This is a sample project for a fictional business. Fable & Foam Brewing doesn't exist; we built it to show what our AI Workflow Automation & Audit delivers from start to finish. The workflow, the logic and the sandbox are real.

Fable & Foam is a small brewery with two event spaces. It gets about 40 event enquiries a month (birthdays, company parties, rehearsal dinners, tour requests) by email, Instagram, phone and the website form. Before the automation, one events coordinator handled all of them, and replies took one to three days. By then, a lot of those leads had already booked somewhere else.

An AI workflow audit is built for this kind of problem that has the same characteristics: (1) the job is repetitive, (2) doing it slowly has a clear cost, and (3) time spent by the events coordinator is better spent elsewhere.

The workflow before

We mapped the process from a single walkthrough with the coordinator. Written out, it looked like this:

  1. Check four inboxes: email, Instagram messages, voicemail and the website form.
  2. Copy each enquiry into a spreadsheet: the occasion, the date, the number of guests, food, contact details. About 10 minutes per enquiry.
  3. Check the calendar to see if the date is free, and which space fits the group.
  4. Write a reply with a quote by hand, answering the same questions about food, deposits and timings each time.
  5. Remember to follow up if the customer goes quiet.

Most of the time went into steps 1, 2 and 4. Step 5 depended entirely on the coordinator's memory.

What we automated

The audit ranked this workflow first. It had plenty of volume, it was mostly rule-based, and the information it needed was already written down somewhere. After the build:

  • Every channel lands in one queue, so nothing gets missed in a fourth inbox.
  • AI reads each message and fills in the record (occasion, date, guests, food, contact), so nobody copies anything.
  • The automation checks the calendar and capacity rules and picks the space that fits, or offers the next free dates if the requested one is taken.
  • AI drafts the reply in the brewery's own tone, using its real food options, policies and templates.
  • If the customer hasn't replied, follow-ups go out on day 2 and day 5.

How the messages get in

The automation never connects to the brewery's main inbox or its phone account. It only sees what arrives at a front door set up for it:

  • Website form: each enquiry goes straight to the automation.
  • Email: the brewery lists a dedicated address, events@, on its site and in its email signature. For enquiries that still land in the main inbox, the brewery has set up a forwarding rule it controls: if a message mentions an event, a party or a booking, it's forwarded to events@. A business that runs on one personal Gmail account can do the same with a free address such as examplewineryevents@gmail.com and a Gmail filter.
  • Voicemail: the phone system emails a transcript of each voicemail to events@. The automation never logs into the phone account.
  • Instagram: the brewery has connected its business account with Meta's messaging permission, which covers direct messages only. The automation can't post or change anything on the account.

Approved replies go out from the events address too. The customer's answer comes back to the same place, so the automation can tell when a follow-up is due.

Everything runs in the brewery's own accounts: the events mailbox, the enquiry spreadsheet and the automation platform. We had access while we built it, and we removed that access at handover. To switch it all off, the brewery deletes the forwarding rule or the address.

To read the messages and draft the replies, the automation uses Claude through Anthropic’s business API, which doesn’t use customer data for training, with the API key held in the brewery’s own account.

Where a person still decides

Nothing goes out until the coordinator approves it. They read the draft, edit it if needed, and send. People are often tempted to skip this step. We keep it because a quote is a promise, and a wrong one costs more than the minutes it saves. Approving a good draft takes a moment; writing one from scratch takes ten minutes or more.

The sandbox runs five sample enquiries, including the awkward ones. A holiday party on a date that's already booked gets the next free dates. A wedding for 250, more than either space holds, gets a polite no with an alternative.

The result

The wait for a reply went from one to three days to a few minutes, with a drafted quote ready soon after each enquiry arrives. The coordinator now reviews drafts instead of re-typing details, and follow-ups go out on schedule.

The brewery's prices, spaces, policies and reply templates also live in one place the team can update. Before, a lot of that was in the coordinator's head.

Try it yourself

You can watch the walkthrough and see the full write-up, or open the sandbox and run the five sample enquiries in your browser. Nothing is sent anywhere.

If your business answers the same kind of enquiry over and over, the free AI Opportunity Scan is a quick first look, and AI Workflow Automation & Audit explains how we find and build the right first automation for you. We covered the audit step by step in What an AI Workflow Audit Covers.