What a sales AI agent actually does
Reads an opportunity's call transcripts, emails and CRM fields, scores how healthy the deal actually is, and drafts the next follow-up or CRM update for a rep to approve.
A sales AI agent for pipeline work sits on top of your CRM, not your prospecting list, and runs a loop that starts after a call or email lands: pull the transcript or message thread, read the opportunity's current stage, amount and close date, and check that against real signals (days since last contact, number of stakeholders engaged, how the conversation tone shifted) to flag whether the deal is actually on track. It then drafts the next step: a follow-up email, a CRM field update, a one-line deal-risk note for the manager, or a next-best-action recommendation, and routes anything consequential (a stage change, a discount, a close-date slip) to the rep for approval rather than acting alone. The good versions log every suggestion and every override, so you can see where the agent's read of a deal diverged from what actually happened.
This is a different job from prospecting. Outreach now bundles a dedicated Deal Agent and a Kaia call-intelligence agent directly into its Amplify platform rather than selling deal-management as a bolt-on, and Salesforce prices its own Sales Cloud agent add-on at a flat $125 per user a month on top of Agentforce's usage-based Flex Credits. Both signal that deal-stage agent work has moved from an experimental feature to a standard line item on the CRM bill.
Where it still fails: an agent can summarize a transcript accurately and still miss the real reason a deal stalled, which is almost always political (a champion lost internal support, a budget got reallocated) rather than visible in the data it has access to. And because it reads from your CRM, an agent built on top of stale or wrong pipeline data just automates bad forecasting faster instead of fixing it; it can't tell you a field is wrong, only that the field says what it says.
AI capabilities involved
Summarizing call and email history into a deal-health read
Drafting follow-up emails and next-step recommendations
Updating CRM stage and fields from call or email context
Forecasting deal risk from engagement and timeline signals
Who uses this
- B2B SaaS account executives managing 20-60 open opportunities who can't manually log every call and follow-up
- Sales managers wanting automatic deal-risk flags before the weekly forecast call instead of chasing reps for updates
- RevOps teams trying to keep CRM stage and field data accurate without nagging every rep individually
- Mid-market teams already on Salesforce or Outreach deciding whether to turn on the native deal agent before buying a separate tool
What the sales pipeline agent handles (and what it shouldn't)
| Task | Agent handles it? | Systems it touches | Human role |
|---|---|---|---|
| Logging calls and emails into the CRM timeline | Fully | Salesforce/HubSpot, email, calendar | None |
| Updating opportunity stage from call content | With approval | CRM | Rep confirms before the stage change posts |
| Drafting the next follow-up email after a call | Fully (as a draft) | CRM, email | Rep reviews and sends |
| Flagging at-risk deals before a forecast call | Fully | CRM, call intelligence | Manager reviews flagged deals |
| Generating a deal summary for exec or QBR review | Fully | CRM | None |
| Recommending a next-best-action on a stalled deal | With approval | CRM, engagement data | Rep decides whether to act on it |
| Negotiating price or contract terms | No | CRM, CPQ | Rep and manager handle it directly |
| Marking a deal won or lost | No | CRM | Rep confirms the final outcome |
Off-the-shelf AI agents on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
Outreach (Deal Agent, Research Agent, Kaia)
Teams already on Outreach for outbound wanting deal-management in the same platform
No public free tier; demo only
Contact sales; third-party estimates report Amplify Core ~$100, Plus ~$130, Pro ~$160 per user/mo, with Kaia call intelligence adding $30-$50/user/mo
Implementation reported at $5K-$25K depending on scope
Pros
- +Dedicated Deal Agent and Research Agent built specifically for pipeline work, not just prospecting
- +Kaia call intelligence feeds deal-health scoring directly from real call audio
- +Deep native fit for teams already running Outreach sequences
Cons
- −No public pricing; you have to negotiate a quote
- −Implementation fees on top of per-seat cost for larger rollouts
Salesforce Agentforce (Sales Cloud)
Teams already running Salesforce Sales Cloud
Salesforce Foundations starts free, includes Agent Builder
$125/user/mo Sales Cloud add-on, plus Flex Credits at $500 per 100k credits ($0.10/standard action, $0.15/voice action)
Agentforce 1 ("Max") edition from $550/user/mo, including 2.75M Flex Credits per org per year
Pros
- +Works natively against Salesforce's own opportunity and account data model, no separate sync needed
- +Flex Credits are usable across service, sales and marketing agents on the same org
- +Backed by Salesforce's Data Cloud for a fuller customer picture
Cons
- −Requires the Salesforce ecosystem; not usable standalone
- −Flex Credit consumption per action is hard to forecast until you've run it for a month
Apollo.io
Teams wanting account and contact research folded into the same tool reps already use for outbound
Free plan with limited credits
Basic $49, Professional $79, Organization $119 per user/mo (annual); monthly billing runs $59-$149
Custom, scoped to seats and database credits
Pros
- +AI research and account context help reps prep for calls, not just find leads
- +Database of 200M+ contacts means deal-related account research doesn't require a separate tool
- +Per-seat pricing is transparent and published
Cons
- −Built primarily as a prospecting and sequencing tool, so deal-stage scoring is lighter than Outreach's or Agentforce's
- −Export and enrichment credit limits can bite a heavy pipeline-research workflow
Clay
RevOps teams wanting a flexible research layer feeding deal context into the CRM
Free (100 Data Credits + 500 Actions)
Launch $185/mo, Growth $495/mo (repriced Mar 11, 2026); Enterprise custom
Custom, scoped to credit volume
Pros
- +"Claygent" research agent can pull fresh account and stakeholder context into a deal before a big call
- +Waterfall enrichment from 150+ data providers keeps account records current
- +AI columns let reps build custom deal-context lookups without code
Cons
- −Two separate credit meters (data credits and actions) make cost forecasting harder
- −Steeper learning curve than a packaged deal-agent product
The AI stack
A pipeline agent needs a reasoning model, access to your actual CRM and call data, tools to write back safely, and a way to catch bad scoring before it reaches a forecast call.
Reasoning model
Reads transcripts and CRM context, scores deal health, and drafts the next step
Claude Haiku 5.5
$0.10/M input tokens (up to 100K), $0.50/M outputFirst-pass logging and field extraction
Claude Sonnet 5.5
$2/M input, $10/M outputThe main deal-scoring and drafting loop
Claude Opus 5.5
$4/M input, $20/M outputWeekly or pre-forecast deep reviews of top opportunities
Our pick: Claude Sonnet 5.5 or GPT-6.1 Sol for the main scoring loop, reserving Claude Opus 5.5 for a deeper weekly pass on your largest open deals.
Knowledge & retrieval
Grounds every deal-health read in the actual CRM record and call history, not a generic guess
Native CRM data model (Salesforce, HubSpot)
Included in your CRM seatThe buy-SaaS path
Custom retrieval over CRM + call transcripts
Roughly $0.15/mo on Supabase pgvector for a small team's historyCustom builds
Our pick: Start by reading directly from your CRM's API; add a vector store only once you need to search across months of call transcripts at once.
Integrations / tools
Lets the agent write CRM updates and draft communications instead of just summarizing
Native connectors (Salesforce, Outreach, Kaia)
Included on most SaaS tiersBuy-SaaS
CRM API + email/calendar tools
Developer timeCustom builds and n8n
Our pick: Use native connectors until your stage definitions or forecast process outgrow them, then add custom tools one field at a time.
Observability & evals
Catches a mis-scored deal before it shows up wrong on a forecast call
Vendor dashboards (deal-risk scores, win-rate accuracy)
IncludedBuy-SaaS
Langfuse
Free up to 50K units/mo, then $29+/moCustom builds
Our pick: Spot-check a sample of agent-flagged deals against what actually happened every week, whichever path you take.
Reference architecture
The hardest engineering problem is separating real deal risk from noisy CRM data: an agent that trusts a stale close date or a field nobody updated will confidently forecast a deal that's already dead, so the agent needs its own confidence signal, not just the CRM's.
Call or email event
CRM, call recorder, or email syncA completed call, sent email, or calendar event triggers the agent
Pull context
CRM API plus transcript/email retrievalThe agent reads the opportunity's current fields alongside the new call or email content
Score deal health
Reasoning modelEngagement recency, stakeholder count and conversation tone combine into a risk read, flagged against the CRM's own stated stage and close date
Draft the next step
Reasoning modelA follow-up email, a next-best-action note, or a proposed field update gets drafted, not sent
Human approval
Rep's inbox or CRM task queueStage changes, discounts and close-date moves route to the rep for a yes/no before anything posts
CRM update
CRM write-backApproved updates post to the opportunity record automatically
Manager rollup
Forecast dashboardFlagged at-risk deals surface on the manager's view ahead of the forecast call
Logging and eval
Observability toolEvery score and draft is logged, with a weekly sample checked against what actually happened to the deal
Estimated cost per request
~$0.02-$0.08 per CRM-touching interaction (a call summary, a field update, or a drafted follow-up) in model costs alone, before any vendor per-seat fee
Cost calculator
Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.
Estimate your monthly cost at your own team size and call volume.
Estimated monthly cost
$1,532
≈ $18.4k per year
Calculator notes
- The per-seat line blends Outreach's reported $100-160/rep Amplify rate with Kaia's $30-50/rep add-on into a single $150/rep estimate; Agentforce's $125/user/mo plus Flex Credits can land higher or lower depending on action volume.
- Model cost applies on every path, whether you buy SaaS or build custom, since someone has to pay for the underlying reasoning calls.
- A custom build swaps the vendor per-seat fee for your own model and infra bill, typically $150-$600/mo at this volume (see build_vs_buy).
- This estimate assumes your CRM seats are already paid for separately; add those on top for a full picture.
Build it yourself with vibe-coding tools
n8n gets you a working deal-assistant in an afternoon: a trigger on completed calls or new emails, an AI Agent node with CRM read/write access, and a human-approval step before anything changes stage.
Time to MVP
1-3 days for a first version, 2-4 weeks to production quality with scoring evals and edge-case handling
Total cost to MVP
$0-$70 (n8n Cloud Starter at roughly 20 EUR/mo plus model API costs)
You'll need
Starter prompt
Set up a trigger on completed calls (a webhook from your call recorder) or new emails tied to an open opportunity (an email-sync trigger on your CRM). Feed the transcript or email into an AI Agent node using Claude Sonnet 5.5 as the model, with an HTTP Request tool for reading and writing CRM opportunity fields and a Vector Store tool over past call transcripts for the same deal. Give the AI Agent node this system prompt: 'You are a sales pipeline assistant for {Company Name}. Read the new call transcript or email alongside the opportunity's current stage, amount, close date and last-contact date from the CRM. Score the deal as on-track, at-risk, or stalled based on: days since last stakeholder contact, whether new stakeholders have joined the conversation, and whether the tone or urgency has shifted since the last interaction. If you recommend a stage change, a close-date move, or any CRM field update, do not write it directly: output a proposed update and wait for approval. Draft a specific next-step follow-up email referencing what was actually discussed, never a generic template. If you are not confident in your read of the deal, say so explicitly rather than guessing. Never invent facts about the deal that are not in the transcript, email, or CRM record.' Add a Switch node that routes proposed CRM updates to a Slack approval channel or an email digest, and only writes to the CRM after a rep approves. Log every score and draft to a Google Sheet or Postgres table so you can review a sample every week against what actually happened to the deal.
Paste this into n8n
Follow-up prompts (run in order)
- 1
Add a weekly digest email summarizing every deal the agent flagged as at-risk and why
- 2
Add a tool that pulls recent news or funding events on the account before scoring the deal
- 3
Add a check that compares the agent's risk score against the deal's actual outcome 30 days later
- 4
Add a fallback model call if the primary model times out or errors
Expected output
A working assistant that logs every call and email to the CRM, flags at-risk deals in a Slack or email digest, and drafts (but never sends) the next follow-up for a rep to approve.
Known gotchas
- !n8n bills per workflow execution, not per step, so a single call that triggers 10 tool calls still counts as one execution, but a busy pipeline week can still hit the Starter plan's execution cap
- !The AI Agent node will confidently score a deal from its training data's general sales patterns instead of your actual CRM fields unless you explicitly instruct it to only use tool results
- !Letting the agent write directly to CRM stage or close-date fields without approval is the single fastest way to corrupt your forecast
- !There's no built-in eval harness: you have to manually sample scored deals or wire in Langfuse yourself
- !Self-hosting removes the monthly fee but makes you responsible for uptime, security patches and backups
Compliance & risk reality check
A pipeline agent carries lighter compliance weight than a customer-facing one, but a few things still matter once it's writing to your CRM and drafting outbound email.
Outbound email compliance
Drafted follow-up emails still have to comply with CAN-SPAM and equivalent rules (accurate sender info, a working opt-out) once a rep sends them, regardless of who drafted the text.
Mitigation: Keep a human in the send loop, and make sure your email platform's standard compliance footer and opt-out handling apply to every agent-drafted message.
Customer data captured in call transcripts
Call transcripts often contain customer PII and sometimes sensitive business details shared in confidence during a sales conversation.
Mitigation: Restrict transcript and CRM-write access to the tools and people who need it, and add a no-training clause to any vendor contract covering call data.
Forecast and reporting accuracy
An agent that writes a wrong stage, amount or close date into the CRM can distort a forecast that feeds board or investor reporting, and the mistake compounds every time the agent re-reads its own earlier update.
Mitigation: Require human approval on every stage, amount or close-date change, and log every agent-proposed update separately from what a human actually approved.
Vendor training on your pipeline and call data
Some per-seat vendors process your deal and call data to improve their own models, and contract terms vary on whether that's allowed.
Mitigation: Add a no-training clause to vendor contracts and prefer zero-retention settings where offered.
Build vs buy: the real math
3-6 weeks for a RapidDev build; 1-3 days for a basic n8n version
Custom build time
$13K-$25K for an agency build (up to $35K for multi-system or compliance-heavy work); $0-$70/mo for DIY
One-time investment
Roughly 20-25 reps, where stacked per-seat vendor fees exceed the amortized first-year cost of a custom build
Breakeven vs buying
At 10 reps, Outreach's Amplify plus Kaia bundle runs an estimated $1,500/mo ($18,000/year), close to a $13K-$25K custom build amortized over year one at $1,083-$2,083/mo. At 25 reps, that same per-seat math hits $3,750/mo ($45,000/year), where a custom build's marginal cost per extra interaction is just model tokens, roughly $0.02-$0.08. Below about 10 reps, buying the native CRM add-on or a per-seat tool is almost always cheaper once you include an agency's ongoing maintenance cost.
Skip the DIY — RapidDev builds the production version
A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship through our AI app development service.
Discovery call (free)
30 minWe map your exact sales AI agent use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.
AI-accelerated build
3-6 weeks for a RapidDev build; 1-3 days for a basic n8n versionOur engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.
Launch + handoff
1 weekWe deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.
What you get
Timeline
3-6 weeks for a RapidDev build; 1-3 days for a basic n8n version
Investment
$13K-$25K for an agency build (up to $35K for multi-system or compliance-heavy work); $0-$70/mo for DIY
vs SaaS
ROI in Roughly 20-25 reps, where stacked per-seat vendor fees exceed the amortized first-year cost of a custom build
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
How much does an AI agent for sales cost?
SaaS pricing runs roughly $130-$160/rep/mo for Outreach's Amplify plus Kaia bundle, or $125/user/mo plus Flex Credits for Salesforce Agentforce. A custom RapidDev build runs $13K-$25K upfront (up to $35K for complex, multi-system work), plus $150-$600/mo in model and infra costs. A DIY n8n version costs $0-$70/mo but takes your own time to build and maintain.
How long does it take to set up?
A vendor add-on connected to an existing CRM takes 2-6 weeks depending on how clean your data is. A custom agency build takes 3-6 weeks because it has to be wired into your exact stage definitions and forecast process. A basic n8n version can be working within 1-3 days, though production-ready scoring takes 2-4 weeks.
Can RapidDev build this for us?
Yes. RapidDev has shipped 600+ apps and offers a free 30-minute consultation to scope your CRM, stage definitions and forecast process before you commit to anything.
Will it update my CRM automatically?
The safest setups route stage changes, close-date moves and discounts through a human approval step first, while letting the agent log calls and draft follow-ups fully automatically. Letting an agent write consequential fields without review is the fastest way to corrupt a forecast.
Can it work with Salesforce or HubSpot?
Yes. Salesforce's own Agentforce Sales add-on works natively against Salesforce data, Outreach integrates with both major CRMs, and a custom build can read and write to either platform's API directly, including fields no off-the-shelf vendor touches.
What's the difference between this and an AI SDR?
An AI SDR agent focuses on outbound prospecting: finding leads and sending the first cold outreach. This sales pipeline agent works on deals already in motion, scoring deal health, logging activity and drafting follow-ups on opportunities a rep is actively working.
Can it replace my sales reps?
No. It removes manual CRM logging and surfaces risk signals a rep would otherwise miss, but negotiation, relationship management and reading the political dynamics inside a buying committee stay human tasks.
Is my deal data safe?
Treat it the way you'd treat any vendor with access to customer and revenue data: add a no-training clause to the contract, restrict access to what the agent actually needs, and prefer vendors offering zero-retention settings.
Want the production version?
- Delivered in 3-6 weeks for a RapidDev build; 1-3 days for a basic n8n version
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.
