1. Our position, in one paragraph
Some of our business-to-business outreach calls are made by a conversational AI rather than a human caller. We believe this is lawful, legitimate, and — done the way we do it — more accountable than conventional human telemarketing, because every factual claim our caller may make is pre-approved and source-grounded, every call is recorded and its transcript machine-checked against those approved claims, and every gate — statutory and self-imposed alike (TPS/CTPS screening, calling hours, do-not-call suppression) — is enforced in software that cannot be talked out of it. This document sets out the whole argument, the rules we hold ourselves to, and how to complain if we fall short.
Scope. This policy governs Friam’s marketing and outreach calls. Separately, we operate AI service-assessment calls (“mystery shopping”) that a business commissions to assess its own front-of-house — those are governed by the commissioning business’s instructions and our AI Ethics & Transparency Policy, and are out of scope here.
2. What these calls are — and are not
The phrase “robocall” describes a pre-recorded message blasted at a list: one-way, deaf, unstoppable, the same regardless of who answers. Our calls are the opposite of that. They are live, fully interactive conversations: the caller listens, understands what is actually said, answers unscripted questions accurately, handles objections it has never seen phrased that way before, and ends the call immediately if asked. A person who says “not interested” is heard the first time.
The underlying technology is a frontier large language model driving real-time speech — the class of system whose published benchmark results on graduate and professional examinations are a matter of public record. We won’t decorate that with our own superlatives: because every call is recorded and transcribed, the quality of the conversation is checkable call by call, and we would rather be judged on transcripts than adjectives. What matters here is the structure: this is not a talking answering machine but a fully conversational representative whose permitted assertions are written down in advance — it may state as fact only what we have sourced and approved (§4).
3. The legal frame
3.1 PECR and live marketing calls
The Privacy and Electronic Communications Regulations 2003 (PECR) draw the line between live unsolicited marketing calls (regulation 21 — permitted to businesses, subject to TPS/CTPS screening, identification and opt-out) and automated calling systems (regulation 19 — prior consent required). Regulation 19 speaks of communications “comprising recorded matter”, and regulation 19(4) defines an automated calling system as one capable of automatically initiating a sequence of calls from stored instructions and of “transmitting sounds which are not live speech”. We quote that second limb deliberately: it is the strongest counter-argument to our own position, and a candid document doesn’t pretend it isn’t there.
Our reading: regulation 19 targets the one-way transmission of pre-prepared material — recorded matter delivered identically regardless of who answers. A generative conversation contains nothing pre-prepared and nothing recorded-in-advance: the words are composed in the moment, responsively, in a two-way exchange whose course the called person’s own words control — the functional substance of live speech. In substance these are live calls, and we conduct them under regulation 21’s duties in full. We acknowledge plainly that no court or ICO decision has yet settled the classification. Our response to that uncertainty is not to seek cover in it: we meet every duty regulation 21 imposes, add controls neither regime requires (§4), monitor ICO and Ofcom output, and will adjust promptly — up to and including pausing calling — if the position hardens against our reading.
3.2 The duties we treat as non-negotiable
- TPS/CTPS screening, fail-safe. Every number is screened against the Telephone Preference Service and Corporate TPS before any cold call. Our dialling software refuses to place a cold call to a number that has not been screened and come back clean — an unscreened number cannot be dialled at all. Numbers on either register are not cold-called, full stop.
- Identification, every call. The caller opens by saying who we are and that the call may be recorded, from fixed wording that the system enforces (see §4). We present a valid, callable number — never a withheld or spoofed line.
- Instant, permanent opt-out. “Don’t call again” takes effect at once: the number goes onto a global do-not-call suppression list that our dialling software checks — in the same software gate as the TPS screen — before any future cold call, across all our products.
- Business targeting and business hours. These are business-to-business calls to trading businesses on public registers, placed only inside configured working hours (UK time, working days), with daily volume caps and per-recipient frequency caps.
- No abandoned or silent calls by design. Ofcom’s persistent-misuse policy exists mainly because human call centres dial more lines than they have agents. Our caller is never busy: one call, one fully-present conversational agent, every time — the over-dialling that produces abandoned calls has no equivalent in our architecture. Calls that reach voicemail are ended rather than played a recorded message.
- Data protection. Contact data comes from public business registers and firms’ own public websites, processed under legitimate interests as set out in our Privacy Policy and Web Scraping & Data Collection Policy. Recordings and transcripts are held securely and retained only as long as needed for compliance evidence.
4. The control architecture — why this is more controlled, not less
The honest comparison is not “AI call versus no call”. It is “AI caller versus human telemarketer”. Here is what our system enforces that no human calling floor can:
- Immutable disclosure. The opening disclosure — who we are, that the call may be recorded, and that we are not the regulator — is stored as fixed, hash-pinned text and carried into every marketing call’s configuration. Our script-testing machinery is forbidden from touching it: a candidate script whose disclosure, claims-scope or offer text no longer matches the pinned fingerprint is refused before it can ever make a call. A human agent can skip the disclosure on a tired Friday afternoon; our configuration can’t lose it.
- A closed claims library. Every factual claim the caller is permitted to make is written down in advance and tied to a primary source (legislation, official guidance, the public register entry). Anything not in the library may not be asserted. There is no equivalent guarantee for a human SDR working a quota.
- Fixed offer bounds. The price and the offer are pinned text, including the words “no guarantee of any regulatory outcome”. The caller cannot invent a discount, inflate a risk, or promise a result.
- Every call recorded, transcribed, and machine-audited. Each transcript is checked after the fact against the approved claims library: any fact the caller asserted that is not in the library is flagged as a violation, and a campaign whose calls drift is frozen automatically. (The disclosure and offer text are enforced the other way round — as fixed inputs to every call — rather than checked afterwards.) A kill-switch stops a campaign instantly, including calls already queued. Ask a conventional telemarketing operation for a per-utterance audit of every call it made last month.
- No incentive to lie. Human boiler-room misconduct is driven by commission pressure, fatigue and fear. Our caller has no quota, no commission, no bad days, and no capacity to decide the rules don’t apply just this once.
We think the ethical question about any outbound call is not who is speaking but whether what is said is true, permitted, and respectful of the person called. On those criteria our architecture outperforms the industry it replaces — measurably, because unlike that industry, we keep the evidence.
5. Honesty commitments
- The caller never claims to be from DVSA, HMRC, or any regulator or public body — the fixed disclosure says the opposite, on every call.
- On being an AI, candidly: our marketing calls do not open with an announcement that the caller is artificial — no current UK rule requires one, and we’d rather say that plainly than leave it unsaid. What we treat as immovable is the flip side: our calling configurations require the caller to answer truthfully if asked whether it is an AI, and it never denies it. If ICO or Ofcom guidance moves to require proactive disclosure, we will adopt it — this is exactly the kind of point the six-monthly review of this policy exists for.
- The caller never impersonates a real, named human being.
- No invented statistics, no scare claims, no urgency theatre, no pressure loops. If the answer is no, the call ends politely and the number is not worked again.
- We do not make pre-recorded message calls at all, and we do not make consumer cold calls at all — our cold outreach is to trading businesses on public registers only, with corporate numbers still screened against TPS/CTPS.
- Welfare before selling. If the person we call is in distress, mentions a personal crisis, or asks for help with anything beyond our product, the sales purpose of the call ends immediately and permanently. The caller is configured to respond as a decent human would — briefly, without ever playing counsellor — and to signpost real help: Samaritans on 116 123 (free, 24 hours a day), 999 in an emergency, NHS 111 for urgent medical advice, and a GP or qualified professional for everything else. The call then ends gently and the number is suppressed from all future outreach. Someone who is angry at being called gets de-escalation only: an apology, an immediate permanent opt-out, and a polite end — never argument, never humour, never a return to the pitch.
6. If we call you and you’d rather we hadn’t
Say so on the call — suppression is immediate — or email legal@everyguard.uk with the number we called, and we will suppress it and confirm within 5 working days. You can also complain to the Information Commissioner’s Office (ico.org.uk), which regulates PECR. We keep the recording and transcript of every call, so a complaint can be checked against exactly what was said — by us or by the regulator.
7. Review
Because regulatory treatment of AI calling is evolving, this policy is reviewed six-monthly rather than annually, and promptly upon any relevant ICO or Ofcom guidance. The directors of Friam Limited own this policy.