How our systems work

Everything we set up for you starts with a single business profile. It's the one place that describes your business to every agent. Your website assistant, phone receptionist, booking agent and follow-ups all read from it, so they never disagree with each other.

When something changes, such as new hours, a new service or a new team member, we update the profile once and every agent follows. The profile is checked against strict rules every time it changes. A bookable service without a duration, or a setting we don't recognize, is caught before it can cause a problem.

Our AI agents don't make things up about your business. They answer from the information you've approved in your profile and nothing else. Starter questions from an industry template stay unapproved until you sign off, and only approved answers ever reach the AI or a public page.

When a question falls outside what you've approved, the agent says so and hands it to a person. It takes the customer's details, writes a clear summary, and routes it to the right member of your team. Questions about pricing, contracts, complaints and other sensitive topics are held for you. Unless you've approved specific pricing wording, the AI explains that pricing is confirmed by your team.

The AI also never claims something happened unless the system confirms it. It can't tell a customer "you're booked" unless the booking service actually saved a confirmed booking. If the booking is a request, the customer is told it's a request. Anything a visitor or caller types is treated as information, never as an instruction. A message can't talk the AI into changing its rules, its prices or what it's allowed to do.

Compliance isn't an add-on we bolt on at the end. It's part of how every message is sent, and the system checks it on every attempt. That protects your customers, and it protects your phone number and your reputation.

Our AI says it's an AI. The phone receptionist tells callers up front that they're speaking with an AI assistant and that the call may be transcribed. Texts go only to people who agreed to receive them. Replying STOP suppresses the number right away, START restores it only if consent had been given before, and HELP replies with your business name and contact details. Quiet hours hold texts and follow-up emails until morning in your time zone.

Every customer email includes an unsubscribe link, and follow-up emails show a one-click Unsubscribe button in Gmail and Outlook. Marketing email goes only to people who opted in, and it always includes your postal address. Website analytics use no cookies, record no typed text or IP addresses, and skip visitors who've turned on Global Privacy Control or Do Not Track. We never create fake reviews or filter which customers are asked for one.

Each client's information is kept strictly apart. Every record in our system is tagged with the business it belongs to, and every request to read or write data is limited to that one business. There's no back door that lets one client's leads, messages or settings show up for another. The AI only ever sees your business's approved knowledge.

You own your accounts and your data. Your phone number, email sending, CRM and ad accounts are set up in your name, and any ad spend is paid directly to the platform, separate from our fees. Your records can be exported, and if you pause or end service, your data is kept until you decide what happens to it.

You also decide how long records are kept. Each night, records older than your retention setting are removed, with a dry-run preview available first. Anything tied to an upcoming appointment is never removed, and opt-out records are kept permanently so nobody who said stop is ever contacted again.

Your passwords, API keys and access tokens never sit in your business profile, your website code, an AI prompt or a log file. The profile holds only a reference to where a secret is stored. The real value is looked up on our server at the moment it's needed, and only there.

If a value that looks like a provider key is pasted into a profile, the system rejects it. Logs and audit records are automatically scrubbed of anything sensitive before they're written. The widget on your website contains no secrets at all. It only works from the website addresses you've approved.

Each connection gets only the permissions it needs. Messages coming in from providers like Twilio are checked for a valid signature, and each one is accepted only once, so a forged or replayed message is refused.

Every outbound action, whether it's an alert, a confirmation, a follow-up, a reminder or a CRM update, is saved as a job with a unique key before anything is sent. That key is what lets the system know for certain whether a message already went out.

If a provider has a hiccup, the job retries on its own with growing waits between attempts. Before every attempt, the system checks the key. If the message was already sent, nothing happens. A job that runs out of retries is marked failed, listed on your dashboard with the reason, and triggers an alert. Once the cause is fixed, Replay sends it once. A message is only ever marked as sent when a live provider accepted it.

Kill switches let us pause sending instantly, for one feature, for your whole business, or for one feature across every client. Paused messages wait without being lost and resume when the switch is cleared, with nothing doubled. Opt-outs and unsubscribes keep working even while sending is paused. Every change to a lead, a booking or a setting is written to an audit log with the time and who made it.

Nothing reaches your customers until you've seen it work. Every agent runs first in dry run, our safe mode. It follows exactly the same rules as live sending: consent, quiet hours, stop conditions and limits. The difference is that each message is recorded as "dry run" instead of being sent.

That lets you review real wording, real timing and real decisions before anything goes out. You see which follow-ups stop when a customer replies, which texts wait for quiet hours to end, and how the AI answers a tricky question. Your daily summary is labeled DRY RUN for as long as safe mode is on, so there's never any confusion.

As part of every setup, we also run a full test: a sample inquiry becomes a lead, then a booking that's cancelled right away, then notifications, then a dashboard update. These test records are tagged and kept out of your real numbers.

Setup follows the same repeatable process for every client, so nothing is skipped and every step can be checked. Scope and price are confirmed in writing after a short conversation, before setup begins.

Once you're live, your client dashboard is the one place to see what your agents did and what needs you. It shows leads by status, bookings requested and confirmed, calls with summaries and transcripts, follow-ups in progress, failures and usage against your limits. Replies on sensitive topics wait in an approval queue for you to send, edit or discard. Booking requests wait for your approval.

The dashboard is honest by design. Every number comes from real records. A feature that isn't on shows as unavailable, never as zero. A missing connection shows as disconnected. Test messages are counted separately from messages actually sent.

Each morning, your team can get a short summary email with yesterday's new leads, bookings, cancellations, replies, opt-outs, failures and usage. On a day with no activity, no email is sent unless you ask for one.

Your systems keep getting better after launch. Website analytics show where visitors get stuck, which forms they abandon and where they click repeatedly without result. A daily AI-written report turns that into a plain summary and the three most useful fixes.

We review that report, your failures, the questions the AI handed to a person and your opt-out patterns with you. When the same question keeps coming up, we draft a new answer for your approval. Every change to what your customers see or hear goes through you first.