AI SDR Outbound Playbook: Signals to Booked Meetings

Outbound for an AI product has a specific problem. Every buyer already receives messages claiming AI will change their team. The message that gets a reply is the one that names something true and specific about the buyer's company, and explains why it matters now. That requires research, and research takes time an SDR does not have unless the process is built for it.

This playbook turns outbound into a weekly cycle of seven steps. It uses Clay for enrichment, the site's open-source GTM skills for research and writing, and an optional n8n workflow to move records around. The rules and requirements it cites were checked on September 26, 2026. It has not been measured on a live campaign, so it contains no reply-rate promises.

The weekly cycle

The days are a starting point. What matters is that research happens in a batch before sending, not one account at a time between calls.

Step 1: Define the target

Start with who buys, not who might. Write down the industries, company size, systems they use and the job title that owns the problem your product solves. Then list what disqualifies an account. A company with no support team is not a buyer for a support agent, however large it is.

The define-icp skill scores accounts against written criteria with evidence for each score. The plan-prospecting skill turns that into a prioritized list. Install the full set with npx skills add zarif3624/gtm-skills and run them inside an agent that supports the Agent Skills format.

Keep the list small enough to research. Research is where the reply comes from, so a list you cannot research is a list you will spam.

Step 2: Collect signals, each with a source

A signal is a public, recent change at the account that makes your product more relevant. For AI products, the useful ones tend to be:

• Hiring. A company posting several support, operations or data roles may be scaling a workflow your product automates. Job postings are public and dated.
• Launches and announcements. A new product line, new market or new regulation that creates work.
• Technical choices. An engineering blog post about building an internal AI tool, or a job posting that names a model provider.
• Leadership changes. A new head of support or data often reviews tools in the first months.
• Public statements. An earnings call or annual report that names cost pressure in a function you serve.

Record every signal the same way, so an AE or a reviewer can check it:

account: Example Co (example.com)
signal: posted 4 support specialist roles
source: https://example.com/careers (job board)
observed_on: 2026-09-21
published_on: unknown
relevance: support volume may be rising faster than headcount
confidence: medium, roles could be backfills

Keep observed_on and published_on separate. A funding announcement you found today may be eight months old. Keep "unknown" when you do not know. An empty field is honest. A guessed one ends up in an email.

Step 3: Research the account

For the top accounts, write a research note that fits on one screen:

1. What the company does, in one sentence, in your words.
2. The two or three signals from Step 2, with links.
3. The workflow your product would touch, and who likely owns it.
4. One open question you would want answered on a first call.
5. What you do not know.

The research-account skill is built for this. Its description is "Research accounts without inventing facts." That is the standard. If a model drafts the note, check every claim against its source before it goes near an email.

Do not research every account to this depth. Do it for the accounts with the strongest signals. Send a lighter, still accurate message to the rest, or wait until a signal appears.

Step 4: Find and verify the people

Pick two or three contacts per account: the person who owns the problem, their manager, and sometimes a technical evaluator. At an AI company that evaluator is often an engineer or data lead. Cohere's BDR posting, opened September 26, 2026, names "CTOs, CIOs, and Business Unit leads" as the people to reach.

Clay's documentation describes a waterfall as a way to "utilize multiple data providers in a predetermined sequence, so you don't duplicate tasks or spend extra credits." In practice, you ask one provider for a verified work email, fall through to the next only if the first returns nothing valid, and stop when one does.

Two rules keep enrichment honest:

• No result means unknown. A contact with no verified email stays unverified. Do not send to a guessed address.
• Keep the source. Record which provider returned the email and when. If it bounces, you know which source to distrust.

Clay's own ClayDR posting describes a Growth team that "sources leads and handles the manual work (using Clay!)" so SDRs can focus on conversations. If your company has a GTM engineer, Steps 2 to 4 are their natural territory. If not, the n8n lead-generation workflow shows how to capture, enrich and deliver records to a CRM yourself, with error handling.

Step 5: Write evidence-first sequences

A sequence is a planned series of touches across email, phone and LinkedIn. Build each message from the research note, in this order:

1. The observation. The signal, stated plainly. "Saw you're hiring four support specialists in Austin."
2. The implication. Why it might matter to this person. "Usually that means ticket volume is growing faster than the team."
3. The relevance. One sentence on what your product does about it, with a claim you can prove.
4. The ask. One small next step.

A simple starting structure:

Space the touches over two to three weeks. Change one thing at a time between batches, so you can tell what worked in Step 7.

The write-outbound skill ("Write relevant, responsible outbound sequences") drafts from a research note and refuses to invent a pain point the note does not support. Use it for a first draft. Read every message before it sends.

Avoid three habits that damage AI outbound in particular. Do not claim a result your company cannot document. Do not pretend a templated email is personal. And do not write "I noticed" about something a tool noticed and you did not check.

Step 6: Send within the rules

Sender rules are not optional, and the big mailbox providers enforce them.

Google's email sender guidelines say senders should keep the spam rate reported in Postmaster Tools below 0.30 percent. Since February 1, 2024, senders of more than 5,000 messages a day to Gmail must authenticate with SPF, DKIM and DMARC and support one-click unsubscribe for marketing messages. Yahoo's sender best practices likewise require bulk senders to support one-click unsubscribe and to "Honor unsubscribes within 2 days."

In the United States, the FTC's CAN-SPAM compliance guide requires accurate header information, a non-deceptive subject line, a valid physical postal address, and a clear way to opt out. Opt-outs must be honored within 10 business days, and each violating email can cost up to $53,088. Other countries set stricter rules. Canada's anti-spam law, for example, is built on consent. Check with your company before emailing outside the US.

In practice:

• Send from a domain with SPF, DKIM and DMARC set up, and warm new domains slowly.
• Remove bounced addresses immediately.
• Honor every opt-out across every tool, not just the one it arrived in.
• Keep daily volume at a level where you can still research and personalize.

Step 7: Review every Friday

Track four numbers per message variant: sent, replied, meetings held, and opportunities the AE accepted. Accepted opportunities are the number that matters. Fireworks' BDR posting measures "new business opportunity and pipeline generation targets," not emails sent.

Each Friday, answer three questions:

1. Which signal type produced the meetings the AE accepted?
2. Which meetings did the AE reject, and why?
3. What one thing changes next week?

Write the answers down. After a month you will know which signals are worth researching and which are noise. That record is also the best evidence you can bring to an AE interview, as the SDR and BDR career guide explains.

Where to go next

• See what happens after your meeting in the AI sales cycle playbook.
• Prepare for the mock call and email in the AI SDR and BDR interview guide.
• Go deeper on the systems side with the GTM engineer career guide.
• Read what is different about selling AI before writing your first sequence.