Compliance

Built for compliance.

Every insurer asks us the same four questions: will it make things up, will it promise things we can't deliver, how does it treat vulnerable customers, and how do we evidence all of this? The sharpest teams ask a fifth: can we test it ourselves? Good questions. Here are the answers.

Accuracy

It says what's true, because it works from your facts.

The model does the talking. Your configuration does the deciding.

  • Prices come from your rate tables. The agent presents them. It never invents them.
  • What it can say about your products comes from your product configuration and approved journey copy, not from the model's general knowledge.
  • Journeys follow vetted, versioned templates, checked automatically before they ship.
  • Answers that go stale (income, health, address) are re-confirmed before they're relied on again.
  • Scripted customer journeys run against the live model before changes go out, checking what the agent said, what it did, and in what order.

Promises

It only offers what it can deliver.

The agent acts through a fixed set of tools: quote, bind, take payment, schedule a call-back, open a claim, bring in a human. Its promises are limited to its tools, and its tools are limited by you.

  • Guardrails screen every reply for the classic failure cases: guaranteed claim outcomes, cover confirmed without its conditions, invented policy terms.
  • When it isn't sure, it says so, and brings a person in rather than improvising.
  • Payments above your threshold wait for a named person to approve them.
  • Commitments are kept because they're scheduled, not remembered: "call me tomorrow at five" becomes a diarised call, on the record.

Vulnerable customers

Extra care, exactly when it matters.

  • Conversations are screened for signs of vulnerability: financial difficulty, bereavement, ill health, confusion, distress.
  • When the agent sees them, it slows down, simplifies its language, drops anything that resembles selling, and offers a human.
  • Escalation rules bring your team in on the triggers you define; urgent escalations stop the conversation until a person picks it up.
  • Your team can watch any live conversation from the inbox and step in at any point.
  • There are no pressure mechanics anywhere: contact is capped, messages have cooldowns, and opt-outs are honoured immediately.
F Float Insure Web chat
Sorry if I'm slow today. My husband passed away last month and I'm trying to sort everything out.
I'm so sorry for your loss. There's no rush at all. We can go at whatever pace suits you, and I can ask a colleague to call you instead if you'd prefer.
Vulnerability flagged · routed for human review
Chat is fine. I just need to know what happens to his policy.
Of course. I have the policy in front of me. Let's go through it together, one step at a time.

Consumer Duty

Evidence, not assurances.

Consumer Duty asks you to demonstrate good outcomes, not just intend them. The platform is built so the demonstration is already there when you're asked.

  • Guardrail packs written around FCA expectations ship with the platform: clear, fair and not misleading; no regulated advice; plain language; no pressure selling.
  • Every conversation is stored and reviewable, with exportable transcripts.
  • Every lifecycle event (quote issued, policy bound, claim approved) lands in an append-only log with who did it and when.
  • Every guardrail check and every compliance decision is logged, message by message.
  • Agents run a pinned, numbered version per conversation. Months later, you can reconstruct exactly which configuration said what, and why.

Testing

Ask it the hard questions before your customers do.

The platform ships with a compliance test harness. Your compliance team writes the questions a regulator would ask, and the answers they'd expect, and runs them against the live agent from the dashboard. No engineers required.

  • Checks run as real sandboxed conversations against the real agent: same model, same prompts, same tools, same guardrails. On the channel you choose, from web chat to SMS, email or voice.
  • Every answer is graded against the answer you expected. Doubt counts as failure.
  • Beyond single questions, full scripted journeys: record a real conversation once, replay it as a script, with expectations on every turn.
  • Schedule any question group to run daily, weekly or monthly against your latest agent version, and compare runs side by side to catch a regression before a customer meets it.
  • Every run exports an audit-ready PDF: verdicts, the judge's reasoning, tool calls, and the full transcript behind every answer.

Bring us the hard questions.

These are the conversations we like having. Bring your compliance team. We'll walk through any of it in as much depth as they want.

Book a call Talk to our agent