AI BDR: What It Does, What It Can't, and How to Run One With a Human in the Loop

An AI BDR is software that does the prospecting half of a business development rep’s job: it builds account lists, researches contacts, drafts outbound emails and LinkedIn messages, sends follow-ups and sorts the replies. It does not run discovery calls or carry a quota. The teams that get meetings from one keep a person approving every message before it goes out.

That last point is where we disagree with most of what ranks for this term. When we checked in October 2026, seven of the top nine Google results for this term came from companies that sell AI sales software, and the pitch is autonomy: switch it on and it prospects for you around the clock. We run outbound for B2B SaaS companies with AI doing the research and drafting, and a person reviewing each draft before anything is sent. This guide explains why, what the software is good at, where it breaks, and how to set up the review loop.

What does an AI BDR actually do?

Most of a BDR’s week is research and writing. Finding accounts that fit, working out who to contact, looking for a reason to reach out now, writing the first message, writing the follow-ups, logging everything in the CRM. Salesforce’s sixth State of Sales report found that reps spend 70% of their time on non-selling tasks. An AI BDR takes most of that work. The person keeps the parts that need judgment.

Task	AI BDR handles it	A person still owns it
Building the account list from your ideal customer profile	Yes, from firmographic data and buying signals	Approving the profile and the exclusions
Researching each contact (role, recent hires, funding, product news)	Yes	Spot-checking that the facts are right
Writing the first email and LinkedIn message	Yes, one per contact	Approving claims, tone and the reason for reaching out
Follow-up sequence	Yes	Deciding when to stop
Sorting replies (interested, not now, wrong person, unsubscribe, out of office)	Yes	Answering the interested ones
Booking the meeting	Partly, through a calendar link	Running the discovery call
Handling objections in a live conversation	Poorly	Yes
Owning pipeline and quota	No	Yes

The pattern in that table matters for how you buy. The software is strong on volume work where a mistake is cheap to catch before sending. It is weak wherever a mistake happens live, in front of the buyer, with no chance to review it.

How is an AI BDR different from an AI SDR?

Vendors use the two names for the same products, and some sell one tool under both. The useful split is the one used for human reps, which we cover in our BDR vs SDR guide: an SDR follows up on inbound leads, a BDR starts conversations with accounts that never raised their hand.

For software, the practical test is what triggers the first message. If it fires when someone fills in a demo form or downloads a guide, it is doing SDR work, and speed is what counts. If it starts from a list of target accounts, it is doing BDR work, and relevance is what counts. The two jobs fail in different ways. Slow inbound follow-up loses leads you already paid for. Careless outbound burns your sending domain and your reputation with accounts you have not met yet.

What can’t an AI BDR do yet?

Five gaps show up in practice. None of them is fixed by a better model. They are fixed by how you run the tool.

It does not know your unwritten rules. Which accounts are in an open deal, which are a partner’s customers, which churned last quarter and should not hear from sales right now. Unless the tool reads your CRM and those fields are kept up to date, it will email them.

It does not check its own facts. Research agents pull from news, LinkedIn and company sites, and they get things wrong: a funding round attributed to the wrong company, a job title from two roles ago, a product launch that was a rumor. A wrong fact in the opening line of a cold email is worse than no personalization at all, because it tells the reader nobody looked.

It cannot hold a sales conversation. It can reply to “send me more info” or offer times to meet. It cannot run discovery, read a hesitant buyer, or decide that this account is worth a second meeting with a different stakeholder.

It will send as much as you let it. Software has no instinct to protect your domain. Google’s rules for anyone sending 5,000 or more messages a day to Gmail accounts require SPF, DKIM and DMARC, one-click unsubscribe, and a spam rate kept below 0.3%, with 0.1% as the target. A bad batch can push a domain past that line in days, and every sales email from that domain suffers, including the ones your human reps write.

It cannot carry the legal responsibility. The FTC’s CAN-SPAM guide says the law makes no exception for business-to-business email, puts the penalty at up to $53,088 for each violating email, and tells companies to monitor what others send on their behalf. Handing outbound to a vendor or an agent does not hand off the liability.

How do you run an AI BDR with a human in the loop?

Run it on a weekly rhythm with one review gate before anything is sent. This is the loop we use for client outbound, and it works the same way with an in-house tool.

1. Monday: score new accounts. The AI scores new accounts against your closed-won customers: same size, same tech stack, same buying trigger. A person approves the list and removes anything in an open deal or on the do-not-contact list.
2. Tuesday: research and draft. For each contact, the AI finds a reason to talk now (a new head of sales, a funding round, a job post that names the problem you solve) and drafts a first message that opens with it. Every draft carries the link to its source.
3. Wednesday: review, then launch. A person reads each draft next to its source link and approves, edits or kills it. Only approved drafts enter the sequence.
4. Thursday: route replies. The AI sorts replies. Interested ones go to a seller the same day, with the thread and the research attached, so the seller does not start from zero.
5. Friday: tune. Look at which triggers and which openings got replies, and change next week’s list and prompts. Drop the angles that got nothing.

The review on Wednesday is the step vendors call optional. Do not skip it. It is where wrong facts, risky claims and accounts that should not be contacted get caught, and it is the only point in the loop where a mistake costs nothing.

What should the reviewer check in each draft?

• The reason for reaching out is real, current and matches the source link.
• Name, title and company are right.
• The email makes one claim, and you could prove it on a call.
• No fake familiarity. “Loved your recent post” when nobody read the post is the line buyers spot first.
• The account is not a customer, an open opportunity or a partner’s client.
• The unsubscribe link and your postal address are in the template.

What a review catches: an illustrative example

Illustrative, not a benchmark. The company is fictional.

Northwind Payroll sells payroll software to companies with 200 to 1,000 employees. Its AI BDR found that a target account had just posted a job for a payroll manager and drafted this opener:

“Congrats on the Series C! As you scale past 1,000 employees, payroll gets harder…”

The reviewer opened the source link. The Series C belonged to a different company with a similar name, and the account had about 400 employees. The approved version:

“Saw you’re hiring a payroll manager. When teams around 400 people add that role, it’s usually because the current setup needs a person to babysit it. Is that what’s happening on your side?”

The research and the trigger stayed the same. The reviewer removed one wrong fact and added a question. The first version would have told the buyer that nobody checked. The second gives them a reason to answer.

What does an AI BDR cost compared with a human BDR?

Start with the human benchmark. The Bridge Group’s 2025 report on sales development, based on 351 B2B companies, puts median on-target earnings at $80,000, split about $55,000 base and $25,000 variable. Average ramp is 3.0 months and average tenure is 1.9 years, and only 60% of reps hit quota, the lowest share in the study’s history. On top of salary come management time, tools and the months of pipeline lost every time a rep leaves and a new one ramps.

AI BDR pricing is harder to compare because most vendors quote annual contracts on a sales call rather than publishing prices, and they charge in different units: per seat, per contact, per email or per meeting. Put every option on the same footing before you compare.

Cost line	Human BDR	AI BDR, fully autonomous	AI BDR with human review
Pay or license	Salary plus variable	Software contract	Software contract
Data and sending tools	Yes	Usually extra	Usually extra
Ramp before useful output	About 3 months (Bridge Group)	Setup weeks	Setup weeks
Reviewer time	Manager coaching	None, which is the risk	A weekly review session
Risk to your sending domain	Low, volume is limited by hours	Highest	Low, a person gates volume
Discovery calls	Yes	No	No, a seller takes them

The column that usually wins is the third one. It keeps most of the software’s speed and puts a person back where mistakes are expensive. If you do not have someone with the time and judgment to review drafts every week, that is the role a service can fill. Our ABM and outbound service runs this exact loop: AI researches accounts and drafts each message every week, our team reviews the drafts before anything is sent, and interested replies reach your sellers the same day.

Will AI replace BDRs?

AI is taking over the research and drafting part of the role, which is where most of the hours go. Gartner predicted in 2023 that by 2028, 60% of B2B seller work will be executed through conversational interfaces powered by generative AI, up from less than 5% in 2023. It is a forecast from 2023, so treat the number loosely. The direction is the useful part: fewer hours on lists and first drafts, more on review and conversations.

The conversation part stays human. Discovery calls, multi-threading into a buying committee and the call on which accounts deserve a second try all need someone who carries the outcome. The BDRs who do well over the next few years will be the ones who can review fifty drafts quickly, catch the wrong fact in the opening line, and turn a reply into a meeting. If you are hiring, our guide to what a BDR does covers the role as it stands today.

How do you choose an AI BDR tool or service?

Ask every vendor the same questions, and ask to see the answer in the product rather than on a slide.

• Can every message be held for approval? If approval is a paid add-on or a setting that resets, treat it as missing.
• Does each draft show its source? A reviewer cannot check a fact they cannot see.
• Does it read your CRM before it writes? It should skip customers, open opportunities and anyone who unsubscribed, without you exporting lists by hand.
• What happens to an interested reply? It should reach a named seller within hours, with the thread and the research.
• How does it protect your domain? Ask about per-mailbox sending limits, warm-up, one-click unsubscribe and spam-rate monitoring.
• How is it priced? Per seat, per contact, per email or per meeting. Model the cost at the volume you plan to send in month three, not the trial volume.

If the honest answer to the first two questions is no, the tool was built for sending at volume and review was an afterthought. That can work for very high volume at low stakes. For B2B SaaS deals where one account can be worth a year of a rep’s salary, it is the wrong trade.