The ICO's Code of Practice on AI is now a statutory instrument. From May 2026, any B2B operation using AI for recruitment, credit assessment, or customer profiling has to align with it or accept regulatory exposure. This is no longer guidance you can skim and file. It's law, and it applies to the tools sitting in your outbound stack right now.
We've spent the last few weeks going through our own product against the Code, because we build AI that scores inboxes and drafts emails. Here's what actually changed, what it means for a two-person sales team running their own pipeline, and the compliance work worth doing this quarter.
Guidance was optional. Legislation is not.
The distinction matters more than it sounds. A code of practice published by a regulator is a strong signal of what "good" looks like. A code enshrined as a statutory instrument is something a court or regulator can hold you to directly. The ICO can point at it. A complainant can point at it. Your lawyer will point at it the moment anything goes sideways.
For most founders reading this, the practical shift is this: the question is no longer "are we broadly GDPR-compliant?" It's "can we show, line by line, how our AI tools handle personal data, and who is accountable when they get it wrong?"
That's a documentation problem as much as a technical one. And it's the kind of problem that gets ignored until a prospect's procurement team asks for it in a security review.
Three uses of AI that now sit squarely in scope
The Code's scope is narrower than the headlines suggest. It targets automated decision-making and AI systems that materially affect individuals. In a B2B outbound context, three patterns land in scope:
- Lead scoring that gates outreach. If your AI decides which prospects get contacted and which get dropped, that's a decision about a person, even if the person is a business contact.
- Profiling for personalisation. Inferring seniority, budget authority, or buying intent from public data and using it to shape the message is profiling. The Code wants you to be able to explain how that inference was made.
- Recruitment and credit-adjacent workflows. If you sell into HR or fintech and your product touches hiring or credit decisions, you're in the deepest part of the Code. Read it properly.
What's not in scope: a human reading a prospect's LinkedIn and writing a bespoke email. The Code is about automated systems, not about people doing their job. The line is whether a machine is making or materially shaping the decision.
The compliance work is mostly unglamorous documentation
Here's the honest version. Nobody is going to shut you down in May for running a lead-scoring model. The realistic risk is a procurement blocker, a complaint, or a regulator asking questions you can't answer cleanly. The work to prevent that is boring and takes a week, not a quarter.
What we did, and what we'd suggest any team running AI in outbound do:
1. Inventory every AI touchpoint in the pipeline. Not just the obvious ones. The enrichment tool that appends firmographic data. The scoring model that ranks your list. The drafting assistant that writes the first line. The meeting-brief generator. Write them down with the vendor name and what data each one touches.
2. Map each one to a lawful basis. Most B2B outbound sits on legitimate interests. That's fine, but the Code expects you to have done the balancing test and to be able to produce it. If you can't explain why your legitimate interest outweighs the prospect's rights, you have a problem the Code will surface.
3. Document human oversight. This is the part most teams skip. The Code cares about whether a human is meaningfully in the loop, not whether a human clicked a button. If your AI drafts and a rep sends without reading, that's not oversight. If your AI scores and a rep reviews the top 50 before sending, that is. Write down which one you do.
4. Get the vendor paperwork in order. Every AI tool processing personal data on your behalf needs a data processing agreement that reflects the Code's expectations. Most standard SaaS DPAs are fine. Some aren't. Read them.
That's it. Four steps. A week of work if you're organised, two if you're not. The teams that skip it will be the ones scrambling when a prospect's legal team sends a 40-question vendor assessment in September.
What this means for how you actually run outbound
The Code doesn't change what good outbound looks like. It changes what you have to be able to prove about it.
We've written before about why AI scrutiny, not fines, is the real threat to cold email. That piece holds up. The Code just makes the scrutiny formal. A prospect's compliance team now has a document they can hold you to, and they will.
The teams that come out of this well are the ones who can say, in plain English: here's every AI system in our pipeline, here's what it does, here's the human check on it, here's the paperwork. That's a sales asset as much as a compliance one. If you can answer those four questions in a security review, you win deals your competitors lose to procurement delays.
If you're running a small team and this feels like a lot, it is. But it's a bounded amount of work, and the alternative is finding out in the middle of a deal cycle that you can't answer a basic question about your own stack.
What we'd do next
If you use AI anywhere in your outbound, block two hours this week to inventory every tool that touches prospect data. That's the whole first step, and it's the one that makes the rest tractable. If you want to see how MiraReach handles human-in-the-loop drafting and scoring, give MiraReach a try. Nothing sends without a human pressing the button, and every AI step is documented.
— Mira