FAQ
Frequently asked questions
The questions worth asking before you decide, answered straight.
How does Civilly make sure it's talking to the right person?
It asks knowledge-based questions from your own records — for example a postcode, a reference number or a recent payment amount — and only proceeds once enough of them match your policy. Nothing about the account, not even the caller's name, is revealed until verification passes. Caller ID is logged for audit but never counted as proof, because it's trivial to spoof.
What happens if Civilly can't verify a caller?
It offers a fair, uniform retry on a mistyped or mis-heard answer, but it doesn't hint at how close someone got — that would let a caller probe for the real value. After enough failed attempts, further attempts are locked out and the call is offered a human transfer or callback instead.
Can Civilly change or waive someone's balance?
No. There is no tool that can zero, waive or otherwise edit a balance — that's a deliberate design decision, not a policy the AI is just told to follow. Payment plans use your own calculation rules, and the model can only relay the numbers your rules produce, never invent its own.
What happens outside business hours?
Civilly is grounded in your real opening hours. Outside them, transfers are switched off and it offers a callback instead of pretending someone is available to take the call.
What stops it from just making something up?
Four layers, and only one of them is the model's own judgement. The actions that could actually harm someone don't exist as functions, so there's nothing to be talked into: there is no way to alter a balance, and payment plans are calculated in code from your rules. Every figure, date and reference the agent is about to say is checked against what your systems actually returned — anything it can't trace isn't spoken. A second, stronger model reviews each exchange while the reply is playing and can put a correction in front of the next one. And the whole call is reviewed again afterwards, with the findings on your record. The full explanation is here, including where those controls stop being useful.
Do you only work with UK businesses?
UK and Ireland today. That's a deliberate limit rather than a technical one — the platform is English-language and the telephony provider sells numbers in most countries, but address validation, number formatting, currency, the data-protection framing and, most practically, the support hours behind our SLA are all built around UK and Irish businesses. We'd rather do that properly than claim a global service we couldn't answer the phone for at 3am. If you're elsewhere and the fit looks right, ask — we'll give you an honest answer about what it would take.
Where is our data processed, and is it used to train AI models?
It isn't used for training: our speech and language suppliers are engaged on commercial API terms that prohibit training on data submitted through them, and that's a condition of the contract rather than an assumption. Processing does involve suppliers in the United States, under the UK International Data Transfer Addendum, with data encrypted in transit and personal identifiers redacted before anything is stored. Every sub-processor is named in our Data Processing Agreement — if UK-only residency is a hard requirement for your sector, tell us early and we'll tell you straight whether we can meet it.
What happens when a caller shares sensitive personal information, like health details?
It gets handled well, then it gets minimised. On the call, the agent slows down, flags the account for careful handling and drops any payment pressure. What's stored afterwards is deliberately thin: the care flag is a fixed category — "bereavement", "serious illness" — applied in code, never the caller's own words; an automated post-call review withholds transcript lines containing special-category data from storage; identifiers are redacted before anything is stored at all; and the lot is purged on a fixed retention window. The recognition is automated, so we describe this as minimisation rather than a guarantee — the precise version, limits included, is in our Data Processing Agreement.
Which CRMs and case-management systems does it work with?
Salesforce and HubSpot connectors are built, and each one is proved against your own org during setup — we don't call an integration done until it has read your real records. Anything else with an API gets a bespoke adapter, or a database Civilly can query safely. Building and proving the adapter is part of the setup work.
Is call data stored, and for how long?
Yes, for quality and dispute-resolution purposes, with personal data redacted by default in what's stored. A fixed retention window applies and expired records are purged automatically by a nightly job, not a manual process someone might forget. If a call becomes evidence in a complaint, it can be placed on hold so it survives the purge for as long as the complaint is open.
What happens when a caller can't be heard clearly?
The same things a good call handler does. A long code gets one free re-read, on the same terms for everyone. A reference that keeps getting mangled can be keyed in on the phone's keypad instead. If the line itself is failing, the agent says so — "if you said something, it didn't reach me" — rather than leaving silence or asking the same question a third time. And if the line drops and reconnects within a few minutes, it says so and picks up where it was rather than greeting the caller as a stranger; for safety, identity is checked again.
How long does setup actually take?
It depends on how complex your CRM integration is — a standard Salesforce or HubSpot connection is quicker than a bespoke system needing a custom adapter. We scope an actual timeline on the first call rather than quoting a generic number.
What does Civilly cost to run?
Two parts: a one-off setup fee for the integration and tuning work, and an ongoing cost — a flat platform fee plus the real, metered cost of telephony, speech and AI usage. See the pricing page for a calculator using current provider rates.
Why does Civilly cost more than low-cost AI receptionists?
Because it's doing a different job. The low-cost tools read your website and take messages — genuinely useful when that's all a line needs and nothing it says has to come from your systems. Civilly is wired into your CRM and talks to callers about their own case, which means verifying identity before disclosing anything, actions gated in code, an audit trail, and a person who built the integration, tested it on live calls and stays accountable for it. The price reflects the work it takes off your team — the hours a member of staff would otherwise spend on the phones — not what it superficially resembles. And if message-taking honestly covers your needs, we'll tell you so rather than sell you the wrong thing.
Can we keep our existing phone number?
Yes — that's the normal setup. You forward your existing number to Civilly from your current provider's settings: live in minutes, no porting involved, and reversed just as quickly if you ever want the line back. Porting the number across properly is available later, once the service has proved itself — but nothing requires it, and nothing locks you in.
Does Civilly replace my team?
It's built to take the repetitive "what's my balance / where's my case up to" calls off your team, and to hand off cleanly — disputes, vulnerable callers, anything genuinely out of scope — rather than to replace judgement calls a person should make.
What happens if the AI genuinely doesn't know the answer?
It says so, and offers a callback or transfer. It's built not to answer its own hypothetical questions or fill gaps with a plausible-sounding guess — if the CRM doesn't have the answer, Civilly doesn't invent one.