The pipeline: Classify, Context, Reason, Execute

Munro processes every incoming event through a four-stage AI pipeline: classify the event by type and urgency, assemble full context about the property and people involved, reason about the best course of action using genuine AI judgment, then execute the action or escalate to the human agent.

1. Classify

Every incoming message, enquiry, portal event, and system trigger is classified by Munro's classifier agent. This runs at low temperature (high precision) to accurately categorise each event: is it a buyer enquiry, a solicitor update, a viewing request, a complaint, an offer, or something else entirely? The classification determines which specialist skill handles the event.

2. Context

Before reasoning, Munro assembles the full context. For a buyer enquiry about 14 Argyll Street, that means: the listing details, current status, asking price, all previous offers, the seller's preferences, every past communication with this buyer, the buyer's qualification score, and any relevant memories from previous interactions. Context assembly pulls from the listing, buyer pipeline, transaction history, communication log, and Munro's domain memory system.

3. Reason

With classification and context in hand, Munro reasons about what to do. This is genuine AI reasoning, not "if buyer enquiry then send template." Munro reads the tone of the message, understands the stage of the sale, considers the seller's communication preferences, and decides on the most appropriate response. If the situation is ambiguous or high-stakes, Munro decides to escalate rather than act.

4. Execute or Escalate

If Munro is confident in its decision, it executes: send the email, schedule the viewing, update the portal, notify the seller, chase the solicitor. If not, it escalates to you with a clear explanation of what happened, what it considered, and why it needs your judgment. Escalated items appear in the "Requires Your Action" section of the Command Centre.

Eight specialist skills

Munro has eight specialist skills that are loaded on demand based on event classification. Each skill contains domain-specific instructions, context requirements, and action capabilities tailored to a particular aspect of estate agency operations.

Seller communications

Manages all outbound communication with sellers: progress updates, viewing feedback, market activity reports, and proactive check-ins. Understands the seller's portal, communication preferences, and the current stage of their sale.

Buyer handling

Processes buyer enquiries, performs AI pre-qualification (budget, mortgage, chain position), schedules viewings, sends reminders, and collects post-viewing feedback. Matches buyers to listings based on requirements.

Offer management

Records offers, runs AI analysis (scoring 0-100), presents to sellers via their portal with strengths, concerns, and recommendations. Handles acceptance, counter-offers, and rejection communications. Manages closing date scenarios with multiple competing bids.

Listing quality assurance

Reviews listing completeness before go-live. Checks descriptions, photos, Home Report, pricing strategy, and all required details. Ensures compliance with portal requirements before syndication.

Command centre

Powers the AI chat in the Command Centre dashboard, WhatsApp, and Telegram. Answers questions about the business, provides summaries, and executes 75+ action types on behalf of the agent, from creating leads to recording offers to checking analytics.

Portal monitoring

Monitors Rightmove, Zoopla, and OnTheMarket sync status. Detects failed uploads, stale listings, and data mismatches. Retries failed syncs and escalates persistent issues.

Compliance

Monitors regulatory compliance: Home Report validity (12-week expiry), listing agreement status, tenancy compliance (deposit protection, gas safety, EICR), and communication audit requirements.

Inbox triage

Reads every incoming email, SMS, and portal message. Classifies by urgency, matches to the relevant property and contact, and routes to the appropriate skill for handling. The first line of defence that ensures nothing falls through the cracks.

Domain memory

Munro maintains a persistent memory system with temporal decay, giving it contextual awareness across every sale and letting in your agency. Memories are scored by importance and recency, so recent high-priority events surface first while old routine activity naturally fades.

When Munro processes a buyer enquiry, it does not just see the current message. It remembers that this buyer viewed two other properties last month, that they mentioned needing a home office, and that their mortgage was pre-approved. This context makes every interaction more intelligent.

Memory types include: activity records (what happened), insights (what Munro observed: buyer sentiment, seller urgency), profile data (contact preferences, budget), and preferences (communication channel, response style). Each type has a different half-life: activities fade after 14 days, insights persist for 30 days, profiles for 90 days.

Communication approval workflow

Every communication Munro sends can be routed through an approval queue before delivery, giving agents full control over what goes out under their name. Agents can approve, edit, or reject any proposed message.

By default, Munro's communications go through the approval queue. The agent sees proposed emails, SMS messages, and portal updates in the Communications dashboard, where they can approve them with one click, edit the content before sending, or reject them with feedback.

Once you trust Munro's judgment in specific areas, you can enable auto-send for routine communications while keeping approval required for high-stakes messages like offer presentations or solicitor correspondence.

You can also pause Munro on a per-listing basis. If you are handling a sensitive negotiation personally, toggle the Munro pause on that listing and all AI communications for that property are held until you unpause.

When Munro escalates

Munro is designed to know its limits. It escalates to the human agent whenever a situation requires genuine judgment, emotional intelligence beyond AI capability, regulatory decisions, or actions with significant financial consequences.

Common escalation triggers include:

Complaints or disputes that require human empathy
Unusual offer conditions that fall outside standard patterns
Seller requests that conflict with the agent's instructions
Portal sync failures that persist after 3 retries
Transactions stuck at the same missives milestone for too long
Compliance deadlines that require immediate agent attention
Any situation where Munro's confidence in the right action is below threshold

Escalated items appear in the "Requires Your Action" section of the Command Centre with full context: what happened, what Munro considered, and why it needs your judgment.

Multi-agent architecture

Munro uses a multi-agent routing system that sends each task to an optimally configured AI agent. Classification uses high-precision settings, creative tasks like description writing use higher temperature, and monitoring uses conservative settings to minimise false positives.

Five agent configurations handle different task types: the classifier (high precision for accurate routing), the inbox processor (balanced for communication drafting), the quality assurance agent (thorough for listing review), the command centre agent (creative for conversational chat), and the monitor (conservative for compliance checking).

This means Munro is not a single AI prompt trying to do everything. It is a coordinated team of specialised agents, each tuned for their specific job, working together through a shared context and memory system.

See the full platform

Munro powers all 16 modules of Saltire Systems. See the complete feature breakdown.

View Platform →

How Munro Works