The brief
DBRE Property took pride in the depth of their appraisal reports. Every prospect got a detailed document — comparable sales analysis, suburb-level context, photography notes, recommendation narrative. The team ran four of them a month. Five hours each, minimum. Often more.
That pride became a bottleneck.
- Prospects waited weeks to receive an appraisal.
- Agents pushed off the work because each one took half a day.
- The team could not scale appraisal volume without breaking the quality bar.
The CEO James Davie was clear about the constraint: DBRE was not willing to lower the quality of the reports to ship faster. The work was too important to the brand. The hours, however, were unsustainable. If the team wanted to double appraisal volume — which the local market was demanding — they had to find a way to keep the quality and shed the time.
What we found in the audit
The audit week (week one of the trial) mapped how the appraisal report actually got written. The pattern was consistent across the team.
Every appraisal followed the same structure. Same sections. Same depth in each section. Same recommendation logic. Same tone. The reports were not bespoke artefacts — they were structured documents with substantial content that an experienced agent assembled by hand, from a finite set of inputs, every single time.
The audit surfaced three facts:
- The structure of the report was completely consistent across the team’s previous 80+ appraisals
- The inputs the agent collected at the property visit were also consistent — property type, condition, comparable sales, suburb context
- The “expensive” part of the five-hour build was not the thinking, it was the writing
That third point unlocked the build. The agent did the analysis at the property. The five hours of follow-up was almost entirely transcription, formatting, and document assembly. Work that AI is good at, given the right training data.
What we built
We trained an AI model on DBRE’s previous appraisal reports. The language, the structure, the comparable sales logic, the formatting conventions, the recommendation framing — all of it. Then we wrapped the model in a simple form.
Staff fill in the property details from the visit. Address, property type, condition notes, comparable sales references, recommendation. The model produces a full appraisal report in DBRE’s voice, with their structure, their formatting, and their analytical depth. A human reviews it. Done.
Then we made it mobile-friendly. Agents fill in the form from the car after the property visit. The report is ready by the time they get back to the office.
The architecture, on a page:
- Agent finishes the property visit
- Agent opens the appraisal form on their phone in the car
- Agent fills in property details, condition, and comparable sales
- The AI model produces a complete appraisal in DBRE’s voice
- Agent reviews and tweaks the draft back at the office (15 minutes)
- Final report sent to the prospect the same day
The stack
- OpenAI ChatGPT as the core model, prompt-engineered on DBRE’s historical reports
- A custom form interface (web app) built for mobile-first input
- Google Workspace for final report storage and prospect delivery
- Make for the orchestration between form submission, AI generation, and document creation
Total monthly platform cost: under $250 AUD.
The results
- 80% time saved. From 20 hours per month to 1 hour.
- Built in 14 days. From kickoff to a tool the team uses every week.
- Wait time collapsed. Prospects went from weeks to hours between visit and report.
- Quality held. Reports only need light review before going out.
- Mobile-friendly. Agents fill in the form from the car. The report is ready by the time they get back to the office.
- Appraisal volume capacity doubled at the same headcount.
What we learned shipping it
The first version produced reports that were technically correct but slightly off in tone. DBRE’s previous reports had a particular voice — confident without being salesy, detailed without being long-winded. Our first AI drafts hit the structure but missed the voice.
We fixed it by adding 15 additional anchor reports as voice examples in the prompt, with explicit notes on what made the tone distinctive. The next round of drafts came back in voice. The CEO signed off.
The second issue was edge cases. Properties with unusual features — heritage listings, mixed-use, off-market sales — produced drafts that needed more rework. We added an “exception path” in the form where the agent could flag unusual features upfront, which triggered a different prompt template. After that, even the edge cases came back close to ship-ready.
What changed for the team
DBRE stopped trading speed for quality. They got both.
The agents who used to dread appraisal weeks now run more of them. The growth bottleneck became a growth lever. The team has the operational headroom to chase listings the old workflow could not support.
“OFO Collective and their team helped turn a regular 5 hour report into a 15 minute task. It’s an exciting future ahead.” — James Davie, CEO, DBRE Property
Why this build matters for service businesses with structured outputs
DBRE’s appraisal was a perfect AI use case and most service businesses have at least one. Look for the pattern:
- The output has a consistent structure
- The inputs are finite and capturable in a form
- The “expensive” work is writing, not thinking
- You have a body of historical outputs to train on
If that describes a deliverable in your business — appraisal, audit report, proposal, brief, recommendation, plan — there is probably a 14-day build that takes 80% of the hours out without touching the quality.
OFO Collective ships this kind of build on the 30-day trial. For a similar engagement, book a call.