What a prospecting AI SDR actually does
Builds a target-account list from your ICP, researches each contact, writes a personalized cold email or LinkedIn message, and runs the follow-up cadence until someone replies.
An AI SDR agent runs the outbound motion a human sales development rep used to run manually: pull a list of accounts and contacts matching your ideal-customer profile from a B2B database, research each one for a real personalization hook (recent funding, a job posting, a tech-stack signal), draft a first-touch message and a multi-step follow-up sequence, and send it through a connected email or LinkedIn account. When a reply comes in, the agent classifies it as interested, not interested, or out of office, and routes anything promising straight to a human account executive instead of trying to close the deal itself. The agent's job ends at the first real conversation; what happens with the opportunity after that belongs to pipeline-management work, a different job entirely.
This market has had a rough year on trust. TechCrunch's March 2025 investigation of 11x found the company had listed companies that were never paying customers (including ZoomInfo and Airtable) as customers, and counted free trial usage toward its reported ARR. Partly in response to buyer skepticism about annual lock-in contracts, HubSpot moved its own Breeze Prospecting Agent to outcome pricing on April 14, 2026, charging $1.00 per recommended lead instead of a flat subscription, a sign the category is shifting toward pay-for-results billing.
Where it still fails: reps and recipients both complain that AI-written cold outreach reads as generic once you've seen a few examples, no matter how much account research went into it. High-volume automated sending without proper domain warm-up and throttling can tank your email deliverability for weeks. And vendor instability is real: Salesforce rebuilt its own Agentforce SDR agent into a different "Lead Nurturing agent" with no migration path for teams that had tuned the original.
AI capabilities involved
Researching target accounts and contacts for real personalization
Drafting first-touch and follow-up outbound messages
Classifying replies as interested, not interested, or out of office
Multi-step sequencing logic across email and LinkedIn
Who uses this
- Early-stage B2B SaaS companies without a dedicated SDR team wanting outbound running from day one
- Sales leaders wanting to scale cold-outreach volume without hiring and ramping junior reps
- RevOps teams layering AI-written sequences on top of an existing contact database like Apollo or Clay
- Teams burned by a generic-sounding AI SDR vendor looking for a more controllable, custom-prompted alternative
What the AI SDR agent handles (and what it shouldn't)
| Task | Agent handles it? | Systems it touches | Human role |
|---|---|---|---|
| Building a target-account and contact list from your ICP | Fully | B2B contact database, CRM | Reviews the list when ICP criteria are ambiguous |
| Writing personalized first-touch emails | Fully | CRM, email sequencer | Spot-checks tone and brand voice periodically |
| Sending follow-up touches on a fixed cadence | Fully | Email sequencer, LinkedIn automation | None |
| Classifying and routing replies | With approval | Shared inbox, CRM | AE takes over the moment a reply shows interest |
| Booking a meeting from an interested reply | With approval | Calendar tool | AE confirms the prospect is actually a fit |
| Answering detailed product or pricing questions | No | AE responds directly | |
| Removing a contact who asks to be removed | Fully | CRM or email suppression list | None |
| Deciding whether to pursue a non-ICP inbound reply | No | CRM | Sales manager makes the call |
Off-the-shelf AI agents on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
Artisan (Ava)
Early-stage teams wanting a self-serve AI SDR without a sales call
Free plan with 300 credits/mo
Intern $250/mo annual ($280 monthly), Employee $600/mo annual ($660 monthly)
Custom; enterprise deals reported at $1.5K-$3K/mo
Pros
- +Self-serve tiers with transparent published pricing, unusual in this category
- +Built-in B2B database and deliverability tooling, no separate data subscription needed
- +Free tier lets you test messaging before paying anything
Cons
- −Self-serve tiers are relatively new; older enterprise deals were reported as opaque and annual
- −Free tier's 300 monthly credits run out fast at real outbound volume
11x (Alice, Julian)
Teams wanting a full-service outbound product and comfortable with an annual commitment
None published
Alice Growth from $3,750/mo billed annually; another vendor page lists plans from $36,000/yr
Pro/Enterprise custom; third-party estimates put real-world cost around $50K-$60K/yr
Pros
- +Markets itself as a full autonomous outbound "AI employee" covering research, sequencing and replies
- +Julian adds a voice-agent channel alongside Alice's email and LinkedIn outreach
- +Established brand with a long market presence in this category
Cons
- −TechCrunch's March 2025 investigation found 11x listed non-customers as customers and counted trial usage as ARR
- −Annual billing with no obvious self-serve month-to-month option
- −Users report copy that reads as generic without heavy prompt tuning
Regie.ai
Teams wanting a lower-cost entry point focused on AI-written sequence content first
RegieGO free tier
Pro $49/mo; Team $150/seat (2-seat minimum); AI SEP $180/user/mo (10-seat minimum); Force Multiplier $499/user/mo (5-seat minimum)
A packaged "AI Agents" offering is reported from roughly $35K/yr
Pros
- +Content-generation tier (RegieGO/Pro) is cheap for teams that just want AI-assisted sequence writing
- +Auto-Pilot mode can run fully autonomous sending once sequences are approved
- +Native integrations with Salesforce, HubSpot and major sales engagement platforms
Cons
- −Seat minimums on the higher tiers (5-10 seats) rule out very small teams
- −Pricing tiers jump sharply between the content-only plan and the autonomous-sending plans
HubSpot Breeze Prospecting Agent
Teams already paying for HubSpot Sales Hub Pro or Enterprise
Credits included in paid Sales Hub plans
$1.00 per recommended lead (100 HubSpot Credits) since April 14, 2026; credits included: Starter 500, Pro 3,000, Enterprise 5,000/mo, extra credits $10 per 1,000
Sales Hub Enterprise tier
Pros
- +Outcome-based pricing means you pay per recommended lead rather than a flat subscription
- +Native fit if you're already running HubSpot's CRM and Sales Hub
- +Sends only through your already-connected inbox, which helps keep sender reputation consistent
Cons
- −Requires Sales Hub Pro or Enterprise (roughly $90+/seat) on top of the credit cost
- −Locked to HubSpot's own contact and sequencing data model
The AI stack
An AI SDR needs a reasoning model for research and copy, a contact and account data source, sending infrastructure that protects deliverability, and a way to catch reply-classification mistakes.
Reasoning model
Researches each contact, drafts personalized copy, and classifies replies
Claude Haiku 5.5
$0.10/M input tokens (up to 100K), $0.50/M outputHigh-volume first-touch drafting
Claude Sonnet 5.5
$2/M input, $10/M outputFirst-touch messages to your highest-value target accounts
GPT-6.1 Sol
$2/M input, $10/M outputTeams already standardized on OpenAI
Our pick: Claude Haiku 5.5 or GPT-6 Luna for volume drafting and reply classification, with Claude Sonnet 5.5 reserved for your top-tier target accounts.
Contact and account data
Supplies the list and the real signals (funding, hiring, tech stack) that make personalization credible instead of generic
Apollo.io
Free plan; Basic $49, Professional $79, Organization $119/user/mo (annual)Teams wanting an all-in-one database and sending tool
Clay
Free; Launch $185/mo; Growth $495/moCustom builds wanting the richest personalization data
Our pick: Start with whichever database your team already pays for; add Clay's research agent only once generic personalization becomes the thing prospects complain about.
Integrations / tools
Sends the actual messages and keeps domain reputation intact
Vendor-native sending (Artisan, 11x, Regie)
Included in the planBuy-SaaS
Your own email/LinkedIn automation via n8n
Developer time plus your existing email/LinkedIn accountsCustom builds and DIY
Our pick: Never skip domain warm-up and sending throttles, whichever path you take; a burned domain takes weeks to recover.
Observability & evals
Catches reply-misclassification and generic-sounding copy before it tanks your reply rate
Vendor dashboards (reply rate, meeting-booked rate)
IncludedBuy-SaaS
Langfuse
Free up to 50K units/mo, then $29+/moCustom builds
Our pick: Read a weekly sample of actual sent messages and classified replies; reply-rate dashboards alone won't catch copy that's technically on-brand but reads as generic.
Reference architecture
The hardest engineering problem is personalization that's actually true and specific, not generic padding dressed up as research: an agent that paraphrases a company's "About" page instead of citing a real, current signal gets ignored or flagged as spam just as fast as a human SDR's lazy template would.
ICP and list building
Contact database (Apollo/Clay-style tool)The agent pulls accounts and contacts matching your defined ICP criteria
Personalization research
Reasoning model plus data toolsThe agent checks for a real, current signal (funding, hiring, tech stack change) before drafting anything
Draft first-touch message
Reasoning modelA personalized email or LinkedIn message is drafted referencing the specific signal found
Send with throttling
Sending infrastructureMessages go out on a warmed-up domain at a volume that protects deliverability
Follow-up cadence
SequencerAdditional touches send on a fixed schedule until a reply or an opt-out
Reply classification
Reasoning modelReplies are sorted into interested, not interested, or out of office
Human handoff
AE inbox or CRMInterested replies route to a human account executive with full context attached
Logging and eval
Observability toolEvery message and classification is logged, with a weekly sample reviewed for copy quality and misrouted replies
Estimated cost per request
~$0.02-$0.08 per contact researched and messaged in model and enrichment costs alone, before any vendor subscription or per-lead 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 outbound volume.
Estimated monthly cost
$660
≈ $7,920 per year
Calculator notes
- The fixed line assumes Artisan's $600/mo Employee tier; Regie's entry tiers run lower, and 11x's $3,750+/mo annual plan sits well above this baseline.
- Model and enrichment cost applies on every path, whether you buy SaaS or build custom, since someone has to pay for research and drafting.
- A custom build swaps the vendor plan for your own model, data and infra bill, typically $150-$600/mo at this volume (see build_vs_buy).
- A separate contact-database subscription (Apollo, Clay) may apply on top of this if your vendor or DIY build doesn't include one.
Build it yourself with vibe-coding tools
n8n gets you a working outbound agent in a couple of days: a trigger that pulls a fresh contact list, an AI Agent node that researches and drafts each message, and a human-approval step before anything sends at volume.
Time to MVP
2-4 days for a first version, 2-4 weeks to get domain warm-up and reply handling production-ready
Total cost to MVP
$0-$100 (n8n Cloud Starter at roughly 20 EUR/mo plus a contact-database subscription and model API costs)
You'll need
Starter prompt
Set up a scheduled trigger that pulls a fresh batch of contacts matching your ICP from your contact-database tool (Apollo or Clay's API). Feed each contact into an AI Agent node using Claude Haiku 5.5 as the model, with an HTTP Request tool for pulling recent company signals (funding, job postings, news) and access to the contact's title, company and industry. Give the AI Agent node this system prompt: 'You are writing a first-touch cold outreach email for {Company Name}, which sells {one-sentence product description}. Using only the real signal provided (funding, hiring, or a specific business detail), write a 3-4 sentence email that references that signal specifically, states one concrete reason {Company Name} is relevant to them right now, and asks a single low-friction question. Never use generic flattery like "I noticed your impressive growth" without a specific fact attached. Never claim a signal you were not given. End with a clear, low-pressure call to action and a one-line unsubscribe note.' Route drafted emails through a human-approval step for the first 50 sends, then switch to auto-send once quality is confirmed, throttled to no more than 20-30 sends per hour per domain. Add a second AI Agent node that classifies incoming replies as interested, not interested, or out of office, and routes interested replies to a Slack channel or CRM task for a human AE. Log every message and classification to a Google Sheet for a weekly quality review.
Paste this into n8n
Follow-up prompts (run in order)
- 1
Add a step that checks a suppression list before sending to any contact
- 2
Add a follow-up sequence with 2-3 additional touches spaced several days apart
- 3
Add a LinkedIn connection-request step alongside the email sequence
- 4
Add a weekly digest summarizing reply rate and meetings booked
Expected output
A working sequence that researches and emails a daily batch of ICP-matched contacts, routes interested replies to a human AE, and logs every message for review.
Known gotchas
- !Skipping domain warm-up and send throttling is the single fastest way to land in spam folders for weeks
- !The AI Agent node will write generic flattery instead of specific personalization unless you explicitly restrict it to only the real signal it was given
- !Reply classification will misroute ambiguous replies (a polite "not now" that's really interested later) without a periodic human spot-check
- !There's no built-in deliverability monitoring: you have to watch bounce and spam-complaint rates yourself or wire in a dedicated tool
- !n8n bills per workflow execution, so a large contact batch can hit the Starter plan's monthly execution cap faster than expected
Compliance & risk reality check
Outbound prospecting carries lighter compliance weight than a customer-facing voice agent, but volume sending and multi-channel sequencing still create real exposure.
Outbound email anti-spam compliance
Automated outbound email at volume still has to meet baseline anti-spam requirements: accurate sender identification, a working opt-out mechanism, and prompt removal when someone asks to stop.
Mitigation: Build a suppression list into every sequence and remove a contact from all future sends as soon as they ask.
TCPA exposure if sequencing adds calls or texts
FCC Declaratory Ruling 24-17 holds that AI-generated voices are "artificial" under the TCPA and requires prior consent for outbound AI calls; the FCC's "one-to-one consent" rule for shared lead-gen consent was vacated in January 2025, but ordinary TCPA consent standards still apply.
Mitigation: If you add an outbound calling or SMS channel to the sequence, capture written consent naming AI-assisted outreach and scrub against DNC lists before sending.
EU AI Act Article 50 transparency
Applies from August 2, 2026 to any agent interacting with people in the EU, including automated outbound messaging to EU-based prospects.
Mitigation: Disclose that outreach is AI-assisted when messaging EU contacts, and keep that disclosure consistent across every channel in the sequence.
Vendor ROI and marketing claims
This category has drawn real scrutiny: the FTC sued Air AI in 2025 over deceptive business-growth and earnings claims aimed at small businesses, and TechCrunch's investigation of 11x found inflated customer and revenue figures.
Mitigation: Ask any vendor for verifiable customer references and real reply-rate data before signing an annual contract based on marketing claims alone.
Build vs buy: the real math
3-6 weeks for a RapidDev build; 2-4 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-$100/mo for DIY
One-time investment
Roughly 8,000-10,000 contacts a month, where self-serve vendor tiers cap out or force an upgrade to custom enterprise pricing
Breakeven vs buying
Artisan's $600/mo Employee tier comfortably covers a few thousand contacts a month for $7,200/year, well under a $13K+ agency build. 11x's $3,750/mo annual tier already runs $45,000/year, close to or above a custom build's amortized year-one cost before you've sent a single message. A custom build's marginal cost per extra contact is just model and enrichment tokens, roughly $0.02-$0.08, which starts to beat a per-seat or per-lead vendor fee once volume climbs into the thousands of contacts a month or your compliance needs outgrow what a self-serve tier exposes.
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 prospecting AI SDR 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; 2-4 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; 2-4 days for a basic n8n version
Investment
$13K-$25K for an agency build (up to $35K for multi-system or compliance-heavy work); $0-$100/mo for DIY
vs SaaS
ROI in Roughly 8,000-10,000 contacts a month, where self-serve vendor tiers cap out or force an upgrade to custom enterprise pricing
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
How much does an AI SDR agent cost?
Self-serve SaaS pricing runs $250-$600/mo (Artisan), while a fuller "AI employee" product like 11x starts at $3,750/mo billed annually. A custom RapidDev build runs $13K-$25K upfront (up to $35K for complex work), plus $150-$600/mo in model and infra costs. A DIY n8n version costs $0-$100/mo but takes your own time to build and maintain.
How long does it take to set up?
A self-serve vendor plan with your ICP and email connected takes 1-3 weeks. A custom agency build takes 3-6 weeks because it has to be wired into your exact ICP, CRM and brand voice. A basic n8n version can send its first sequence within 2-4 days, though production-ready deliverability 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 ICP, data sources and compliance needs before you commit to anything.
Will my outbound emails sound generic?
Only if the agent is allowed to pad messages with generic flattery instead of a real, specific signal. The fix is restricting the prompt to only reference facts it was actually given, and reviewing a weekly sample of sent messages against what a strong human SDR would write.
What's the difference between an AI SDR and a lead-generation agent?
An AI SDR agent runs outbound: it finds accounts and contacts and sends the first cold outreach. A lead-generation agent works the other direction, capturing and qualifying inbound interest from your website, chat widget or marketing campaigns.
Can it also make cold calls?
Some vendors, like 11x's Julian, add a voice-calling channel, but outbound AI calls carry TCPA consent requirements that email and LinkedIn messaging don't. Most teams start with email and LinkedIn sequencing and add calling only with proper consent capture in place.
Will I get my domain blacklisted for spam?
You can, if sending volume ramps up without warming up the domain first and throttling send rates. Vendor-native sending infrastructure handles this for you; a DIY build needs its own warm-up schedule and rate limits, typically starting at 20-30 sends per hour per domain.
Can it work with Apollo or Clay for contact data?
Yes. Most AI SDR vendors either include their own database or connect to Apollo or Clay for enrichment, and a custom build can call either API directly to source and research contacts before drafting outreach.
Want the production version?
- Delivered in 3-6 weeks for a RapidDev build; 2-4 days for a basic n8n version
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.
