How We Build an AI Outbound Campaign
On Tuesday I argued that AI outbound isn’t spam at scale unless it’s done badly, and the difference comes down to how it’s built.
This is the build, step by step, as we run it in Month Two of the Pipeline Sprint. It starts from the ideal customer profile and total addressable market defined earlier, because without those, none of what follows works.
Account and Contact Research
Every campaign starts with the market rather than the message. We take the companies identified in the total addressable market and decide who inside each one should hear from us, and why now.
Research Inputs
For each target account, AI-assisted research pulls together everything a well-prepared salesperson would want to know before picking up the phone:
- What the company does, including its services, markets and customers, taken from its own website.
- What’s changed recently, such as new contracts, expansion, new sites, acquisitions or leadership changes.
- Hiring activity, because job postings often reveal priorities and pressure points before anything is announced.
- Public filings, since Companies House shows a UK company’s growth trajectory, size and filing history.
- The likely problem, which is the issue we have the best reason to raise, based on everything above and the ICP.
Then we choose the contacts. In most B2B sales, several people influence the decision, typically an economic buyer, a technical buyer and sometimes an internal champion, so we identify who matters at each account rather than emailing whoever appears first in the database.
The result is a researched, prioritised list, with companies showing signs of need going first.
Personalisation That Isn’t Creepy
Personalisation has a bad reputation, mostly because of how it’s been done. “I saw you went to Loughborough!” or a comment about someone’s holiday photos doesn’t feel personal, it feels like surveillance.
The personalisation that works is about business relevance. The recipient should be able to tell that the email was written for a company like theirs, by someone who understands their situation.
In practice, that means three layers:
- Segment. The core message is written for a specific sector and role, so a message to an operations director at a precision engineering firm reads differently from one to a managing partner at an accountancy practice.
- Account. One or two lines reference something genuinely relevant to that company, such as a recent expansion, a new market or a change that creates the problem we solve.
- Problem. The email leads with the issue the recipient is likely to be facing, in their language, rather than with a description of our services.
AI drafts the account-level lines from the research, and then a person checks that each one is accurate, relevant and something a professional would actually say.
The test is simple. If the recipient read it aloud to a colleague, would it sound thoughtful or intrusive? Only the thoughtful ones go out.
Sequence Design
A single email rarely gets a reply, because a prospect usually needs several encounters before they respond. The sequence is how those encounters are planned.
Sequence Cadence
An illustrative B2B sequence might look like this:
- Day 0: the opening email, a short message that leads with the likely problem and asks one low-commitment question.
- Day 3: a different angle, looking at the same problem from another perspective or sharing a relevant insight.
- Day 8: proof, with a short example of a similar business and what changed for them.
- Day 14: a useful resource, such as a relevant article or guide, with no ask attached.
- Day 21: a direct, polite close, asking clearly whether a conversation is worthwhile and making it easy to say no.
The cadence varies by market and sales cycle, but a few principles stay the same:
- Every email earns its place, so there’s no “just bumping this up”, and each one adds something new.
- Short and plain, written in plain text across a few short paragraphs, as one professional to another.
- One ask at a time, with a single, simple next step.
- Coordinated with other channels, so that where it helps, outbound is timed alongside LinkedIn activity and remarketing and the prospect meets the same message in more than one place.
- Replies stop everything, because when someone responds, they leave the automated sequence and a person takes over.
Deliverability Basics
The best email in the world is worthless if it lands in spam. Deliverability is the unglamorous part of outbound, and it’s the part most often skipped.
Sender Reputation
Mailbox providers decide where your email lands based largely on your sending reputation, so protecting it is non-negotiable:
- A separate sending domain. Outbound runs from a domain related to, but separate from, your main business domain, so your day-to-day email is never put at risk.
- Authentication set up properly. SPF, DKIM and DMARC records are configured correctly, so providers can verify the email genuinely comes from you.
- Warm-up. New sending accounts start at low volume and build up gradually.
- Controlled batches. We work through the market in batches rather than sending to everyone at once, learning from each one before the next.
- Clean lists. Every address is verified before it’s used, because bounces damage your reputation quickly.
- Monitoring. Bounce rates, spam complaints and reply rates are watched closely, and because major mailbox providers now enforce strict limits on complaints, any problems are fixed early rather than after the damage is done.
None of this is visible to the prospect, but all of it decides whether they ever see the email.
Human Checkpoints
AI does much of the research and drafting, but people make the decisions that carry risk. There are four checkpoints in every campaign:
- Target approval, where a person signs off the account list and contacts against the ICP before anything is written.
- Message approval, where the core argument, the sequence and a sample of personalised emails are reviewed and approved before launch.
- Compliance check, where the lawful basis for contacting each business is documented and every email carries a clear way to opt out.
- Reply handling, where every response is handled by a person, so interested replies are passed to sales quickly, objections are answered properly and opt-outs are honoured straight away.
After launch, performance is reviewed weekly to see which segments respond, which messages earn replies and which opportunities are genuinely qualified. That feeds into the optimisation work in Month Three, where every change is judged on qualified pipeline rather than open rates.
Next week, we’ll look at why one channel never delivers ROI on its own, and what the Rule of 7 means for how outbound fits alongside everything else.
Frequently Asked Questions
How is an AI outbound campaign built?
Research the right accounts and contacts, personalise messaging, design the sequence, protect deliverability and keep human review in the loop.
How personalised should cold email be?
Relevant to the recipient’s business and problem, not superficial personal details.
How do I protect email deliverability?
Warm sending domains, send in controlled batches, keep lists clean and monitor bounce and spam rates.
Where does human judgement fit?
In targeting decisions, message approval and handling replies.
Does the Pipeline Sprint include outbound?
Yes, built and launched in Month Two and optimised in Month Three.
Book the £495 AI Pipeline Audit
If your outbound is going out but conversations aren’t coming back, the AI Pipeline Audit will show whether the problem is the targeting, the message or the deliverability. It takes 90 minutes and costs £495, which is deducted in full if you continue to the Pipeline Sprint.
