UNBOUND — the event called INBOUND for fifteen years — with a clear message: pure seat-based pricing is dead. The company moved to a seats + credits hybrid model and is already testing a simplified version with new customers in the Nordics and Benelux before rolling it out further. The numbers from its Analyst Day deserve attention: 19% of Pro Plus customers were using agents in August, double the start of the year; monthly agentic actions grew 3.5 times; and credit consumption doubled even though HubSpot cut agent prices in April.

A week earlier, at Dreamforce (September 15-17), Salesforce launched seven named agents — Piper for inbound, Hunter for sales, Fin for customers — and laid out its own consumption bet: $2 per conversation or Flex Credits at $500 per 100,000 credits, where every record update and case summary draws from the pool.

This is not a calendar coincidence. It is the same move: CRM software stopped selling access and started selling executed work. HubSpot even published a buyer's guide that admits it plainly: "pure seat-based pricing is fading, credit-based pricing is growing fast, and hybrid models are where the market is settling."

A credit is not a credit: the table you should demand before signing

The conceptual trap is that every platform says "credits" and no credit is worth the same. HubSpot charges $0.01 per credit; Salesforce charges $0.005 per Flex Credit — half as much, but consumed by different actions. Before comparing anything, translate everything to dollars per action:

Two readings of this table. First: the same work costs 4 times more in a CRM than in an open-source orchestrator, because in the CRM every micro-action is billable. Second: the honest comparison is not "HubSpot vs n8n" — it is "which actions in my process are genuinely agentic, and which are just deterministic logic running at a premium".

The four-step exercise: turn your volume into dollars

You don't need a sophisticated spreadsheet. You need four numbers you already know:

1. Inventory of billable actions — list the AI features you plan to enable: support agent conversations, prospecting recommendations, data responses, content assets.
2. Real monthly volume — not the ideal one: how many conversations do you actually handle per month? How many leads does prospecting really touch?
3. Credits × $0.01 — multiply volume by each action's unit cost and add it up.
4. Subtract included credits — your plan ships a few thousand per month (HubSpot's pricing program includes 5,000 on Starter, 10,000 on Professional, 15,000 on Enterprise) and they expire monthly. Overage is bought in 1,000-credit packs at $10.

Applied to three profiles we actually see in LatAm:

Look at where the money goes: in the ecommerce profile, 70% of the bill is support conversations resolved at $0.50 each. In the agency, prospecting ($1.00 per recommendation) is the line that scales. And in all three, the most expensive item on the list — content at $10 an asset — is also the easiest one to produce outside the CRM.

Why the bill grows even when prices fall

Here is the cross-reference almost nobody makes: HubSpot reported that its credit consumption doubled in a year while cutting agent prices. When unit price drops and total bill rises, volume is growing faster than efficiency — and that is exactly what happens with agents: every adoption win (19% of Pro Plus customers using them, 55% using Breeze Assistant) converts into more billable actions.

This is not a HubSpot-only problem. According to the Zylo 2026 SaaS Management Index (218 IT leaders surveyed), 78% experienced unexpected AI or consumption charges in the past year. And the McKinsey State of AI 2026 (September 23) finds that a third of organizations now spend more than 10% of their technology budget on AI, 60% plan to increase that spend, and one in five already feels operational cost pressure. Against that backdrop, Gartner has been warning for months that more than 40% of agentic AI projects will be canceled by 2027, with escalating costs as one of the three main causes.

The client who asks "how much does the agent cost?" is asking the wrong question. The right one is: how many actions will it execute, at what unit price, and who turns off the meter when it runs away? And if "cost per action" sounds new, it is — as we broke down in what an AI agent really costs per task, until now the conversation was about tokens, not actions billed by a CRM.

Our stance: deterministic work should not live inside the meter

After auditing stacks for clients paying CRMs with credits, the rule we apply is simple: keep all deterministic work out of the credit meter, and reserve credits for work that genuinely requires judgment.

• Deterministic and volumetric — sends, syncs, assignments, reports, reminders — outside: run it in n8n or Make, which bill per execution (or don't charge per step at all). A 10-step workflow on self-hosted n8n still costs a fraction-of-a-cent server execution; the same flow billed per action inside a CRM eats the budget.
• Agentic and valuable per interaction — customer question resolution, prospect research, qualification — inside: there, $0.50 or $2.00 per outcome compares favorably with a person doing the same thing at 11 p.m.
• Hard limits from day one: HubSpot lets you cap credits per agent, set alerts, and pause features; Salesforce supports org-level pools. An agent without a cap is an invoice without a ceiling — it is literally how Uber burned its entire annual AI budget in four months.
• Beware "free": Make opened AI Agents to all plans including Free, but every execution consumes credits and the Free plan has 1,000 per month — as we analyzed, the practical ceiling is around 150-160 executions. "Free" always means "free until you scale".

Before you enable any paid agent

1. Budget with the four steps above using real volume, not sales projections.
2. Request the full rate sheet — action by action — and verify the rep did not quote only the per-conversation price.
3. Decide which actions will be agentic and which will run outside the CRM before buying, not after.
4. Configure credit caps and alerts on deployment day.
5. Review consumption at 30 days: if it grows more than 50% without equivalent business growth, the agent is doing too much.
6. Add the human cost — review, corrections, escalations — that credits do not capture.
7. Remember included credits expire: treat them as a trial, not a saving.

Per-action pricing is neither good nor bad: it is exact. The problem starts when you buy "an agent" with a license mentality and a factory of actions sends the invoice. At Mintec we budget agents in credits before implementing them — if you want to see what that number looks like for your case, contact us and we will send back the math, not the demo.

FAQ

Is an agent in the CRM cheaper than hiring someone? It depends on volume and hours. At $0.50 per resolved conversation, 1,000 conversations a month cost $500 in credits — far below a salary, but far above a deterministic automation that resolves half those cases without a language model. The real question is how many of those 1,000 need genuine judgment.

Do credits roll over? No. Credits included with your subscription reset every month and unused ones expire. Additional pack terms vary by contract, but the expiration logic rewards consumption, not savings.

Can I move my credit budget to another platform? Yes, and our framework recommends it: portability is exactly the advantage of keeping volume logic in an orchestrator like n8n or Make, while treating the CRM as the system where data and CRM-context actions live. The CRM automation patterns we actually run and the Make vs n8n breakdown cover that decision with numbers — and if you are unsure whether an action even needs AI, our AI agents vs traditional automation framework answers it.