AI SDR: what an AI sales agent can and can't do

An AI SDR is software that does parts of a sales development rep's job: researching accounts, drafting personalized outreach, running follow-up sequences, answering simple replies and booking meetings. It takes over the repetitive research and admin. It can't reliably judge fit, handle a nuanced objection or protect your sender reputation on its own, so give it a reviewer and keep a human on the conversation.

#What an AI SDR does, step by step

Most AI SDR products and custom builds follow the same chain. The names differ. The steps rarely do.

1. Define the target. Your ideal customer profile: industry, company size, geography, tech stack, and trigger events such as a funding round or a new VP hire.
2. Build the list. Pull accounts and contacts from your CRM and data providers, then remove anyone who should not be contacted.
3. Research. For each account, gather what the company sells, recent news, open job postings and the contact's role.
4. Personalize. Draft an opening line and a short message tied to one real fact you can verify.
5. Sequence. Send over email, LinkedIn or both, with follow-ups that stop the moment the prospect replies.
6. Handle replies. Sort them into interested, not now, wrong person, unsubscribe and out of office.
7. Hand off. When a prospect shows interest, create a task or a meeting for a human and pass the context along.

The last step matters most. Everything before it exists to put a qualified conversation in front of someone who can have it.

#What an AI SDR does well

• Reading the website, job board and news for 500 accounts. That is tedious for a person and cheap for a model.
• Writing a short first draft that mentions a real detail, in your tone, based on examples you approved.
• Sending the third follow-up that a busy rep forgets.
• Sorting replies and acting on unsubscribes immediately.
• Logging every touch in the CRM, updating fields and creating tasks.
• Offering times from a rep's calendar and confirming the slot.

#What it can't do

A model sees only the signals you give it. It doesn't know that a prospect's last vendor relationship ended badly, or that your best customers never come from that segment.

Objections are the next gap. Pricing pushback, politics inside a buying committee and "we're locked into a contract until March" need a person who can read the room. So does a hesitant buyer on a call. No sequence replicates an experienced rep there.

Then there is plain error. Models state outdated facts with confidence and misread job titles. A message that congratulates someone on a role they left last year costs you the contact for good.

And the sending itself: the agent will send as much as you let it. Let it send badly and your domain reputation drops, taking your regular company email down with it.

Key point: An AI SDR multiplies whatever your outbound already is. Good targeting and honest messaging get more volume. Poor targeting and generic copy get more spam complaints, faster.

#Human review: where to keep people in the loop

In the early weeks we usually have a person approve every message before it sends. As a segment proves itself, review drops to sampling. A new segment or a new message pattern goes back to full review.

#Deliverability and compliance

Outbound email lives or dies on your domain's reputation, and AI makes it easy to wreck.

Deliverability basics:

• Send from a dedicated domain or subdomain, separate from your main company domain.
• Set up SPF, DKIM and DMARC correctly, then monitor them.
• Warm new mailboxes gradually and cap daily volume per mailbox.
• Verify addresses before sending so bounce rates stay low.
• Include a working unsubscribe and honor it automatically.
• Watch spam complaint rates in your inbox provider's postmaster tools. Google and Yahoo have published sender requirements for bulk senders, and they change, so read the current ones.

Compliance, in general terms (not legal advice):

• In the US, the CAN-SPAM Act sets rules for commercial email, including a physical address and an opt-out. Calls and texts fall under the TCPA, which is stricter.
• In the UK and EU, rules depend on who you email. The UK's Privacy and Electronic Communications Regulations (PECR) treat emails to individuals more strictly than emails to corporate subscribers, and GDPR governs how you obtain and use contact data. "Legitimate interest" has conditions, and you need to document them.
• LinkedIn's terms restrict automation. Check what your tool does and decide what risk your team accepts.
• Keep a record of where each contact came from, and one suppression list that every tool checks.

#Packaged AI SDR tool or custom build

Ready-made platforms bundle data, sending and AI in one product. You can send your first email within days. The trade-offs show up around month three, when your sales motion stops fitting the product's assumptions.

Ask any vendor to show you the messages it actually sent in a pilot, the bounce and complaint rates, and what happened to the replies. Then ask how you stop it, and how fast.

For how we build this, see sales automation. That work covers CRM updates, routing, follow-up and reporting around the outreach itself. If your real problem is finding the right accounts, our guide to AI lead generation covers enrichment and scoring.

#A realistic pilot

1. Choose one segment. One industry, one role, one clear pain.
2. Write the rules. Who is excluded, how often you make contact, what the message may and may not claim.
3. Seed it with your best message. Give it three or four emails that worked, written by a person.
4. Review every message in the first batch of a few hundred sends.
5. Measure from reply to meeting, not opens. Open tracking is unreliable, partly because email clients preload images.
6. Keep a human control group. A rep handles a matching list, so you know whether the agent helps.
7. Decide on real numbers. Keep it, change it or stop.

Say a pilot sends 300 emails and gets 12 replies. Read all 12 before drawing conclusions. If 8 of them are unsubscribes or "wrong person", fix the list first.

#When an AI SDR is a poor fit

Skip it if your sales cycle depends on a few high-value accounts. Each message there should come from someone who knows the buyer. Skip it if your CRM data is unreliable, because the agent will personalize against wrong facts. And skip it until you've found a message that works for humans. Automating a message that doesn't work gets you more of the same results.

If you're weighing this for your team, book a strategy call. We'll review your ICP, data and CRM setup and tell you whether an AI SDR helps or something else needs fixing first.

#Frequently asked questions

#What is an AI SDR?

An AI SDR is a system that automates parts of the sales development workflow: building lists, researching accounts, writing outreach, running follow-ups, classifying replies and scheduling meetings. It usually works alongside human reps, who take over once a prospect shows interest.

#Can an AI sales agent replace human SDRs?

It can take over research, drafting and follow-up admin. It can't replace judging fit, handling objections or building trust. Most teams do better using AI to give each rep more qualified conversations than trying to remove reps.

#Is AI cold outreach legal?

It depends on the channel, the country and who you are contacting. Email rules (CAN-SPAM in the US, PECR and GDPR in the UK and EU) require clear identification, an opt-out and a lawful basis for processing data. Automated calls and texts are restricted more tightly. Get advice for your markets.

#Will AI outreach hurt my email deliverability?

It can. High volume, weak targeting, unverified lists and missing unsubscribe handling all raise spam complaints. Use a separate sending domain, authenticate it, cap volume, verify addresses and watch complaint rates from day one.

#How do I measure an AI SDR pilot?

Count qualified meetings and the pipeline they create, against a human-run control group. Also track reply quality, unsubscribe and complaint rates, and the rep time spent on review. Meetings that never convert are a cost, so follow them through to outcomes.

#AI SDR vs sales automation software: which one do I need?

Sales automation software runs rules you configure, such as a sequence of templates on a timer or a task when a deal changes stage. An AI SDR reads the account, decides what to say and drafts it. If your problem is consistent follow-up on a known list, a sales outreach tool or CRM workflow may be enough, and it is cheaper. Add an AI SDR when research and personalization are the bottleneck. Our guide to sales automation software shows what a CRM does natively.