Grok Bot Front Desk: The RepairDesk Clerk

What this bot is for

A front-desk Grok Bot is the clerk at the counter, not the owner and not the person who texts the customer on its own. Its job is to turn a walk-in or a website booking into a clear ticket: who is here, what device they brought, what is wrong in one line, which repair tier you recommend, and how they plan to pay. It also keeps open tickets from going stale.

That sounds simple. In a one-person shop it is the work that spills — a note on a pad, a half-finished ticket, a booking that never made it onto the calendar, a customer who has been "waiting on parts" for too long. The bot's value is the same every time: take a short, messy intake and return a complete draft you can check in a minute.

It does not send customer messages. It does not take payments. It does not change a price behind your back. If a text, a reminder, or a review request should go out, it writes the draft and stops.

Intake without chaos

Start with as little as you actually get at the door. A name, a phone number, a device, and one sentence about the problem is enough. The bot should look up a returning customer by phone first, then by name, so you do not create a second record for someone who was already in the system last month.

A good intake draft has:

  • A single-line problem you could read out loud ("iPhone 14, cracked screen, no other damage reported").
  • The device and, when you have it, the color or capacity only if it changes the work.
  • How they found you or whether this is a return visit, if they said so.
  • A recommended tier based on the rules you already use, not a guess invented on the spot.
  • A flag if anything in the request is incomplete (no passcode policy mentioned, liquid damage, a second issue mentioned in passing).

The clerk bot should not pad the ticket with marketing language. It should not promise a time it cannot keep. If the walk-in note is too thin, it should say what is missing rather than invent a cleaner story.

Quotes and tiers in language a customer can see

This shop quotes labor and parts in public tiers — for example an economy screen versus a higher-grade one — with the posted price on the site matching what you say at the counter. The bot's job is to apply those same rules automatically, then write the quote in words a customer can see.

Two hard limits matter here:

  • Never put a supplier or wholesaler name in the ticket note that might be shown to a customer, and never in a text draft.
  • Never show part cost. If a quote would go below cost, the bot flags it for you. It does not display the cost, and it does not silently raise or cut the price.

If the customer asks for the cheapest option, the draft should explain what that tier is, what it is not, and what the posted price is. If they ask for the longest-lasting option you actually sell, the draft should stay inside the same public language you already use on the site. Do not use factory-official or maker-endorsed quality claims. That kind of wording causes trouble and it is not how this shop describes parts.

When a quote is unusual — two repairs at once, a device that is already in poor shape, a rush request — the bot should list the assumptions it made and ask you to confirm them. A clerk bot that hides uncertainty is worse than a clerk bot that asks.

Payments: record, don't guess

The front-desk bot records how the customer plans to pay and, later, the date the money actually landed. Those two facts are easy to mix up. A card run on Tuesday for a job that was quoted on Monday should be dated as the day the money landed, not the day the ticket was opened.

The draft should capture:

  • Payment type (card, cash, or the cash-app method you actually use).
  • Whether a deposit was taken, and on which date.
  • Whether the balance is still open.
  • A short note the Accountant can match later, without account numbers and without a full card PAN.

The bot does not "mark paid" because a customer said they would pay later. It does not complete a ticket because a reminder was written. Recording is not collecting. Collecting stays with you.

Booking to calendar

When the job starts on the website instead of at the door, the same clerk bot should turn the booking into a ticket draft and a calendar event. The event needs the same one-line problem, the length of time you actually want on the calendar, and any reminder you usually send — still as a draft if a send is involved.

A useful pattern is:

  • Create or update the ticket from the booking fields.
  • Draft the calendar event on your shop calendar, not a personal calendar the bot invented.
  • Draft the reminder text and leave it unsent.
  • Put a line in the daily brief if the booking is for today or tomorrow and the ticket is still incomplete.

If the booking is missing a phone number or the device is unclear, the bot should not guess. It should list the gap and wait.

Aging list at the end of the day

Open tickets rot quietly. A customer waiting on a part, a ticket stuck in "waiting on customer," a repair that is done but not closed — those are the jobs that come back as an angry call. Once a day, usually toward close, the front-desk bot should list tickets that have sat in a waiting status too long.

The aging list is a working document, not a shaming report. For each stale ticket it should say: how long it has been waiting, what the last public-facing status was, and what a reasonable next step is (a draft status text, a call reminder for you, or a handoff to Cortana for the morning brief). It should not auto-message the customer. It should not close the ticket to tidy the list.

What stays a draft until you approve it

This is the rule that keeps a clerk bot safe:

  • Status texts, review requests, pickup reminders, and any other customer-facing message are drafts until you say the exact words to send them.
  • Calendar invites that leave the shop stay drafts.
  • Price changes, refunds, and "we can do it cheaper" offers stay drafts.
  • Anything that would put a vendor name, a cost, or a private note into a customer channel is blocked, not just warned.

If you want the bot to prepare three reminder options, that is fine. Sending is a separate step, and it should require you.

Sample prompts

You are my repair-shop front desk clerk. From this walk-in note, draft a ticket: customer name, device, problem in one line, recommended tier, and labor rules we already use. Do not send a message. Do not show vendor names. Flag anything that would go below cost, and give me the draft to approve.

Look up this phone number in our tickets first, then the name. If this is a returning customer, add today's visit to their existing record. If you are not sure it is the same person, show me both options and do not merge anything.

Make tonight's aging list: tickets that have been in a waiting status longer than our usual window. For each one, give the wait time, the last status, and a draft next step. Do not text the customer.

Turn this website booking into a ticket draft and a calendar event on the shop calendar. Draft a reminder and leave it unsent. List anything the booking left out.

Record this payment: type, amount we were told, and the date the money actually landed. Leave a short matching note for the accountant. Do not include account numbers. Do not mark the ticket paid unless I say the money is in.

How this bot works with the rest of the team

Front Desk hands payment notes to the Accountant, aging and "needs a decision" tickets to Cortana for the morning brief, and any "what should we actually say about this tier" questions to the Localizer. It does not invent a warranty policy or a price. Those already live on the site and in your shop rules.

If you only set up one bot after the getting-started guides, this is a strong second bot: short intake in, complete ticket draft out, nothing sent until you say so.

Back to blog