How to Prepare an Agency Knowledge Base for AI Property-Inquiry Follow-Up

An AI assistant that follows up on property inquiries can only state what your agency has approved it to state. Before you configure AI lead follow-up, prepare a knowledge base: one approved record of the facts the assistant may use, with the fast-changing ones marked and owned. Five steps get you there, and all five are agency work rather than software work. TechTribe's Real Estate Platform configures your details, property types and messaging style during onboarding, but the accuracy of the underlying facts stays with you.

TL;DR: A property-inquiry knowledge base is the approved set of facts an AI assistant may use with buyers and tenants. Prepare it in five steps: inventory your existing sources, split facts into "state automatically" and "confirm with an agent", put an effective date on anything that changes fast, name one owner per category, and test with real past inquiries. Contradiction and expiry review is manual and stays with the agency.

Step 1: inventory the sources your agents already answer from

If you have not yet mapped how an inquiry moves through your agency, start with the inquiry-to-agent workflow and come back. The knowledge base slots into that map; it does not replace it.

List every place an agent currently finds an answer: listing sheets and mandate files, the WhatsApp threads where agents answer buyers directly, office policy documents, the agent roster with coverage areas, your viewing procedure, and the documents you require before accepting an offer. For each source, record three things: which facts it holds, whether those facts are current, and who maintains it.

Expect duplicates. Most agencies discover the same fact living in several places with different answers, and the inventory is what exposes that before the assistant does.

Step 2: split facts into "state automatically" and "confirm with an agent"

Not every approved fact should leave the assistant without a human in the loop. Sort your inventory into two columns.

This split is TechTribe's recommended starting point, not fixed product behaviour; like your fallback rules, it is set to your agency's decisions. Keep the left column aligned with the questions that qualify a buyer before a viewing, because asking them is most of the assistant's job between first contact and agent handover. The mechanics of how an assistant grounds its answers in approved documents are a separate topic; this article stays on the preparation work.

Step 3: give fast-changing facts an effective date

Availability, asking prices, listing status and agent assignments change faster than any review cycle. Two rules keep them safe.

First, each volatile fact gets one named source of truth. Prices live in the listing register, not in an agent's memory or a months-old WhatsApp message. When two sources disagree, the named one wins and the other gets corrected.

Second, each volatile entry carries an effective date: the date someone last confirmed it. A modelled example, not a real listing: "Three-bedroom house, Greendale, Harare. Asking price confirmed by the listing agent on 14 August 2026." If a buyer asks after that confirmation has lapsed, the safe response is that the agent will confirm the current price. A deferred answer costs the buyer a short wait. A stale price quoted as current costs trust, and sometimes the sale.

Step 4: name an owner and a review rhythm for every category

An unowned fact decays. Assign one approver per category: the principal signs off policy answers, the listings manager signs off listing facts. Assign one named person to make updates, so nobody assumes someone else did it.

Before launch, run one full manual review pass to find contradictions, gaps and expired entries. After launch, set a recurring review: listing facts whenever a listing changes, plus a standing cycle set to how often your listing register actually moves, and policy facts whenever the policy changes.

Plan this review as a human process and do not rely on software to catch the problems for you. If the knowledge base says a property is available and it sold last week, the assistant will not know until a person corrects the record.

Step 5: test with real inquiries before going live

Pull real past inquiries from your WhatsApp threads and inbox, word for word, spelling mistakes included. For each one, write the answer the assistant should give using only the knowledge base. Three outcomes matter:

• The knowledge base answers it correctly, and nothing needs changing.
• The knowledge base cannot answer it, and it should escalate. Check that behaviour against your fallback rules, which govern what the assistant does at runtime when approved information runs out.
• The knowledge base cannot answer it, but an agent answers it daily. That is a gap. Add the fact, get it approved, and retest.

Include questions the assistant must refuse outright, so you can confirm it refuses them.

Before you launch: what the assistant must never answer

Write the prohibited list into the same document and give it principal-level approval. At minimum:

• Legal or tax advice of any kind.
• Prices or availability an agent has not confirmed.
• The identity or circumstances of a property owner.
• Any screening of buyers by ethnicity, religion, nationality or similar grounds.
• Commitments your agency has not approved, such as holding a property or accepting an offer.

Every future update to the knowledge base gets checked against this list.

You are ready to configure when

• Every information category has one named approver and one named updater, and both people know it.
• Every fact in the "state automatically" column has been approved by the person who owns that category.
• Every fast-changing fact has a single named source of truth and a date someone last confirmed it.
• The prohibited list is written down and signed off at principal level.
• Your real past inquiries have been run through the knowledge base, and every gap they exposed has been filled and retested.

If a line is still open, close it before onboarding rather than after. Configuring an assistant on facts nobody owns moves the problem rather than solving it.

What onboarding covers, and what stays with you

During Real Estate Platform onboarding, TechTribe configures the assistant with your agency's details, property types and preferred messaging style. The assistant is built to respond to inquiries in seconds, qualify budget, timeline and location preferences, and route each lead to an agent with a summary, working around the clock across your website forms and WhatsApp. For a Zimbabwean agency, that configuration is only as good as the facts behind it, and the five steps above are how you get those facts right.

The Real Estate Platform is $350 setup plus $75 per month; the pricing page sets out how that compares with the once-off Real Estate Website. If the knowledge base work is done, book a demo and bring your hardest past inquiry with you.