A six-person Nottingham residential estate agency was losing 28 hours every week to property listing prep, sales progression chasing, and applicant follow-up. That admin was capping the team at 35 active sales at any one time — while a growing pipeline of valuation appointments sat waiting to convert. Three AI agents changed the maths: 62% more sales progressed simultaneously, 24 hours recovered weekly, £85 a month to run.
The Admin Weight Inside a UK Residential Sales Agency
Anchor Property, a six-person independent residential estate agency in Nottingham, had the market position of a firm twice its size. Strong local brand, a high valuation-to-instruction conversion rate, and a consistent stream of buyer enquiries from their Rightmove and OnTheMarket profiles. The pipeline was there. The hours to work it weren't.
In a residential sales agency, there are three distinct admin loads — and all three compound as the active sales list grows.
The first is listing preparation. Every new instruction requires a full property description: Rightmove-ready copy, a key features list, and a detailed room-by-room narrative. The firm's negotiators spent an average of 90 minutes per new instruction writing copy from their valuation notes. At 12–15 new instructions per month, that's 18–22 hours of writing time that required the judgement of a senior negotiator but the skills of a content writer. The copy was good. The time cost was not.
The second is sales progression. Once a property goes under offer — Sold Subject to Contract in estate agent language — the real admin begins. The solicitor's chain needs coordinating. Searches need chasing. Survey results need communicating. Mortgage offers need confirming. Each sale in progress requires an average of four to six touchpoints per week across buyers, sellers, and their respective solicitors. With 35 active sales at any one time, that was 140–210 chasing contacts per week — a constant, low-value communication task that was consuming two of the team's six working days.
The third is applicant follow-up. When someone views a property and doesn't offer, they don't disappear from the market — they're still active buyers, likely to transact in three to six months. Keeping them warm with relevant new instructions, circling back when a price reduction makes a previously dismissed property relevant, re-engaging after a sale falls through: all of it required remembering individual applicant preferences and acting on them. In practice, most fell through the cracks.
Total weekly admin drain: 28 hours. Thirty-five active sales was the ceiling — not from lack of stock, but because the admin surrounding each sale was consuming the time the negotiators needed to list more properties and convert more viewings.
We were turning away valuations. Not because we didn't want the business — because taking on another instruction meant another 90 minutes of listing copy and another six or seven progression contacts a week on top of everything else. At some point the maths stops working, and ours stopped working at 35 sales.
— Director, Anchor Property, Nottingham
The Three Agents They Built
The build took four weeks from scoping to all three agents running in production. The work was done in n8n, hosted on a low-cost VPS, using Claude for listing copy and progression summaries. No new software was purchased — all three agents connect to the tools Anchor was already using: their estate agency CRM (Jupix), Gmail, and Google Drive for property files.
Agent 1: The Property Listing Agent. After a valuation visit, the negotiator completes a structured property notes form — a Google Form built around their existing valuation checklist, covering room dimensions, features, condition, and any selling points they want to highlight. When the form is submitted, the Listing Agent reads the structured input and generates a full Rightmove-ready listing: a headline description, a key features list formatted for Rightmove's five-bullet display, and a complete room-by-room narrative. The output matches Anchor's house style — active voice, punchy opening sentence, factual and specific rather than generic. The negotiator reviews the draft, makes any edits, and approves it for upload. Average review time: twelve minutes. Previously: 90 minutes of writing from scratch.
Agent 2: The Sales Progression Agent. This agent monitors every sale in Jupix that has passed the SSTC milestone. It maintains a structured tracker for each sale — current stage, outstanding actions, last contact date with each party, and any flags raised by the negotiator. Every Monday it generates a prioritised progression plan for the week: which sales need a solicitor chase, which buyers need a mortgage update request, which sellers need a chain position update. It also sends automated holding messages on behalf of the negotiator at defined intervals — keeping buyers and sellers informed between active contacts without requiring the negotiator to compose each message manually. For complex chain situations requiring judgement, it flags the case for direct negotiator attention rather than attempting to manage it autonomously.
Agent 3: The Applicant Engagement Agent. Every viewing that doesn't result in an offer feeds this agent. The negotiator logs viewing feedback using a structured form — interested but not ready, price too high, property too small, preferred a different road. The agent maintains a live preference profile for each applicant. When a new instruction comes in, it automatically matches against the preference database and sends personalised alerts to relevant applicants. When a price reduction happens, it re-contacts all applicants who viewed at the higher price. When a sale falls through, it re-activates the original applicant pool. The applicant who viewed six months ago and noted "if the price drops to £250k" gets a contact prompt automatically when the price hits that figure.
What Each Agent Does, Step by Step
All three agents sit inside a single n8n instance. They share connections to Jupix via its API, Gmail via OAuth, and Google Drive for document storage. LLM calls go to Claude via the Anthropic API. Each agent is a separate n8n workflow — they don't interfere with each other and can run simultaneously across multiple instructions, sales, and applicants at once.
The Property Listing Agent is triggered by a new Google Form submission. It runs a structured generation prompt that takes the form data as input and produces the listing copy against a detailed system prompt encoding Anchor's house style. That system prompt includes examples of their best-performing listings — properties that generated above-average enquiry-per-view rates on Rightmove — so the output reflects what actually converts browsers to enquiries, not just what sounds professional. The draft is saved to Google Drive and a Slack notification goes to the instruction negotiator with a review link.
The Sales Progression Agent runs as a scheduled workflow every Monday morning and Thursday afternoon. It reads the current status of all SSTC sales from Jupix, calculates days since last contact for each party, identifies which milestones are overdue, and generates the weekly progression plan. For automated holding messages — the "just wanted to keep you updated" contacts — it drafts the message and queues it in Gmail with a 30-minute send delay, giving the negotiator a window to review or recall before delivery. Nothing sends autonomously without that review window on a first contact in a new sale.
The Applicant Engagement Agent triggers on three events: new instruction added to Jupix, price reduction saved, or sale status updated to fallen through. On each trigger, it runs the match query against the preference database, generates personalised match alerts for relevant applicants, and queues them for send. Alerts are written to the individual applicant — "you viewed 14 Maple Road in March and noted the garden wasn't large enough; we've just listed 22 Maple Road which has a 90-foot south-facing rear garden" — rather than generic new listing notifications. That specificity is what drives re-engagement.
For the broader principles behind testing production-ready agent systems like this before going live with client-facing communications, the AI agent evaluation framework covers the testing approach used. For the observability layer that keeps the progression agent's weekly plan accurate when CRM data is incomplete, the AI agent observability guide covers the monitoring setup that catches stale or missing data before the plan runs.
The Results After 90 Days
Ninety days after all three agents went live, the numbers were consistent enough to report:
- Active sales: 35 → 57. A 63% increase in simultaneous sales progressed. The additional capacity came from recovered negotiator time — the team was no longer spending two full days a week on progression contacts and listing copy.
- 24 hours recovered per week. Listing copy time (18–22 hours per month), progression contact admin (10–12 hours per week), and manual applicant re-engagement (3–4 hours per week) now require around four hours of human oversight across all three agents combined.
- Listing copy time: 90 minutes → 12 minutes per instruction. The Listing Agent drafts are approved with minor edits in the majority of cases. The negotiators describe reviewing the drafts as "correcting, not writing." Rightmove enquiry-per-view rate is up 22% compared to the 12-month prior average.
- Applicant re-engagement rate: 11% → 34%. Of applicants who viewed and didn't offer, 34% are now re-engaged via the automated matching and re-contact system. Previously, re-engagement was manual and largely dependent on whether a negotiator remembered a specific applicant's stated preferences.
- Fall-through rate: 14% → 9%. The Sales Progression Agent's consistent weekly contact cadence caught three potential chain collapses before they happened — including one case where an automated prompt alerted the negotiator to a solicitor who had gone 12 days without a response, sufficient time to intervene and keep the chain intact.
The listing agent alone would have justified the build cost. Getting 90 minutes of a senior negotiator's time back on every instruction — and getting copy that converts better than what we were writing — is the kind of result you don't expect from an £85-a-month system. The progression and applicant agents are everything else on top of that.
— Director, Anchor Property, Nottingham
What It Actually Cost
The build was completed over four weeks. The running cost breaks down as:
- n8n self-hosted (VPS): £12/month
- Anthropic API (Claude) for listing copy, progression summaries, and applicant match alerts: £55–£70/month at current volume
- Google Workspace (already in use): £0 additional
- Total monthly running cost: approximately £85/month
The payback period on the build cost was four weeks. By that point, the additional instructions converted from the recovered valuation time had generated fee income exceeding the cost of building the system.
The strategic return is harder to put on a monthly P&L but more significant. UK residential property transactions are returning to pre-2022 volumes as base rate reductions improve buyer confidence. Independent agencies are competing with national chains deploying AI listing tools and progression software at scale. The negotiator time that Anchor previously spent on admin is now going into valuation appointments, vendor relationships, and buyer conversations — the work that actually determines market share on a competitive patch.
The same three-agent pattern applied to a very different property admin load appears in the letting agency case study — where the bottleneck was compliance documentation under the Renters Rights Act rather than listings and progression. The admin types are different; the underlying dynamic is identical. The compounding advantage post explains why building the system now — rather than waiting until the market fully recovers — is the timing that matters.
For the technical approach to keeping API costs flat as listing volume grows, the AI agent cost optimisation guide covers caching listing templates and routing simpler applicant alerts to cheaper models without sacrificing quality.
If you run a residential estate agency — independent or franchise, sales or lettings or both — and you want to see what a similar build would look like for your team size and active sales volume, book a free 30-minute call. We'll map your current admin load across listing prep, progression, and applicant management, identify the three highest-impact automation points, and give you a clear picture of what the build would cost and recover. The call is free.