On this page
Can an AI agent do outbound prospecting?
Yes, with a person watching it. An AI SDR (an AI sales development representative) can do the repetitive half of outbound prospecting today: pick accounts that show buying signals, find the right contacts, write and send a short sequence of emails, answer simple replies and book meetings with your reps. Salesforce reports that its own Engagement Agent sends outbound introduction emails, answers product questions, books meetings and handles opt-out requests, and has reached 100,000 leads. HubSpot's product page quotes Jotform's Head of Lead Generation: its prospecting agent "tested 1,000 contacts and qualified 26 deals." Both are vendor-published figures.
What the agent should not own is the list, the law and your sender reputation. It writes to strangers under your name, so a wrong fact, an email to someone who opted out or a spike in spam complaints lands on you. Treat it as a supervised hire: the agent drafts and sends inside limits you set, and a person decides who gets emailed, reviews messages until they are reliably right and watches deliverability.
This page covers outbound email prospecting. Researching and enriching accounts before outreach is lead research and enrichment; keeping CRM records clean is CRM data hygiene; the role as a whole is the sales development representative page.
What the agent does and what stays with you
Agent
- Pick target companies from buying signals
- Find contacts at those companies
Human
- Decide who may legally and sensibly be emailed
Agent with review
- Write the first email and follow-ups
Agent
- Send on a schedule within volume limits
Agent with review
- Answer replies and product questions
Agent
- Book a meeting with a rep
Agent with review
- Process opt-outs and keep the suppression list
Human
- Watch sender reputation and deliverability
- Run the first call and hand qualified deals to an account executive
10 tasks: 4 agent, 3 agent with review, 3 human
| Task | Who does it | Why |
|---|---|---|
| Pick target companies from buying signals | Agent | HubSpot documents its prospecting agent monitoring companies in chosen segments for intent signals such as funding news and recommending which to enroll; a person can check each recommendation's stated reason. |
| Find contacts at those companies | Agent | HubSpot's agent suggests contacts by persona from the CRM or a connected data provider (Apollo, Surfe, ZoomInfo or Seamless), up to a daily limit you set. |
| Decide who may legally and sensibly be emailed | Human | Consent and opt-out rules differ by country and recipient type (the ICO treats sole traders differently from companies), and no source checked shows an agent making that call. |
| Write the first email and follow-ups | Agent with review | HubSpot's agent drafts personalized emails from CRM data and your selling context; TechCrunch reports 11x customers complaining of hallucinations, so a person reviews drafts until samples are clean. |
| Send on a schedule within volume limits | Agent | HubSpot lets you cap the emails per contact, the minimum days between them, business days and a daily send window, and the agent sends within those settings. |
| Answer replies and product questions | Agent with review | Salesforce says its agent answers product questions from pricing pages, specifications and messaging guidelines; replies go to prospects, so a person samples them and takes over hard questions. |
| Book a meeting with a rep | Agent | Salesforce describes its agent sending a scheduling link and the meeting landing on the SDR's calendar, with the seller copied on the confirmation. |
| Process opt-outs and keep the suppression list | Agent with review | Salesforce says its agent handles opt-out requests; the FTC requires opt-outs to be honored within 10 business days, so a person checks the suppression list weekly. |
| Watch sender reputation and deliverability | Human | Google's guidelines ask senders to keep the Postmaster Tools spam rate below 0.3% and to cut volume when bounces or deferrals rise; no source checked shows an agent managing this. |
| Run the first call and hand qualified deals to an account executive | Human | Salesforce describes its agent handing a lead to an SDR once the lead is ready to meet, with the meeting on the SDR's calendar; the conversation from there is the rep's. |
The split follows what a reviewer can check. Choosing accounts from signals, finding contacts and sending on a schedule are configurable and auditable: HubSpot documents a daily cap on new contact suggestions, a maximum number of emails per contact, a minimum gap in days and a send window. Writing and replying are where errors reach a prospect, so a person reviews them first. HubSpot builds this in: its Automation tab offers "Review emails," where you approve each message, or "Send without review."
Judgment stays human. Whether a contact may be emailed at all depends on where they are and what kind of recipient they are, and protecting your domain's reputation means slowing down when the numbers turn, which a sending agent has no reason to do on its own.
A worked example: one lead
This is an illustration, not a customer story. Assume, for this example, that you sell scheduling software to dental practices in the U.S., you have approved three claims the agent may make, and a rep reviews every email during the first month.
The agent sees that a 12-location dental group posted jobs for two practice managers, which matches a hiring signal you selected. It finds the operations director in your contact data source, checks that the address is not on your suppression list and drafts a 90-word email: it mentions the new locations, makes one of your approved claims and offers a meeting link. The email carries your postal address and an opt-out line.
The rep's review catches one problem: the draft calls the group "family-owned," which nothing in the CRM supports. The rep deletes the phrase, approves the email and adds a line to the agent's instructions: no claims about ownership. Six days later the director replies, "Not now, try us in Q3." Under the rules in this example, the rep unenrolls the contact and sets a reminder for July. Nobody is emailed again until a person decides to.
What you need to provide
The agent touches personal data about prospects (names, work email addresses, job titles, company news) and needs write access to your CRM, a sending mailbox and, if it finds new contacts, a data provider account. Give it:
- Your ideal customer profile and the job titles you sell to.
- Selling context: what you sell, the problems it solves and a written list of claims it may make. HubSpot's setup asks for exactly this (what you sell, your value proposition, the pain points you solve).
- A sending mailbox on an authenticated domain. Google requires every sender to Gmail accounts to set up SPF or DKIM and keep the Postmaster Tools spam rate below 0.3%. Senders of more than 5,000 messages a day to Gmail accounts must set up SPF and DKIM, DMARC, and one-click unsubscribe for marketing and subscribed messages.
- Your suppression list and a rule for who is off-limits (customers, open deals, competitors, past opt-outs).
Commercial email rules vary by jurisdiction. In the U.S., the FTC says CAN-SPAM "makes no exception for business-to-business email": each commercial message needs accurate headers, a non-deceptive subject line, a clear disclosure that it is an ad, your physical postal address and an opt-out you honor within 10 business days. The FTC also says you cannot contract away that responsibility by hiring another company to send for you. In the UK, the ICO's PECR guidance says you can send marketing email to corporate bodies but need specific consent (or the soft opt-in for your own customers) to email individuals, sole traders and some partnerships, and that the soft opt-in does not cover bought-in lists. The ICO notes this guidance is under review after the Data (Use and Access) Act. Ask counsel which obligations apply to your role and jurisdiction.
Job brief template
Use this as the brief for a vendor, an operator or an internal team setting up the agent. Replace the bracketed values with your own.
Copy-ready brief
- Goal
- Every account in the agreed target segment that shows a buying signal gets a relevant, accurate first email and follow-ups within 5 business days, and every interested reply turns into a booked meeting with a named rep.
- Deliverable
- Outreach sent from your mailbox under a real sender name, every email and reply logged on the CRM record, booked meetings on the rep's calendar with context notes, and a weekly report of sends, bounces, spam complaints, replies, opt-outs and meetings.
- Inputs
- Your ideal customer profile: industries, company sizes, countries and the job titles you sell to
- Selling context: what you sell, the problems it solves, approved claims, case studies you may cite and claims the agent must never make
- Read and write access to the CRM for leads, contacts and activity logging
- A connected sending mailbox on an authenticated domain (SPF, DKIM and DMARC set up) and access to its reputation data
- A contact data source, if the agent should find new contacts
- Your suppression list: past opt-outs, customers, competitors, open deals and anyone a rep is already working
- Meeting links for the reps who take booked calls
- Acceptance criteria
- On a weekly sample of 50 sent emails, none states a fact about the prospect or your product that is false or not in the approved selling context.
- Zero emails go to an address on the suppression list; every opt-out is on the list within 2 business days.
- Spam rate in Google Postmaster Tools stays below 0.3% and the hard-bounce rate stays below a threshold you set before the trial.
- Every email includes your postal address and a working opt-out, and every interested reply reaches a rep or a booked meeting within 1 business day.
- No email offers a price, discount or contract term that is not in the approved selling context.
- Systems
- CRM (for example HubSpot or Salesforce)
- sending mailbox on your own domain (for example Google Workspace or Microsoft 365)
- contact data provider (for example Apollo or ZoomInfo)
- calendar scheduling link
- Constraints
- The agent processes personal data about prospects (names, work emails, job titles, company signals); limit it to work contact data.
- Commercial email rules apply where you and your recipients are; for example CAN-SPAM in the U.S. and PECR in the UK. Ask counsel which apply.
- The agent may not buy contact lists, change prices, promise terms or email anyone on the suppression list.
- Sending stays within volume caps that rise only after the reputation numbers hold.
- Volume
- Example: 200–2,000 new contacts enrolled a month, 3–5 emails each
- Cadence
- Example: daily sending in business hours, weekly sample review
- Human checkpoints
- Before launch: a sales lead approves the target segment, the contact sources and the suppression list.
- Until the sample is clean for two weeks: a rep approves every email before it sends.
- Weekly: a sales lead reads 50 sent emails and every negative reply, and checks spam rate, bounces and the opt-out list.
- Every booked meeting: the rep confirms fit before the call and owns the conversation from there.
The two criteria to keep, whatever else you change, are the 50-email weekly sample and the spam rate. One shows whether the agent tells the truth; the other shows whether anyone will receive your email next quarter.
What it costs
Human benchmark
Sales representatives, wholesale and manufacturing, except technical and scientific products
$72,080
per year (median annual wage)
BLS May 2025 national median for SOC 41-4012, as published in the Occupational Outlook Handbook, which lists inside sales representatives who cold-call among this occupation; wage only, no commission, benefits or overhead
Source: Wholesale and Manufacturing Sales Representatives: Occupational Outlook Handbook
Agent price range
$1
per lead the agent writes outreach for
HubSpot prospecting agent: 100 HubSpot Credits per lead × $0.010 per additional credit = $1.00. Requires a Starter, Professional or Enterprise subscription (Core Seat from $20 per seat a month on Starter); the catalog's included credits for Smart CRM and Marketing, Sales, Service, Content or Revenue Hub are 500, 3,000 or 5,000 a month by edition (Data Hub and Customer Platform differ); they are one shared monthly pool and cover up to 5, 30 or 50 prospecting leads only if nothing else spends them. Flexible seats-and-credits pricing uses different credit rates and has no Core Seats, so $1.00 and $20 are the standard catalog rates. Contact data from third-party providers and seats are excluded.
Sources: HubSpot Product & Services Catalog, AI Prospecting Agent for Sales Teams
The two figures use different units. Compare them at your own volume.
The human figure is the May 2025 national median annual wage for wholesale and manufacturing sales representatives outside technical and scientific products (SOC 41-4012), $72,080, from the BLS Occupational Outlook Handbook. BLS has no separate SDR category; it lists inside sales representatives who cold-call as one type within this occupation. It is wage only, without commission or benefits.
The agent figure is HubSpot's published rate: its prospecting agent uses 100 HubSpot Credits each time it recommends outreach for a lead, and additional credits cost $0.010 each, so $1.00 per lead. It needs a HubSpot Starter, Professional or Enterprise subscription; Starter Core Seats are $20 per seat a month, and the catalog's included credits for Smart CRM and Marketing, Sales, Service, Content or Revenue Hub (500, 3,000 or 5,000 a month by edition; Data Hub and Customer Platform differ) are one shared monthly pool that covers up to 5, 30 or 50 prospecting leads only if nothing else spends them. Customers on HubSpot's flexible seats-and-credits pricing have different credit rates and no Core Seats, so $1.00 and $20 are the standard catalog rates. Salesforce prices its agents in Flex Credits at $500 per 100,000 credits, with 20 credits per agent action, which works out to $0.10 per action ($500 ÷ 100,000 × 20). Salesforce does not say on the pages read how many actions one lead takes, so its price per lead is unknown. Prices for standalone AI SDR products were not published on the vendor pages checked.
Neither number includes contact data, the reviewer's time or a damaged domain. Measure those in your trial.
Risks and human checkpoints
The agent states something false about the prospect's company or about your product.
How you notice it
Review every draft at first, then sample 50 sent emails a week against the CRM record and the approved selling context; TechCrunch reports 11x customers citing hallucinations.
Spam complaints and bounces damage your domain's reputation, and later email, including to customers, lands in spam.
How you notice it
Watch the spam rate in Google Postmaster Tools daily; Google asks senders to stay below 0.3% and to reduce volume when bounces or deferrals rise. Send prospecting from a mailbox you can afford to slow down.
Email reaches someone you may not contact, or keeps reaching someone who opted out.
How you notice it
Screen every enrollment against the suppression list and audit it weekly; the FTC sets a 10-business-day deadline to honor opt-outs and the ICO says the soft opt-in does not cover bought-in lists.
Volume goes up while results do not, and you pay for leads that never reply.
How you notice it
Track replies and meetings per 100 leads enrolled from the first week, and set a credit or usage limit; TechCrunch quotes a company that tried 11x calling the ratio of automated emails to meetings disappointing.
Vendor results do not hold for your market.
How you notice it
Treat case studies as vendor claims and judge the agent only on your own trial numbers; TechCrunch found companies listed as 11x customers that said they had only run short trials.
Human checkpoints
- Before launch: a sales lead approves the target segment, the contact sources and the suppression list.
- Until the sample is clean for two weeks: a rep approves every email before it sends.
- Weekly: a sales lead reads 50 sent emails and every negative reply, and checks spam rate, bounces and the opt-out list.
- Every booked meeting: the rep confirms fit before the call and owns the conversation from there.
One public record of what goes wrong is TechCrunch's March 2025 investigation of 11x, which sells an AI SDR. Three current and former employees told TechCrunch that most early customers used break clauses to leave, and sources cited email that did not work as expected and hallucinations. ZoomInfo said that during a one-month pilot, "11x's product performed significantly worse than our SDR employees." 11x told TechCrunch that performance "ultimately relies on the quality of user input" and that its retention rate was 79%. TechCrunch also reported that Pleo and Rho confirmed they are using 11x products. The report is from 2025; it stays here because it was the only independent, reported account of an AI SDR in customer hands found for this page.
Our recommendation: write down who may be emailed and what may be claimed before the agent sends anything, and give one person the job of reading samples and the reputation numbers every week.
How to run a trial
Run the agent on one segment for six weeks before widening it.
- Pick one segment and one offer, with 200–500 contacts. Keep your existing outbound running elsewhere so you have a comparison.
- Set up authentication (SPF, DKIM, DMARC) on the sending domain and check it in Google Postmaster Tools before the first send.
- Load the suppression list and the approved claims. Turn on review mode: a rep approves every email.
- Start with low volume and raise it slowly, as Google advises, reduce volume if bounces or deferrals rise, then raise it slowly again.
- Each week, read 50 sent emails and every negative reply against the brief's criteria, and record spam rate, bounces, replies, opt-outs and meetings per 100 contacts.
- After two clean weeks in a row, allow sending without review for that segment only, and keep sampling.
- At the end, compare meetings and cost per meeting with your existing outbound, using your numbers, not the vendor's.
For structuring the test, see how to run a paid trial with an AI agent; for writing the brief, see how to write a job description for an AI agent. For the other sales jobs, see the sales overview.
Sources
- Set up and use the prospecting agent
HubSpotPrimaryAccessed Nov 14, 2026
- AI Prospecting Agent for Sales Teams
HubSpotVendorAccessed Nov 14, 2026
- HubSpot Product & Services Catalog
HubSpotVendorAccessed Nov 14, 2026
- Salesforce's sellers close more deals in less time with Agentforce.
SalesforceVendorAccessed Nov 14, 2026
- Agentforce Pricing
SalesforceVendorAccessed Nov 14, 2026
- a16z- and Benchmark-backed 11x has been claiming customers it doesn't have
TechCrunchPressAccessed Nov 14, 2026
- Email sender guidelines
Google (Gmail Help)PrimaryAccessed Nov 14, 2026
- CAN-SPAM Act: A Compliance Guide for Business
U.S. Federal Trade CommissionPrimaryAccessed Nov 14, 2026
- Electronic mail marketing (Guide to PECR)
Information Commissioner's Office (UK)PrimaryAccessed Nov 14, 2026
- Wholesale and Manufacturing Sales Representatives: Occupational Outlook Handbook
U.S. Bureau of Labor StatisticsPrimaryAccessed Nov 14, 2026
What we could not verify
- Independent measurements of reply rates, meetings booked or pipeline from AI SDRs. Every outcome figure on this page (Salesforce's 100,000 leads and Jotform's 26 qualified deals from 1,000 contacts) is published by the vendor.
- The text of the EU GDPR on EUR-Lex blocked automated access, so the page does not describe GDPR obligations for B2B email; only the UK ICO's PECR guidance and the U.S. FTC's CAN-SPAM guide were read.
- Salesforce's help pages on its SDR agent (setup, sending limits) did not render for automated reading; Salesforce's agent is described only from its own customer story and its pricing page.
- How many Flex Credit actions Salesforce's agent uses per lead or per email is not published on the pages read, so a per-lead price for Salesforce cannot be derived and it is left out of the cost range.
- Prices for 11x, Artisan and similar standalone AI SDR products were not found on vendor-owned pricing pages; figures quoted in third-party round-ups were not used.
- Error rates (wrong facts in emails, emails to the wrong person) for any AI SDR. The risk is shown through TechCrunch's reporting on 11x, not a measured rate.
Related
- AI agents for sales: which jobs to hand over first
- Can AI replace SDRs? The sales development job, task by task
- AI agent for lead research and enrichment
- AI agent for CRM data hygiene: clean and dedupe records
- AI agent job description: how to write one, with template
- How to test an AI agent before buying: run a paid trial
