Skip to content
Primores AI wiki

AI SDR — What an AI Sales Agent Does, Costs and Proves

AI SDR (AI Sales Agent)

TL;DR: An AI SDR is software that does the work of a sales development representative: it finds and researches leads, writes and sends outreach, answers replies and books meetings. It comes in four forms: agents built into a CRM (HubSpot, Salesforce, Microsoft), standalone products (11x, Artisan, AiSDR), agents that talk to website visitors (Qualified), and workflows you build yourself (Clay). No independent test compares named products on the same leads. The one aggregate available shows AI getting fewer replies per email than human reps (2.9% against 4.7%). Test one on your own leads for four to six weeks, counting held meetings, before signing an annual contract.

Simple Explanation

A sales development rep (SDR) is the person who starts sales conversations: builds a list, writes the first email, follows up, and hands a booked meeting to the salesperson. An AI SDR is an glossary/ai-agent given that job. The cheap part of the work (research, drafting, sending) is what the software does well. The parts that decide whether a meeting happens (a good list, a real offer, a reply that gets the facts right) still depend on what you feed it.

The four kinds

KindExamplesWhat it works fromWhen it fits
CRM-native agentHubSpot Prospecting Agent, Salesforce Agentforce, Microsoft Dynamics 365 Sales agentsThe contacts, deals and history already in your CRMYou’re already on that CRM and want the lowest-effort test
Standalone AI SDR11x (Alice), Artisan (Ava), AiSDR, Regie.aiYour list or the vendor’s contact database, plus sending mailboxesCold outbound at volume
Inbound website agentQualified (Piper)Your site traffic and CRMYou have real traffic; it talks to people who came to you, so far fewer email and legal problems
Build-your-ownClay plus a sequencerData providers and prompts you chooseYou want control and have someone to build it

Voice agents (Retell and others) do the same work by phone, under stricter rules (below).

What they cost (listed prices, November 2026)

ProductListed priceRead
HubSpot Prospecting Agent100 credits per lead at $10 per 1,000 credits, so about $1 per lead, plus HubSpot seats11 Nov
HubSpot Customer Agent (support, not prospecting)50 credits per resolution (about $0.50); 50 credits per voice minute11 Nov
Salesforce Agentforce$2 per conversation; or $500 per 100,000 Flex Credits (20 credits per standard action) with a $5 per user per month licence; or $125 per user per month for unmetered employee use. The pricing page names no separate SDR agent11 Nov
Microsoft Dynamics 365 SalesSeats at $65, $105 or $150 per user per month, paid yearly; agents run on Copilot Credits (Premium includes 1,000 per user a month)9 Nov
AiSDR$1.50 per person prospected, pay as you go, no monthly fee; warmed-up mailbox $33; fully managed $2,500 a month11 Nov
Regie.aiFree; Pro $41 a month; Team $125 per seat a month (billed yearly)9 Nov
ClayFrom $54 a month (yearly) for the smallest paid tier; the page shows conflicting figures, so get a quote9 Nov
Retell AI (voice)$0.07–0.31 per minute9 Nov
11x, Artisan, QualifiedNo public price9 Nov

Two things follow from the table. Pricing per outcome (HubSpot per lead, AiSDR per person, Salesforce per conversation) makes a small test cheap. A vendor that offers only a demo and an annual contract puts the whole risk on you. These prices change often; every figure above is a list price on the date shown.

Does it work? The evidence is thin

We found no independent test of named products on the same lead list, no audited retention figure from any vendor, and no named customer publishing cost per held meeting. What exists:

  • Vendor-reported (HubSpot, Q3 2026 call). Customers using its prospecting agents see “76% more qualified leads and 80% more meetings than those who don’t”. That compares adopters with non-adopters, with no method, baseline or sample published. Teams that adopt early are probably better run to begin with.
  • An aggregate from an agency that sells AI services (Digital Applied, April 2026). Reply rates of 4.7% for human reps, 2.9% for AI and 3.6% for a mix. It publishes no raw data. Read it as a direction: AI is cheaper per email and gets fewer replies per email.
  • A cautionary case (TechCrunch, March 2025). 11x listed customers who said they weren’t customers. ZoomInfo: “We did not give them permission to use our logo in any manner, and we are not a customer.” Airtable said the product “was never used in production”. A former employee claimed 11x was “losing 70-80% of customers that came through the door”; 11x said its “retention rate is currently 79%”. Anonymous claims against a company statement, and old, but a reason to check any logo on a vendor’s homepage.

So whether an AI SDR is worth it comes down to cost per held meeting on your own list, which is the same logic as glossary/cost-per-completed-task: price the finished result, not the unit of activity.

What goes wrong

  • Your sending domain. Google requires anyone sending more than 5,000 messages a day to Gmail to use SPF, DKIM and DMARC and one-click unsubscribe, and to keep spam complaints “below 0.30%” (its monitoring target is below 0.10%). Yahoo sets the same 0.3%. Send from a separate domain, so a climbing complaint rate doesn’t touch the one your team uses. The full picture is on marketing/email-deliverability.
  • The law covers B2B email. The FTC’s CAN-SPAM guide: “The law makes no exception for business-to-business email”, with penalties of up to $53,088 per email, and a duty to “monitor what others are doing on your behalf”. EU and UK rules on B2B email differ by country; we haven’t verified them from a regulator’s page, so check your own authority’s guidance.
  • LinkedIn automation. LinkedIn doesn’t permit third-party software that automates activity on its site, and accounts that use it can be restricted.
  • AI voices on the phone (US). In February 2024 the FCC ruled that AI-generated voices are “artificial” under the TCPA, per the law firm Cooley’s summary, so robocall consent rules apply to AI calling. That is a consent requirement, not a ban.
  • Wrong facts in replies. No one has measured how often AI SDRs misstate things when a prospect answers. Read every reply thread in the first two weeks.
  • Text that reads as AI. Covered on marketing/ai-human-voice-prompting and marketing/ai-tells-in-sales-copy.

The pilot: test on your own leads first

This is our method, built from the evidence above, for outbound email. Allow four to six weeks.

  1. Take at least 1,000 leads of one customer type from your CRM and split them in half: one half to the agent, one to your best person, same offer, same weeks. With a few hundred leads, a difference of two or three meetings is chance.
  2. Give the agent its own sending domain with SPF, DKIM and DMARC, and cap daily volume. Watch the complaint rate every few days.
  3. Agree the target in writing before starting. Count held meetings and opportunities, not opens or replies. A workable target: at least as many held meetings per 100 leads as your person, at a lower cost per meeting.
  4. Read the replies yourself for two weeks. Stop if complaints climb or the agent says something false about your product.
  5. Compare cost per held meeting: what the tool cost plus the time someone spent minding it, against your person’s cost for the same leads.

For an inbound or CRM agent the test is simpler: one queue, 30 days, compared with the 30 days before.

Common Misconceptions

  • ❌ Myth: An AI SDR replaces the SDR team.

  • ✅ Reality: In the only aggregate available, AI alone gets fewer replies than people, and the mixed setup sits between the two. It changes what a meeting costs, which you can only learn on your own list.

  • ❌ Myth: A vendor’s customer logos and ARR prove the product works.

  • ✅ Reality: The 11x case shows both can be overstated. Ask for a paid pilot with a meeting target instead.

  • ❌ Myth: B2B cold email is outside spam law.

  • ✅ Reality: CAN-SPAM makes no B2B exception, and you answer for what a tool sends in your name.

Key Takeaways

  • An AI SDR automates lead research, outreach, reply handling and meeting booking; it comes built into a CRM, standalone, as a website agent, or self-built.
  • Listed prices in November 2026: about $1 per lead (HubSpot), $1.50 per person prospected (AiSDR), $2 per conversation (Salesforce); 11x, Artisan and Qualified publish none.
  • No independent head-to-head test exists. The one aggregate shows 2.9% replies for AI against 4.7% for human reps.
  • The main risks are the sending domain, spam and calling law, LinkedIn account restrictions and wrong facts in replies.
  • Run a split test on at least 1,000 of your own leads and judge it on cost per held meeting before any annual contract.

Sources

Do-not-cite: HubSpot’s 76% / 80% as the agent’s effect (adopters against non-adopters, no method); “AI SDRs book more meetings than people” (no independent test; the one aggregate points the other way); an unverified 100,000-email comparison in circulation (source page gone); 11x’s $5,000-a-month price (third-party, unverified); “the FCC banned AI voice calls” (it applied existing consent rules).

By Andrej Ruckij