Skip to main content
Maidpilot
Log in Start free

← Help Center

Configure the AI Secretary

Teach the AI about your business, set escalation rules, and review call summaries.

The AI Secretary is the control center for how MaidPilot behaves on calls and in AI-assisted conversations. If the AI Receptionist is enabled, this is also where you tune the personality, knowledge, and handoff rules.

1. Add the knowledge the AI can use

Open Settings → AI Secretary (requires the AI Receptionist add-on or a receptionist plan). Provide a knowledge document covering services, pricing guidance, FAQs, and policies. Write the way you want the AI to explain things, and include the rules that matter most: minimum job sizes, travel fees, supplies, pets, deposits, cancellations, and when a caller should be handed to a human.

2. Set escalation rules

Add escalation rules for complaints, damage, refunds, legal or insurance topics, and angry callers so the AI hands off instead of improvising. Decide who should be notified and what context the follow-up person needs, such as the caller name, phone number, and transcript summary.

3. Match the voice to your brand

Review sample prompts and greeting language so the voice matches your brand. Use names and details from your company notes, avoid promises the team cannot keep, and keep answers short enough for a phone conversation.

4. Test before going live

Save changes before forwarding live calls. Run several test calls that cover booking, pricing, availability, an angry caller, and a question not in your knowledge document. Check the summaries and refine anything that sounds wrong or that promises work you cannot fulfill.

5. Review after go-live

After go-live, check Inbox call transcripts and summaries, and review Handoffs when the AI asks a human to take over. Tighten knowledge or rules whenever answers drift, callers repeatedly ask the same clarifying question, or your pricing and availability change.

Best practices:

  • Keep the knowledge document in plain language with current prices and policies.
  • Put the most common questions near the top so the AI finds them quickly.
  • Use tests on a quiet line before opening the full call volume.
  • Review at least the first few live calls and each new escalation.
  • Update company hours, holiday rules, and service area at the same time you change schedules.

Write knowledge in answer-shaped blocks

Organize the knowledge document as short answer blocks rather than a long wall of text. A block should begin with the question the caller is likely to ask, then give the answer, then state the exception. For example, ask about minimum visits, answer with the minimum and rate, and add that special pricing requires a human quote.

Set up a realistic escalation test

Use a test caller to walk through the handoff path before any live call: the AI should identify the trigger, stop negotiating, collect the callback details, and transfer or notify the right person. Verify the notification reaches the team member who can act, not only the account owner.

Keep a weekly tuning routine

Once live, schedule a short weekly review: read the newest call summaries, note questions the AI could not answer, check any handoffs for missed context, and make one update to knowledge or escalation settings. Small weekly fixes keep the call quality high without turning review into a project.

Write example knowledge blocks

A balanced knowledge document answers the questions callers actually ask. Start with service menu, availability, pricing guidance, travel or minimums, and what to do outside business hours. Add separate blocks for cancellation, pets, supplies, property access, and jobs that need a human quote. Write each answer as a short paragraph the same way you would say it on a call, then add a one-line exception when the rule does not always apply.

Launch checklist before forwarding live calls

  • Confirm the AI Receptionist add-on or receptionist plan is active.
  • Review the greeting and capture flow in a test call.
  • Ask a teammate to make a booking request from the caller side.
  • Verify the request reaches the dashboard and creates the right records.
  • Check that escalation names, phone numbers, and notification preferences are current.
  • Save a copy of the launch version in case you need to compare answers later.

Review the first live answers as a team

Do not judge the system only from the dashboard. Listen to or read the first live calls and compare the AI's wording with the policy you intended. If a caller has to repeat themselves, the greeting or question order needs a small change before more traffic arrives.

Create a one-page operating brief

Write one short page that contains the greeting style, out-of-hours rule, the three most common services with pricing guidance, the handoff trigger, and the person who handles each handoff. Keep that brief next to the AI Secretary settings and update it whenever the business changes. It gives a second person a fast way to understand the intended behavior without reading the full knowledge document.

Keep a tuning log

Record each tuning decision in a simple log: date, caller question, what the AI said, what it should have said, and the knowledge or escalation change made. Include the test call result after the update. A small log makes future review faster and prevents the same correction from being made twice after a new person takes over.

Review one metric per week

Choose one simple number to review each week, such as calls that ended in a booking request, calls that needed a handoff, or repeat questions about the same policy. Review the calls behind that number and make one change if the pattern repeats. One metric keeps tuning useful without turning every call into a detailed report.

Example knowledge block: minimum booking

Write common answers as blocks that start with the caller's question. For a minimum booking, the block might say: for a standard weekly clean, the minimum is one hour and pricing starts at the rate shown for the service; smaller homes can combine bathroom, kitchen, and light tidying in one visit; a customer who wants only a single room or a two-hour deep clean should receive the deep clean rates. The answer is short enough to say on a call and clear enough for the AI to repeat without negotiating an unapproved price.

Build an escalation matrix

List each handoff type with three fields: trigger, action, and owner. A complaint about damage triggers stop the conversation, collect the caller's name and property, and send the owner a summary. A refund or pricing dispute triggers a handoff to the person who can approve a change. A caller asking about insurance or liability triggers a polite reply that a human must answer. Keep the owner names and phone numbers current so the handoff notification leads to action instead of another search.

Test the AI Secretary with realistic calls

  • A standard booking request with a date that is available.
  • A booking request on a day or in a neighborhood the company does not serve.
  • A pricing question using a service that is not in the knowledge document.
  • A caller who asks for a refund, a complaint, or a damage claim.
  • A caller who wants to repeat an existing weekly booking.
  • A call after hours with no one available to confirm immediately.

For each scenario, write down what the AI should say, what record or notification should be created, and who should follow up. A test passes only when all three parts match the intended workflow.

Common tuning mistakes

  • Adding many policies at once without testing whether the AI can distinguish them.
  • Writing knowledge in dense paragraphs instead of answer-shaped blocks.
  • Keeping old prices, hours, or service names in the document after a real change.
  • Creating escalation rules without current owner names or notification targets.
  • Reviewing only dashboard summaries instead of comparing the AI's wording with the approved policy.

Run a monthly AI Secretary refresh

At the end of each month, open the knowledge document, the escalation rules, and the first page of call summaries together. Update prices and hours first, then remove any question block that no caller asked, and add one block for a question that repeated. Confirm that the owner names in escalation rules are still the people who can act. A monthly refresh keeps the AI aligned with the business without waiting for a customer to expose a stale answer.

Turn this guide into a working step

A help article becomes useful when you apply it to a real part of your operation. Open the matching dashboard page, make one small change, and verify the effect before moving to the next setting.

  1. Open the dashboard section named in the guide and compare what you see with the steps above.
  2. Use a realistic example from your business, such as one customer, one service, or one upcoming week.
  3. Ask a teammate to complete the same step and confirm they understand the result.
  4. Check notifications or the linked record afterward so the change is reflected where the team works.

If something looks different in your account

The exact label or placement can change when features are added or refined. Start by checking your plan, because some settings, channels, add-ons, and AI tools are only available after they are enabled for your company.

  • Confirm the feature is active on your plan or add-on.
  • Check whether the setting should be changed at company level, user level, or service level.
  • Refresh the page and reproduce the issue once with a recent browser.
  • Note the exact page, setting, and expected outcome before contacting support.

What to include when you need help

Write down the goal you wanted to accomplish, the steps you tried, the message or behavior you saw, and whether the problem affects one customer, one user, or the whole company. That context lets support connect the issue to the right setting instead of starting a generic investigation.

Continue exploring

AI Receptionist

Set up 24/7 inbound call answering, lead capture, and booking requests.

Can't find what you're looking for?

Contact Support FAQ