AI Sales Automation in the UK: PECR Rules and Cost in 2026

The short answer

AI sales automation is legal in the UK in 2026, but the rules that bite are not AI rules — they are the Privacy and Electronic Communications Regulations (PECR) and the UK GDPR, and the penalty for getting PECR wrong rose from £500,000 to £17.5 million or 4% of worldwide turnover under the Data (Use and Access) Act 2025. In practice that means an AI agent may research accounts, draft personalised messages, score leads and handle inbound replies with very little friction, while cold email and text to sole traders, automated voice calls, and calls to numbers on the TPS or CTPS stay tightly restricted whether a human or a model pressed send.

So the design question is not "can we use AI for outbound?" but "which parts of our pipeline are regulated channels, and which are internal work the AI can take over outright?" That split is the whole basis of a workable AI sales automation build, and it is where most UK teams get it wrong — they automate the regulated part and leave the unregulated grunt work to people.

What actually changed for UK outbound in 2026

Three things matter to anyone running automated sales outreach from or into the UK.

PECR fines caught up with the UK GDPR. The Data (Use and Access) Act 2025 received Royal Assent on 19 June 2025 and lifted the maximum penalty for breaching PECR's direct marketing rules to £17.5 million or 4% of total annual worldwide turnover, whichever is higher — in line with the UK GDPR ceiling. The old £500,000 cap was, for a funded scale-up, a cost of doing business. The new one is not. Note that the increase does not extend to the separate PECR duty to notify the Commissioner of a personal data breach.

Most of the remaining data protection provisions commenced on 5 February 2026. The ICO confirmed the timing in its own commencement statement, with the new complaints-handling procedure requirement following on 19 June 2026. If your last PECR review was written against the pre-DUAA text, it is out of date.

The ICO is now looking directly at agentic AI. On 8 October 2026 the regulator opened an agentic AI call for evidence, closing 20 November 2026, covering data security, transparency, accountability, automated decision-making, fairness and purpose limitation, and lawfulness of processing. Final guidance has not been published, so anyone telling you the compliance position for autonomous sales agents is settled is guessing. Build so that the guidance, when it lands, does not force a rewrite.

The channel rules, in plain terms

PECR does not care whether a model or a person composed the message. It cares who the recipient is and which channel you used. The ICO's business-to-business marketing guidance sets out the distinctions that decide most of this.

Two practical traps. First, if you cannot tell whether a contact is a corporate or an individual subscriber, the ICO's position is to treat them as an individual — so a list that mixes "info@" company addresses with named sole traders needs classifying before, not after, the AI sends. Second, the soft opt-in is narrower than most teams assume: you must have obtained the details during a sale or negotiation for your own product, be marketing something similar, have offered an opt-out at collection, and offer one in every message. All four, not any one.

Where an AI voice agent is in scope, the rules are tighter again and worth reading alongside our breakdown of AI voice agents in the UK.

Lead scoring, automated decisions and the DUAA relaxation

AI sales stacks score and route people. That is profiling, and the DUAA changed the rules around it. The old restriction applied to all solely automated decisions with legal or similarly significant effects; it now applies only to significant decisions based wholly or partly on special category data. For ordinary B2B lead scoring, that is a genuine loosening.

It is not a free pass. Safeguards are still required for any significant decision taken solely by automated means: the person must be able to get information about the decision, make representations, obtain human intervention and contest the outcome — and the ICO has been consulting on updated automated decision-making guidance reflecting the reforms. Two tests worth applying to your own scoring model:

• Does the score decide anything significant for the person, or only how your team spends its time? Prioritising a queue is usually the latter.
• Could special category data leak into the features — health, trade union membership, beliefs — via scraped bios, enrichment vendors or free-text notes? If yes, you are in the restricted category whether you meant to be or not.

Transparency is the part teams skip. If you buy or enrich contact data rather than collecting it directly, the UK GDPR requires you to provide privacy information within a reasonable period and no later than one month. An AI agent that quietly builds profiles from third-party sources without that notice is the clearest compliance gap we see in UK outbound stacks.

What it costs to build and run

The figures below are Stratgik's own indicative ranges for UK and India-delivered work, not vendor list prices, and they assume you already have a CRM. Treat them as planning numbers and pressure-test them against your own volumes with our app cost estimator.

Two cost realities worth naming. Model and data costs are rarely the problem — screening, consent records and audit logging are where the engineering hours go, and they are also what makes the system defensible. And the run cost scales with volume, not seats, so a pipeline that triples its sends triples its inference bill. Budget for the second year, not the first month.

If you are weighing a purpose-built platform against something assembled around your own data, the trade-offs are the same ones in our build vs buy tool and in custom software engagements generally.

A practical checklist before you switch anything on

1. Classify every contact record by subscriber type — corporate, sole trader, partnership — and make the classification a hard gate the agent cannot bypass.
2. Wire the suppression list in as a pre-send check, not a post-send report. Suppress rather than delete, so future lists can be screened.
3. Screen call lists against both the CTPS and the TPS, plus your own do-not-call list, on a schedule — not once at import.
4. Write down the lawful basis for each processing purpose, and run a legitimate interests assessment where you rely on it.
5. Check whether your privacy information actually reaches people whose data you bought or enriched, inside one month.
6. Audit your scoring features for special category data leakage from scraped or enriched sources.
7. Give every significant automated decision a human-intervention route that someone is actually accountable for.
8. Log what the agent did, with which data, under which policy version. If you cannot reconstruct a send six months later, you cannot defend it.
9. Set a review date against the ICO's agentic AI guidance once it is published.
10. Put a kill switch on the agent, and test that it works.

Stratgik is a technology firm, not a law firm. We implement the controls, logging and gating that make a sales stack auditable, and we work alongside your own legal advisers on the questions that need a legal opinion. Nothing here is legal advice.

Where AI sales automation pays off fastest

Across UK engagements, the fastest return is almost never cold outbound volume. It is the unregulated middle of the pipeline: inbound replies answered in minutes rather than days, account research that no longer eats a third of an SDR's week, CRM data that stays accurate without nagging, and call notes and follow-ups written automatically. None of that touches PECR, all of it is measurable, and it buys the credibility to do the regulated parts carefully.

That sequencing — unregulated first, regulated once the controls are proven — is also how you avoid the failure mode where a compliance incident kills the whole programme. If outbound is only one part of a wider operations problem, managed technology and our UK practice cover the surrounding ground, and UK GDPR compliance work can run in parallel.

Frequently asked questions

Can I send cold AI-written emails to UK businesses?

To corporate subscribers — limited companies, LLPs, Scottish partnerships — yes, PECR's consent rule for email and text does not apply, provided you do not conceal your identity, you give a valid opt-out address and you honour opt-outs. Sole traders and most partnerships are treated as individuals and need consent unless the soft opt-in conditions are all met. The UK GDPR still applies to any named individual throughout.

Does it matter that an AI wrote the message rather than a person?

Not to PECR. The channel and the recipient determine the rules. What AI changes is scale and auditability: a model sending thousands of messages makes a classification error thousands of times, so the gating and logging around it carry much more weight than they did at human volumes.

Are AI voice agents allowed for UK sales calls?

Automated or recorded marketing calls need consent that specifically covers automated marketing calls, which is a high bar. Live calls — including those an AI agent places and a human conducts — require screening against the CTPS and TPS and honouring objections. Treat voice as the last channel to automate, not the first.

Does the UK have an AI Act we need to comply with?

No. The UK has not enacted an EU-style AI Act; sales automation is governed through existing law, mainly the UK GDPR and PECR, with the ICO as the active regulator. Its agentic AI call for evidence closed on 20 November 2026 and final guidance is still pending, so the position may firm up.

How long does a compliant AI sales stack take to build?

An inbound reply and routing agent is typically four to eight weeks. Outbound orchestration with subscriber gating, TPS/CTPS screening and a consent ledger is usually ten to sixteen weeks, most of it on the control layer rather than the models.

What is the single biggest compliance risk?

Unclassified lists. Sending to sole traders in the belief that they are companies is the error with the largest downside now that PECR penalties reach £17.5 million or 4% of worldwide turnover.

If you are planning AI in your sales pipeline and want a straight answer on what is safe to automate first, what it will cost and where the controls need to sit, book a free 30-minute session through contact us. We will map your pipeline against the channel rules and tell you which parts of AI sales automation will pay back this quarter and which should wait — and if the honest answer is that you need a quarter of data work before any of it, we will say so.