The short answer: an AI sales/CRM agent is usually billed in three layers: CRM seats, an AI licence/add-on, and usage by action, conversation, or outcome. At public list prices, Salesforce Agentforce add-ons are USD 125/user/month, Flex Credits USD 500 per 100,000 (about USD 0.10 per action), and conversations USD 2 each; HubSpot Breeze charges USD 1 per recommended lead and USD 0.50 per resolved support conversation; Microsoft 365 Copilot is USD 30/user/month for enterprise. For 20 sales users and 5,000 leads a month, the AI usage can range from hundreds to thousands of dollars — before CRM seats, implementation, data work, and human review.
This guide is for B2B manufacturing sales, marketing, IT, and procurement leaders. Data cut-off: 25 October 2026. Prices come from Salesforce, HubSpot, and Microsoft public pages; they are USD list prices under annual/contract terms, before tax and discounts. The models are editorial estimates, not quotes, and do not promise lead growth, win rate, or payback.
A seat fee is not an AI fee. An AI fee is not the total cost. Without CRM data and human review, an agent only produces wrong follow-ups faster.
How are AI sales agents priced?
The common units are CRM seats, an AI add-on/licence, actions, conversations, recommended leads, and outcomes. Most deployments stack three or more; comparing one unit price is misleading.
Sales agents usually sit inside a CRM: they read accounts, opportunities, emails, calls, and tickets; recommend next steps; score leads; draft follow-ups; or schedule meetings. Vendors separate the CRM seat, AI capability, and AI usage. Some charge per user, some per action/conversation/recommended lead, and some include basic AI in a higher tier.
| Vendor/product | Public pricing | Good fit | What is often missed |
|---|---|---|---|
| Salesforce Agentforce | Flex Credits USD 500/100k (20 credits/action ≈ USD 0.10); Conversations USD 2 each; Agentforce add-on USD 125/user/month; User License USD 5/user/month plus credits; Max from USD 550/user/month | Existing Sales/Service Cloud, complex processes, larger teams | CRM base seats, Data 360, implementation, minimum commitment, credits that do not roll over |
| HubSpot Breeze | Prospecting Agent USD 1 per recommended lead; Customer Agent USD 0.50 per resolved conversation; Pro/Enterprise required | Mid-sized B2B already on HubSpot; outcome-aligned buying | HubSpot seat/tier, credit usage, whether a recommended lead is qualified |
| Microsoft 365 Copilot | Enterprise USD 30/user/month; Business USD 23.50/user/month with Copilot; advanced agents/usage may be extra | Microsoft 365, Office/Teams-centric sales teams | Base M365 licence, Copilot tier differences, usage-based advanced features |
Public list prices on 25 Oct 2026, USD, annual/contract terms; tax, discounts, minimum commitments, and regional prices are extra. Product scopes differ; the lowest unit price is not a ranking.
What do 20 reps and 5,000 leads a month cost?
At public usage prices, the AI layer is roughly USD 600–4,000/month for 20 reps and 5,000 leads, before CRM seats, implementation, and human review.
Assumptions: 20 sales users; 5,000 inbound/outbound leads per month; 1,000 enter recommendation/follow-up; Salesforce averages 3 actions per lead; HubSpot charges for 1,000 recommended leads; Microsoft is priced only by 20 Copilot licences. Actual actions, resolutions, lead definitions, and tiers change the result.
| Option | Calculation | AI usage/licence per month | Not included |
|---|---|---|---|
| HubSpot Breeze Prospecting | 1,000 recommended leads × USD 1 | ≈ USD 1,000 | HubSpot seats/tier, implementation, human follow-up |
| Salesforce Agentforce (Flex Credits) | 5,000 leads × 3 actions × 20 credits = 300,000 credits; USD 500/100k | ≈ USD 1,500 | Sales Cloud seats, Agentforce add-on, Data 360, implementation |
| Salesforce Agentforce (Conversations) | 2,000 sales conversations × USD 2 | ≈ USD 4,000 | Sales Cloud seats, conversation definition, implementation |
| Microsoft 365 Copilot | 20 users × USD 30 | ≈ USD 600 | Base M365, usage-based advanced agents, implementation |
Editorial model, not a quote. Excludes tax, CRM base subscription, integration, data governance, human review, and retries. A HubSpot recommended lead is not a closed deal. Salesforce credits and conversations cannot be mixed in the same org.
CRM-native, standalone, or build?
Use a native agent if you already have a mature CRM; use a standalone agent when data is fragmented; build only when the sales process is truly differentiated and you can sustain engineering.
| Option | Advantages | Limits | Best for |
|---|---|---|---|
| CRM-native (Salesforce/HubSpot/Microsoft) | Account, opportunity, email, and ticket context already present; permissions, audit, and reporting integrated | Seats + add-on + usage; lock-in and minimums; customisation constrained | Existing CRM, standard sales process, fast launch |
| Standalone sales agent | Can connect several CRMs, email, and forms; choose models per task | Data sync, permissions, duplicates, and audit are your job; vendor risk | Fragmented CRM, cross-system follow-up |
| Build in-house | Full control of data, models, routing, and cost | Engineering, operations, evaluation, security; easy to end up with a chat box | Sales process is core differentiation; existing AI team |
Editorial judgement, not a vendor ranking. Native agents mainly add value through data and process, not a “smarter model”; self-build costs are often underestimated in integration, permissions, and ongoing evaluation.
Per seat, per action, or per outcome?
Seats suit stable internal use; actions suit complex, predictable processes; outcomes suit clearly measurable results with an acceptable minimum commitment.
| Model | Advantage | Risk | Ask the vendor |
|---|---|---|---|
| Per seat/user | Predictable budget; broad adoption | Paying for unused seats; advanced features/usage extra | Does it include AI usage and advanced agents? |
| Per action/conversation | Aligned to usage; start small | Action definition, retries, and long workflows can inflate the bill | What counts as one action? Are tools/retries included? |
| Per outcome | Tied to business results; easier to explain | Outcome definition, minimums, attribution disputes | How is a recommended lead/resolution/opportunity defined? |
There is no universally “cheapest” model. Price the same lead volume, follow-up actions, and outcome definition before comparing.
Why is “AI email writing” not a sales agent?
Writing is only generation. A sales agent must read the CRM, prioritise, plan follow-up, update records, obey permissions, and stop on opt-out or sensitive promises.
A usable sales agent should identify the account/opportunity, read history and preferences, recommend the next action, draft a reviewable email/call script, write back to the CRM, create tasks, and stop or hand off on opt-out, sensitive commitments, or low confidence. A model connected only to an inbox lacks context and a closed loop; it can create duplicate outreach, wrong promises, and dirty data.
What should acceptance measure?
Measure qualified-lead rate, response time, meeting/opportunity conversion, CRM completeness, outreach compliance, cost per qualified meeting, and human edit rate — not email volume.
| Metric | Suggested definition | What to verify |
|---|---|---|
| Qualified-lead rate | AI-recommended leads accepted/advanced by sales / recommended leads | Recommendation criteria match the ICP |
| Response time | Lead arrival to first effective touch | Faster, not just more mass email |
| Meeting/opportunity conversion | AI-assisted leads vs baseline, same definition | Attribution window, rep skill, lead source |
| CRM completeness | Records with required fields / new records | Duplicates, wrong ownership, opt-outs |
| Compliance | Opt-out, privacy, promise, and frequency violations | Audit logs and human review |
| Cost per qualified meeting | (AI + CRM + human + implementation) / qualified meetings | Auditable denominator, not just tokens |
| Human edit rate | AI drafts edited/discarded / drafts generated | Whether it creates extra work |
Set thresholds by industry, deal size, and sales cycle; there is no universal “AI recommendation accuracy”. High-ticket B2B should keep human confirmation by default.
How should the POC and contract be structured?
Run a four-to-six-week shadow POC on 10–20 real accounts, with humans sending and logging edits; write data, action/outcome definitions, minimums, opt-out, and exit into the contract.
- Scope: one product line, one sales pod, 10–20 real accounts/leads; no full-list blasts.
- Shadow: AI suggests; sales reviews and sends. Log edits, rejections, replies, and meetings.
- Baseline: same response time, qualified rate, meeting rate, and CRM completeness; do not compare with a vague “before AI” impression.
- Contract: customer data/training use, action/outcome definitions, minimums, overage, opt-out/privacy, model/supplier exit, audit.
- Payment: tie stages to data access, POC, launch, and stable operation. A demo email is not acceptance.
Is Lingyi a fit?
If you need private deployment, CRM/ERP integration, and permission control, Lingyi can provide the AI foundation and workflow integration. If you need a standard CRM-native agent, buy the vendor directly.
Lingyi does not sell a CRM and does not promise lead volume or win rate. LingHarness can serve as a private AI foundation for model access, permissions, logs, and agent execution; LingWork covers internal office workflows. If you already use Salesforce, HubSpot, or Microsoft, evaluate their native agents first; consider custom integration only when data is fragmented, private deployment or complex permissions are required. Related reading: AI customer service cost, LLM API cost, and agent base comparison.
FAQ
How much does an AI sales agent cost?
There is no universal quote. It is usually CRM seats + AI add-on/licence + actions/conversations/outcomes. Salesforce Agentforce add-ons are USD 125/user/month with Flex Credits at USD 500/100k; HubSpot charges USD 1 per recommended lead; Microsoft 365 Copilot is USD 30/user/month enterprise. Add CRM base, implementation, and human review.
Is outcome-based pricing always cheaper?
No. Outcome definitions, minimum commitments, whether a recommended lead is qualified, and attribution windows change the effective unit price. At high volume, action/outcome pricing can exceed a seat plan; at low volume or stable workflows, seats/monthly plans are more predictable.
CRM-native or standalone agent?
Use native when you already have a mature CRM, complete data, and standard processes. Use standalone or custom when CRM systems are fragmented or private deployment is required. Include data sync, permissions, audit, and duplicate handling in the standalone cost.
Can AI send cold email or quotes automatically?
It can draft, but do not send quotes, discounts, or commitments without review. High-ticket B2B should keep human confirmation, opt-out detection, frequency limits, and audit; sales/legal should approve pricing and contract terms.
Is a sales agent worth it for a small team?
Check lead volume, response time, and CRM data first. If leads are few and data is incomplete, fix the CRM and follow-up process before buying an agent; otherwise it only produces duplicate or wrong follow-ups faster.
How should we accept a sales agent?
Run a four-to-six-week shadow POC on real accounts and compare qualified-lead rate, response time, meeting/opportunity conversion, CRM completeness, compliance, cost per qualified meeting, and human edit rate. Also log how often AI creates extra work.
Sources, method, and limits
This is public-source research and an editorial model, not a vendor quote, interviews, or sales testing. Prices come from official pages accessible on 25 October 2026, in USD list prices; tax, discounts, minimum commitments, region, and version differences are extra. The model excludes CRM base subscriptions, implementation, data governance, human review, and retries; it does not promise lead volume, win rate, or payback.
- Salesforce, Agentforce Pricing (accessed 25 Oct 2026): Flex Credits USD 500/100k; 20 credits/action; conversations USD 2; add-on USD 125/user/month; user licence USD 5/user/month.
- HubSpot, Customer Agent and Prospecting Agent pricing (13 Apr 2026): Prospecting USD 1/recommended lead; Customer Agent USD 0.50/resolved conversation; Pro/Enterprise.
- Microsoft, Microsoft 365 Copilot pricing (accessed 25 Oct 2026): Enterprise USD 30/user/month; Business USD 23.50/user/month; advanced AI usage may be extra.
The Chinese version is at AI 销售 Agent 成本与选型.