Your CRM vendor would very much like to sell you an AI agent. HubSpot now charges per resolved conversation, Salesforce per conversation or per action, and both promise agents that answer customers, qualify leads and keep records up to date. At the same time, building your own agent on top of the CRM has never been easier.
So which route is right for a small or mid-sized company? This guide compares HubSpot Breeze, Salesforce Agentforce and custom agents on price, capabilities and risk, walks through a worked cost example, and ends with a six-question framework you can use in your next budget meeting.
What “AI in the CRM” means in 2026
The label covers five quite different jobs:
- Customer-facing agents that resolve support requests or qualify inbound leads in chat, email or voice.
- Seller assistance: meeting summaries, automatic field updates, account research, email drafts.
- Data hygiene: deduplication, enrichment, extracting structured fields from emails and calls.
- Analytics: questions in plain language (“which deals stalled this month and why?”) and auto-generated reports.
- Workflow automation: routing, reminders, follow-ups and hand-offs between teams.
The second category is quickly becoming a standard platform feature: meeting notes and CRM updates are built into Breeze and Agentforce. The money and the difficulty sit in the first and fifth categories, where agents take actions with consequences.
The wider market favours buying over building. In 2025, 76% of enterprise AI use cases were bought rather than built in-house, up from 53% a year earlier (Menlo Ventures). MIT NANDA research found that buying from a specialised vendor or partner succeeds about 67% of the time, versus about 33% for internal builds (Fortune). “Buy” here includes platform agents and partners who build on top of them.
Option 1: HubSpot Breeze
Since 14 April 2026 HubSpot prices its Breeze agents by outcome (Resolve247):
- Customer Agent: $0.50 per resolved conversation.
- Prospecting Agent: $1 per recommended lead.
- Credits: $10 per 1,000 for additional AI usage.
HubSpot also offers a 28-day trial, which makes a low-risk pilot easy.
Where it shines: companies that already run marketing, sales and service in HubSpot, standard support questions answered from a knowledge base, and teams without in-house developers. Outcome pricing means you pay nothing for conversations the agent fails to resolve.
Where it struggles: like every platform agent, Breeze is strongest inside its own data model. As soon as the agent needs to check stock in an ERP, change a booking in another system or issue a refund through a payment provider, you are into integration work — and the quality of that work determines the result.
Option 2: Salesforce Agentforce
Salesforce offers several pricing models (Salesforce):
- $2 per conversation.
- Flex Credits: $500 per 100,000 credits; a standard action costs 20 credits, which is $0.10 per action, and voice actions cost $0.15.
- Per-user add-ons: $125 per user per month, industry editions $150.
- Agentforce Max: from $550 per user per month.
Where it shines: organisations with mature Salesforce instances, complex permission models and strong governance requirements. Agentforce works with Salesforce data, flows and security out of the box.
Where it struggles: cost modelling is genuinely complex, because the same use case can be priced per conversation, per action or per user. Getting value usually needs an experienced Salesforce admin or partner, and for smaller teams the per-user editions can outweigh the benefit.
Option 3: a custom agent connected to your CRM
A custom agent uses model APIs (OpenAI, Anthropic, Google or open-weight models), an orchestration layer — workflow tools such as n8n or plain code — and the CRM’s API. Increasingly the connection runs over the Model Context Protocol; see MCP explained.
Typical market prices: $10–30k to automate one process and $30–80k for several (Digital Applied). Model usage for an average agent at scale runs $50–500 a month (Lets-Viz), plus maintenance — in-house or as a retainer.
Where it shines: processes that cross systems, CRMs without strong native AI (Pipedrive, Zoho, amoCRM, Bitrix24 and many others), strict data residency requirements such as EU-only hosting, the freedom to switch models, and high volumes where per-outcome fees add up.
Where it struggles: you own the quality. Without evals, monitoring and someone responsible after launch, a custom agent degrades quietly.
Side-by-side comparison
| Criterion | HubSpot Breeze | Salesforce Agentforce | Custom agent |
|---|---|---|---|
| Pricing model | Per outcome ($0.50 per resolution, $1 per lead) | Per conversation ($2), per action ($0.10) or per user | Build project plus model usage and maintenance |
| Time to first value | Days to weeks | Weeks | Weeks |
| Best fit | SMBs living in HubSpot | Mid-market and enterprise on Salesforce | Cross-system processes, any CRM |
| Actions in other systems | Via integrations | Via integrations and Salesforce tooling | Native to the design |
| Model choice | Vendor’s choice | Mostly vendor’s choice | Yours |
| Data residency control | Vendor’s terms | Vendor’s terms | You choose hosting, e.g. EU-only |
| Lock-in risk | High (platform) | High (platform) | Low if code, prompts and data are yours |
| Who maintains it | Vendor plus your admin | Vendor plus admin or partner | You or your partner |
The cost math: a worked example
Pricing pages are hard to compare, so put your own numbers into one model. Here is an illustration with assumed volumes: 4,000 customer conversations a month, of which the agent fully resolves 3,000.
| Option | Calculation | Monthly cost |
|---|---|---|
| HubSpot Breeze Customer Agent | 3,000 resolutions × $0.50 | $1,500 |
| Salesforce, per conversation | 4,000 conversations × $2 | $8,000 |
| Salesforce, Flex Credits (assuming 4 actions per conversation) | 16,000 actions × $0.10 | $1,600 |
| Intercom Fin, for reference | 3,000 resolutions × $0.99 | $2,970 |
| Custom agent | One-off build (e.g. $20k) + model usage (a few hundred $) + maintenance | Depends on volume and scope |
Prices from Resolve247, Salesforce and Intercom. Volumes are assumptions for illustration — check current price lists.
Three conclusions follow:
- For standard support inside HubSpot, the platform agent is hard to beat on cost. A custom build would need years to pay back against $1,500 a month — unless it does things Breeze cannot.
- The pricing model matters as much as the vendor. The same Salesforce use case costs $1,600 or $8,000 depending on how it is billed and how many actions each conversation triggers.
- Define “resolved” in writing. HFS Research expects the definition of a resolved interaction to become the next battleground in service contracts (HFS). Does a customer who comes back the next day count? Make it measurable before you sign.
For pricing models beyond this example, see our guide to outcome-based pricing for AI.
Use cases that pay off — and one that usually does not
Inbound speed-to-lead. A human inside sales agent costs $3–5k a month, while AI tools for instant response and qualification cost $99–500 a month (Rework). The CRM is where the lead, the conversation and the booked meeting must end up. More in speed-to-lead with AI.
Support resolution. 66% of service organisations already use AI agents, up from 39% in 2025, and 70% of them report measurable impact within 60 days (Salesforce).
Data hygiene and meeting notes. Low risk, quick wins, and largely included in platform features.
What usually disappoints: replacing your SDR team. AI SDR products show 50–70% annual churn; human-led pipelines are about 2.6 times larger, and meeting show rates are 71% for human-booked versus 52% for AI-booked meetings (Forma Nôrden). Use AI to make sellers faster, not to replace them.
The decision framework: six questions
- Where does the data live? If the customer record, history and knowledge base are already in the CRM, start with the platform agent.
- Does the agent need to act in other systems? Refunds, bookings, stock, contracts — every external action means integration work and is the strongest argument for a custom or hybrid design.
- What do the unit economics look like at your volume? Model 12 months on per-outcome fees versus build and run costs, using your real numbers.
- What are your regulatory constraints? In the EU, Article 50 of the AI Act has required chat and voice agents to disclose that they are AI since 2 August 2026 (Cooley), and the GDPR governs the data the agent processes. In Germany, data protection requirements (66%) and legal uncertainty (56%) are the top barriers for companies using AI (Bitkom).
- How much lock-in can you accept? Platform agents tie your AI logic to your CRM contract. If you might switch CRMs in three years, keep the logic portable.
- Who owns it after launch? Platform agents still need an admin; custom agents need monitoring and evals. Name the owner before you start.
A pragmatic default for most SMBs: platform first, custom where the platform stops. Many good setups are hybrids — the CRM’s agent handles standard questions, while a custom agent handles the cross-system actions. Our build vs buy framework goes deeper.
Questions to ask before you sign
Whichever route you take, these questions surface the hidden costs and risks early:
For platform agents (Breeze, Agentforce):
- What exactly counts as a billable resolution, conversation or action, and can we audit the count?
- Which actions can the agent take in external systems out of the box, and which need custom work?
- Can we cap monthly spend, and what happens when the cap is reached — does the agent stop or hand over to humans?
- Where is conversation data processed and stored, and which sub-processors are involved?
- Can we export the agent’s configuration, knowledge sources and conversation logs if we leave?
For custom agents and partners:
- Will the code, prompts, eval sets and logs live in our accounts from day one?
- Which models are used for which steps, and how easily can they be swapped?
- How is quality measured before launch and every month afterwards?
- What does the monthly running cost consist of — model usage, hosting, maintenance — and is there a budget cap in the contract?
- Who is on call when the agent misbehaves, and what is the response time?
A good vendor or partner answers all of these in writing. Vague answers are an early warning sign, regardless of the brand.
A six-step implementation plan
- Clean the data first. Deduplicate contacts, agree field definitions, archive dead records. An agent on dirty data produces confident nonsense.
- Pick one use case with a baseline. Measure today’s resolution rate, response time or lead-to-meeting conversion before anything changes.
- Pilot for two to four weeks. Use vendor trials (HubSpot offers 28 days, Intercom Fin 14) or a fixed-price pilot with an agreed metric.
- Keep humans on risky steps. 49% of large US companies have explicitly defined scenarios where agents may not decide alone (KPMG). Refunds, discounts and contract changes belong there.
- Measure and log everything. Every agent action should be visible in the CRM timeline, with a weekly quality review against your eval set.
- Scale only what works. Add the next use case when the first one meets its metric.
Common mistakes
- Switching the agent on before cleaning data. The agent inherits every duplicate and outdated field.
- Buying per-user add-ons for everyone. Start with the people who will actually use the features.
- Leaving “resolution” undefined. You will pay for outcomes you would not count as outcomes.
- Letting the agent write to records without an audit trail. When something goes wrong, you need to know what changed and why.
- Building custom what the platform already sells for $0.50. Engineering budget belongs where the platform stops.
- Forgetting disclosure and data protection in the EU. Compliance is part of the product, not an afterthought.
Conclusion
There is no universal winner. HubSpot Breeze is the cost-efficient default for SMBs whose work lives in HubSpot. Salesforce Agentforce suits organisations with mature Salesforce operations and the expertise to model its pricing. Custom agents earn their keep when the work crosses systems, when data residency or model choice matter, or when per-outcome fees at your volume outgrow the cost of running your own. In every case, clean data, a measurable definition of success and human control over risky actions decide the outcome more than the logo.
If you want a second opinion on your CRM setup, our Systems team designs and integrates CRM agents, and our Agents team builds the custom parts. Book a free 15-minute audit and we will help you run the numbers on your own volumes.