AI SDR: what works, what doesn't, and how to evaluate one

The three kinds of AI SDR in 2026

Our AI SDR glossary entry has the one-line definition. In practice, vendors using the label sell three different things, and most buying mistakes come from mixing them up.

Notice where one of the best-known AI SDR vendors points its voice agents: inbound demand and expansion, not cold calls. Its homepage also lists a "Human Dialer" so that "when a rep makes a cold call, they get the buyer research and context to make it count" (11x, checked 2026-10-22). That split, AI on consented or warm conversations and humans on cold calls, matches both the legal rules and what buyers report.

The honest state of AI SDRs: what the evidence says

Adoption is high

• Salesforce's State of Sales survey (a vendor survey of 4,050 sales professionals, August to September 2025) reports that 87% of sales organizations use some form of AI, 54% of sellers say they have used AI agents, and 55% use AI for prospecting.
• The same survey reports the average seller spends 40% of their time selling, and that nearly half of reps name cold calling as the worst part of the job. That is the gap AI SDR tools are sold into.

Results are uneven

• Gartner predicts that by 2028 AI agents will outnumber human sellers tenfold, but fewer than 40% of sellers will report that AI agents improved their productivity. Its analyst warns of "a value ceiling" beyond which more AI does not mean more productivity.
• In March 2025, TechCrunch reported that AI SDR startup 11x had listed companies as customers that said they were not, including ZoomInfo, whose spokesperson said that during a one-month pilot the product "performed significantly worse than our SDR employees" (TechCrunch). Sources told TechCrunch that many early customers used three-month break clauses to leave; 11x disputed parts of the report and said its retention was 79%.

The pattern behind both: AI SDRs disappoint when they are bought to replace an outbound team wholesale, and do better when they take specific tasks off a team that still owns the pipeline. Gartner's own recommendations lead with piloting with clear objectives and fixing data quality before scaling.

Which SDR tasks AI does well, and where humans still win

Break the SDR job into tasks and the AI question gets much easier to answer. This map is our judgment from how the products work today; test it on your own funnel.

The phone line: consent rules for AI voices

• In February 2024 the FCC adopted a Declaratory Ruling that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act.
• The TCPA and the FCC's rules (47 CFR 64.1200) restrict calls using an artificial or prerecorded voice to cell phones without the called party's prior express consent, and require prior express written consent for such telemarketing calls to homes and mobiles.
• The FCC's consumer guide puts it plainly: "AI-generated voice calls are illegal unless the consumer has agreed to receive them or the caller is exempt" (FCC).
• These rules turn on the type of line and the consent you hold, not on whether the person is a business buyer. Many B2B cold-call lists are mostly mobiles and direct dials.

The practical upshot: an AI voice agent calling people who filled in a form and agreed to be called is a very different legal position from an AI voice agent cold-calling a purchased list of mobile numbers. Ask any voice AI SDR vendor how consent is captured, stored and checked before each call, and get counsel's view before you dial.

The economics: cost per held meeting, with worked arithmetic

Compare AI SDR options on cost per held, qualified meeting, not cost per seat or meetings booked. The inputs below are hypothetical, chosen to show the method. Replace them with your own.

On these made-up inputs the autonomous option looks cheapest per opportunity. Now change one number: if the autonomous agent's meetings qualify at 15% instead of 30%, its cost per qualified opportunity doubles to about $1,667, no better than the unassisted rep. That sensitivity is the point. Meeting quality, not meeting count, decides the answer, and it is the number vendors are least likely to show you. Track show rate and opportunity conversion by source in your CRM for every option you pilot.

AI SDR evaluation checklist

1. Which kind is it? Autonomous messaging, AI voice, or rep augmentation. Ask for a diagram of who talks to the prospect at each step.
2. Call references yourself. Ask for three current customers in your segment and phone them. Logos on a website are not evidence.
3. Pilot with an exit. Thirty to ninety days, clear success criteria, and the right to leave if they are missed.
4. Measure held meetings and pipeline, by source. Booked meetings alone reward volume over quality.
5. Review a sample of real outputs. Twenty emails or ten call recordings, unedited. Look for wrong facts, off-brand tone and claims you would not make.
6. Consent and compliance. For any AI voice: how consent is captured and checked before each call, how opt-outs and do-not-call requests are honored, how calls are disclosed.
7. CRM write-back. Does every touch, call and outcome land on the right record in your CRM, natively?
8. Data and deliverability. Where contacts come from, how numbers and emails are verified, and how sending domains and caller IDs are protected from spam labels (spam likely).
9. Human escalation. How and how fast a person takes over when a prospect is interested or upset.
10. Pricing model. Per seat, per contact, per meeting or platform fee, and the minimum term. Model your cost per qualified opportunity before signing.

A practical model: AI around the rep, human on the call

Most teams that get value from AI in outbound end up with a hybrid. Here is one way to set it up for a phone-led SDR team.

1. 1
   AI builds and cleans the listResearch, enrichment and de-duplication run automatically against your ICP. Numbers are validated before they reach a rep. See build a cold call list.
2. 2
   AI handles everything before helloA parallel dialer rings several numbers, skips voicemail, phone trees and dead lines, and answers AI call screeners, passing only live people to the rep.
3. 3
   The rep owns the conversationOpener, discovery, objections and the ask, with call guidance on screen if you use it. Train with objection handling talk tracks.
4. 4
   AI handles everything after goodbyeDispositions, notes and recordings are logged to the CRM; follow-up emails are drafted for the rep to approve; the next cadence step is scheduled.
5. 5
   AI voice only where you hold consentUse AI voice or chat agents for inbound leads and opted-in follow-up, where speed matters and consent exists, with a fast handoff to a human.

Frequently asked questions

1. 1.Salesforce (vendor survey), State of Sales report announcement, 2026 (survey of 4,050 sales professionals, Aug to Sep 2025) (2026)
2. 2.Gartner, Gartner Predicts By 2028 AI Agents Will Outnumber Sellers by 10X, Yet Fewer Than 40% of Sellers Will Report AI Agents Improved Productivity (2025)
3. 3.TechCrunch, a16z- and Benchmark-backed 11x has been claiming customers it doesn't have (2025)
4. 4.FCC, FCC Makes AI-Generated Voices in Robocalls Illegal (2024)
5. 5.FCC, Stop Unwanted Robocalls and Texts (consumer guide) (2026)
6. 6.eCFR (FCC rules), 47 CFR 64.1200, Delivery restrictions (2026)
7. 7.Artisan, Homepage (checked 2026-10-22) (2026)
8. 8.11x, Homepage (checked 2026-10-22) (2026)