AI SDRs are good at the top of the funnel and weak everywhere after it. One builds and enriches prospect lists, summarises account research, drafts and sends first-touch email and LinkedIn messages, sorts inbound replies and books meetings, at a volume no person can match. It does not qualify buyers in conversation, handle an unstated objection, make a confirmation call, or notice when a buyer has gone quiet for a reason. Nor will it screen a UK list for the people you are not allowed to email.
That is the short answer. The longer one is that "are AI SDRs any good" cannot be settled at the level of the category, because the category is not one thing. Below is what the software does task by task, what the vendors publish as list prices, the four places the work reliably falls over, and the cases where I would buy one anyway. Every external number here comes from a source I opened, named and linked. Where no source exists, I say so rather than borrow a figure.
What does an AI SDR actually do?
Less than the name promises, and more than the sceptics allow.
Start with the label. Gartner's press release of 25 June 2025 gave the problem a name: "agent washing", which it defines as the rebranding of existing products such as AI assistants, robotic process automation and chatbots "without substantial agentic capabilities". Gartner's estimate in that release is that of the thousands of vendors marketing agentic AI, only about 130 are real. That estimate covers agentic AI generally. Sales is where the marketing is loudest. A fair share of products sold as AI SDRs are sequencing tools with a language model bolted onto the copy step.
It helps to know what the job consists of before asking what a machine can take off it. The Bridge Group's SDR Models, Motions and Metrics study, published by Matt Bertuzzi on 6 February 2025 from 351 B2B companies, puts the median SDR's day at 112 activities: 44 phone, 41 email, 19 LinkedIn and 8 text or other. Those 112 activities produce a median of 4.1 quality conversations a day.
An AI SDR goes after the 41 emails. It does not go after the 44 dials or the 4.1 conversations, whatever the demo suggested.
Task by task, what holds and what does not
| Task | What the software does well | Where it fails | What a person still does |
|---|---|---|---|
| List building | Pulls and dedupes thousands of records from a bundled database in minutes, then enriches them from public sources. | Cannot tell a live buying unit from a stale org chart, and will not know that part of your list are sole traders with different legal protection. | Defines the account criteria, sets the exclusions, and reads fifty rows by hand before anything sends. |
| Account research | Reads a website, a filing, a job advert or a press release and summarises it in seconds, at any volume. | Weighing which fact matters to this buyer this quarter. When the source material is thin it invents. TechCrunch reported 11x customers complaining about "hallucinating" products. | Decides which signal actually justifies an approach, and bins the rest. |
| First touch | Drafts and sends at any hour and in any language, without getting bored on a Thursday afternoon. | The structure is legible. Your buyer has read forty emails built the same way this month. | Writes the offer. The offer is the one thing that cannot be generated from the prospect's own website. |
| Reply handling | Sorts out-of-office, unsubscribe and wrong-person replies without human attention. | The two-line reply that is half interest and half objection, which is most of the replies worth having. | Reads intent and answers in the buyer's own words within the hour. |
| Objection handling | Retrieves an approved answer to a stated objection quickly and consistently. | Recognising that the stated objection is not the real one. "No budget" usually means "no priority". | Effectively all of it. |
| Qualification | Checks stated criteria against form fields and CRM data. | Budget, authority, timing and appetite are disclosed in conversation, not in fields. Software cannot earn the disclosure. | Asks the awkward question that changes the answer. |
| Booking | Offers slots, handles time zones, sends the invite and chases the confirmation. The strongest automated step on this list. | Very little. This part genuinely works. | Almost nothing, which is the point of automating it. |
| No-show recovery | Fires a reminder sequence at a predictable cadence. | A no-show is usually a verdict on how weak the original interest was. Software cannot re-sell the meeting. | Picks up the phone. ORRJO holds 90%+ meeting attendance, and the mechanism is a person ringing the day before. |
Read the right-hand column as a shopping list. Everything in it is work you are still paying for after the licence.
How much do AI SDRs cost?
Published list prices run from $49 a month to $2,500 a month. Two of the five products below publish no price at all. And no two vendors sell the same unit, so the prices cannot be compared without doing conversion work the vendors have chosen not to do for you.
Here is what each vendor puts on its own pricing page. I have listed only products whose pricing page I opened directly.
| Product | Published list price | What the price is measured in | Commitment |
|---|---|---|---|
| AiSDR Solo | $250 a month | 200 AI-researched contacts a month | Monthly |
| AiSDR Explore | $900 a month | 800 contacts a month, with a managed campaign at $149 each on top | Quarterly |
| AiSDR Scale | $2,500 a month | 2,500 contacts a month, with a fully managed service at $2,500 a month on top | Quarterly |
| Reply.io AI SDR Starter | From $500 a month | 1,000 active contacts | Billed annually |
| Reply.io AI SDR Growth | From $1,500 a month | 5,000 active contacts | Billed annually |
| Regie.ai Pro | $49 a month | 5,000 credits a month, no rollover | Monthly |
| Artisan, Ava | Not published | "Leads contacted" per month, band unstated | Not published |
| Qualified, Piper | Not published | Custom plan built with their sales team | Not published |
List prices move. Open each page yourself before you quote any of these back to a finance director.
Four different units in eight rows. Contacts, active contacts, credits and leads contacted are not the same thing, and a credit is whatever the vendor decides it is this quarter. Regie.ai's own page says research, drafting, enrichment and dialling all consume credits, and that credits reset each billing period rather than carrying over. So the headline $49 buys an amount of work nobody outside the company can size in advance.
The most revealing line in that table is AiSDR's. It sells a fully managed service at $2,500 a month on top of its $2,500 software tier. The vendor of an autonomous sales agent will sell you people to run it.
The arithmetic nobody puts on the page
The licence is the smaller half of the cost. The other half is the internal hours it consumes, and those hours are real: defining the account criteria, approving copy, warming domains, watching deliverability, reading the ambiguous replies, screening the list for people you should not be contacting, and cleaning up the CRM afterwards.
To price them in sterling, two figures you can check. The ONS Annual Survey of Hours and Earnings puts median gross hourly earnings excluding overtime for UK full-time employees at £19.67 in April 2025. Employer National Insurance for a standard category A employee runs at 15% in the 2026 to 2027 tax year. That is a fully loaded floor of roughly £22.62 an hour, before pension, software, desk or management.
Work it through. Eight hours a week is about 34.7 hours a month, or roughly £785 a month of your own people. Set that against a $900 licence and you are running a programme costing the licence plus £785, before domains, inboxes, data and the sending infrastructure. Eight hours is not a benchmark and I am not offering it as one, because I have found no primary source that measures it. Log your own hours for a fortnight and use your number.
Two currencies in one paragraph is not carelessness. The licences are priced in dollars and your people are paid in sterling, and reconciling that, plus VAT, plus a moving exchange rate, is a genuine cost of buying American software from a British company. None of the pricing pages mention it.
For comparison, ORRJO's Lead Generation starts at £4,495 a month as one sterling figure, with the research, the sending infrastructure, the reply handling and the confirmation calls inside it. Whether that beats a $900 licence depends entirely on what your eight hours are worth and how many of them you have. Our outsourced SDR cost guide runs the same arithmetic against an in-house hire.
What about cost per meeting?
That is the number every buyer wants, and it is the one the category will not publish.
No AI SDR vendor I checked publishes a reply rate alongside a sample size and a method. Plenty publish a page headed "benchmarks". None of them say how many campaigns, over what period, in which markets, at what deal size. Without that you are reading a claim.
The human side does have a public benchmark. The Bridge Group's 2025 study puts the median monthly SDR quota at 10 Stage 0 meetings held, down 40% since 2018, with 60% of reps hitting quota, the lowest share in the study's history. There is no equivalent figure for software. If someone shows you a cost per meeting for an AI SDR, ask who produced it, and whether those meetings were held or merely booked.
Are AI SDRs replacing human SDRs?
Not at the rate the funding implies. The best evidence available is one line in the Bridge Group's 2025 report: 2025 was the first year "AI SDRs" appeared as a distinct category in the study, at 1% of 351 respondents.
One per cent. That is the adoption figure, from the longest-running survey in sales development, collected from companies rather than from vendors.
Two more dated data points sit alongside it.
Gartner, in a press release dated 25 June 2025, predicted that over 40% of agentic AI projects will be cancelled by the end of 2027, on escalating costs, unclear business value or inadequate risk controls. Anushree Verma, Senior Director Analyst at Gartner, is quoted in that release: "Most agentic AI projects right now are early stage experiments or proof of concepts that are mostly driven by hype and are often misapplied."
And on 24 March 2025, TechCrunch published an investigation by Dominic-Madori Davis and Marina Temkin into 11x, an AI SDR company backed by a16z and Benchmark. Two named companies went on the record. A ZoomInfo spokesperson told TechCrunch that during a one-month pilot, "11x's product performed significantly worse than our SDR employees". Airtable ran a short trial and decided it was not a fit for the business. A former employee told TechCrunch the company was "losing 70-80% of customers that came through the door". 11x disputed the reporting and said its retention rate was 79%.
I would not build a buying decision on one contested investigation. I would note that it is the only piece of adversarial reporting on an AI SDR vendor's actual results that I could find, and that both customers willing to be named said the same thing.
The pitch lands anyway because the human model is genuinely under strain. The same Bridge Group study reports 40% median annual SDR attrition and an average ramp of 3.0 months. Median SDR on-target earnings sit at $80,000, having grown at 0.56% a year over the past decade while CPI grew 43.9% over fourteen years. A sales director reading that is not being foolish when a $900 licence catches their eye.
Where do AI SDRs break?
The inbox has a hard limit, and volume is the product
Google's bulk sender guidelines tell senders to keep the spam rate reported in Postmaster Tools below 0.10%, and never to reach 0.30%. Cross that line and deliverability degrades while the sequence keeps running, which is the dangerous part. Senders of more than 5,000 messages a day need SPF, DKIM and DMARC, and marketing mail must carry one-click unsubscribe headers.
A tool sold on the promise of more sends, running on domains you set up in an afternoon, is an efficient machine for finding that 0.30%. The complaints arrive before the meetings do.
UK law does not care that a machine sent it
This is where American-written comparisons are no use, and it is the section I would read twice if you are buying from the UK.
The ICO's business-to-business marketing guidance splits recipients into corporate subscribers and individual subscribers. Companies, LLPs, Scottish partnerships and some government bodies are corporate subscribers, and the PECR rule on marketing by electronic mail does not apply to them. Sole traders, most English, Welsh and Northern Irish partnerships and other unincorporated bodies are individual subscribers, and that rule does apply. The ICO is explicit that an employee's email address at a corporate body counts as a corporate subscriber, because the subscriber is the employer. The page also carries a notice that the guidance is under review following the Data (Use and Access) Act, so re-read it before you rely on it.
Two consequences for anyone pointing an autonomous agent at a bought list.
- Your list is almost certainly mixed. Any UK list built from company data at volume contains sole traders and ordinary partnerships. They sit under a different rule, and no AI SDR product I have seen screens for legal form.
- UK GDPR applies either way. The ICO states that if you are processing personal data for direct marketing, even in a business context, UK GDPR applies. You need a lawful basis, you must tell people when you have collected their details in their business capacity, and businesses can object to your marketing and withdraw consent at any time. An agent that keeps sending after an objection is your compliance problem, not the vendor's.
If the product also dials, the bar rises. The same ICO guidance requires screening against both the Corporate Telephone Preference Service and the Telephone Preference Service registers, as well as your own do-not-call list, before live marketing calls. Ask any dialler vendor to show you where in the product that screening happens.
The handover is where meetings die
A booked meeting is not a delivered meeting. The gap between the two is filled by confirmation calls and by somebody who can tell that a buyer has gone cold between Tuesday and Thursday. Software sends the reminder. It does not notice the change in tone.
This is why attendance is the number we quote rather than bookings. ORRJO runs at 90%+ meeting attendance across 12,500+ meetings booked, and that comes from a person on the phone the day before. Our client testimonials are more use to you here than our own figures, because they describe what turned up in the diary.
Almost every comparison you read was written by a vendor
I scraped the UK Google results for "are ai sdrs any good" while researching this piece. Setting the Reddit threads aside, page one was a page by an AI SDR vendor, a page by a company selling a competing AI SDR, two "we tested the tools" round-ups published by outbound software companies, a vendor round-up on an AI agent platform's own blog, and a piece by IBM.
None of that is dishonest on its face. All of it is written by someone with a position. When you are reading a ranked list of AI SDR tools, the first question is who owns the site and the second is which product sits at number one.
When is an AI SDR the right choice?
Four situations where I would buy one without much argument.
- Inbound speed to lead. Somebody filled in a form at 11pm. An agent that replies in ninety seconds with a booking link beats a human who replies at nine the next morning. This is the clearest win in the category, and it has nothing to do with cold outbound.
- Research volume feeding human senders. Use it to summarise 500 accounts so a person can choose 40. The machine reads, the person decides. Cheap, low risk, and it makes the reps you already employ faster.
- Re-engaging people you already have a relationship with. Dormant CRM records and closed-lost opportunities from eighteen months ago. You have a lawful basis, the recipients know who you are, and the message does not have to work hard.
- A founder testing whether an audience answers at all. Before you hire anyone, a $250 tier will tell you whether a segment replies. Treat it as a research spend with an end date written in the diary, and cancel it on that date.
I would not buy one as the primary source of new business meetings for a considered, high-value sale with a long cycle and a small addressable market. There, every account is expensive to lose, and the failure mode of automated outreach is burning names you cannot replace. That is the case for an SDR service run by people, and it is the argument on our page for companies whose AI SDR programme has already stalled.
What should you ask before you sign?
Eight questions. If a vendor cannot answer the first four inside a demo, that is your answer.
- What is one unit of your price, precisely? A contact, an active contact, a credit or a send?
- Show me a raw send log from a live account, redacted. Not a case study.
- What spam rate do your customers run at in Google Postmaster Tools?
- Who owns the sending domains and inboxes, and what happens to their reputation when I leave?
- How do you identify sole traders and ordinary partnerships inside a UK list?
- Where does a suppression request get honoured in the product, and how quickly?
- If you dial, where do CTPS and TPS screening happen?
- What is the notice period, and does it match the quarterly or annual commitment on the pricing page?
Question two is the one that separates the category. A company confident in its output will show you the mail it actually sent.
Frequently asked questions
Is AI replacing SDRs?
Not yet, on the available evidence. The Bridge Group's 2025 study of 351 B2B companies recorded AI SDRs as a distinct category for the first time, at 1% of respondents. Gartner predicted in June 2025 that over 40% of agentic AI projects would be cancelled by the end of 2027. What is being replaced is the manual part of a human SDR's day, mainly list building, account research, first-touch drafting and follow-up scheduling.
Which AI SDR platform is the best?
There is no defensible answer, because no independent party publishes performance data on these products and the vendors do not publish reply rates with a method attached. What you can compare is published pricing, the unit it is measured in, the length of the commitment and the notice period. Every ranked list of "best AI SDR tools" on page one of Google is published by a company with a product in the category.
What does an AI SDR do?
It builds and enriches prospect lists, summarises account research, drafts and sends first-touch email and LinkedIn messages, sorts inbound replies into simple categories, and books meetings into a calendar. It does not qualify buyers in conversation, handle an unstated objection, make a confirmation call, or notice when a buyer has gone quiet for a reason. The Bridge Group puts a median human SDR's day at 112 activities, of which 41 are email. Those 41 are what the software takes.
How much do AI SDRs cost?
Published list prices run from $49 a month for a credit-based plan to $2,500 a month for a high-volume tier, with fully managed layers sold on top at up to another $2,500 a month. Several well-known products publish no price at all. Then add the internal hours the tool still needs. At the ONS 2025 median UK hourly rate of £19.67 plus 15% employer National Insurance, eight hours a week costs about £785 a month on top of the licence.
If you want that arithmetic run against your own market instead of a category average, book a strategy call. We will show you the addressable list size and the meeting range behind it. Those are ranges we plan against, not promises.