Two hours back on every report, and the client keeps going

Experts are the bottleneck, not demand

The firm's appraisers are specialists. A commercial property appraisal takes subject matter knowledge of building types, floor plans, building age and engineering detail. It is not work the firm can hand to just anyone, and there is nowhere to hire more appraisers from. There is more work at the door than people to do it.

Part of every appraisal is clerical. The appraiser digs data out of the source documents and copies it into the platform by hand.

The documents are PDFs, brochures and spreadsheets, up to ten per appraisal and thirty to a hundred pages each. Every office formats them its own way. That step took about two and a half hours per report.

The workspace became the way in

First Line Software builds the firm's platform. Its AI-native stream built the extraction, and the agent-to-agent data imports case describes how two people and the agent built it. This case is about what it did on the client's side.

The entry point is an off-the-shelf agentic workspace on the appraiser's own desk, connected to the platform over MCP. The appraiser loads the documents there. The agent extracts the data, runs the quality checks, enriches the data and imports the result into the platform. The classic web interface stays as a second way in.

Every appraiser in the network got the workspace and the training to use it. That makes it infrastructure on every desk, not a tool for a few enthusiasts.

Telemetry instead of surveys

The rollout carries its own feedback loop. When an import fails, the delivery team sees it on its side without the user filing anything. It fixes the cause or reaches out to the user. The signal comes from real usage, not from asking people how it went.

13.5% of a report, 16% of capacity

Measured on the tool, the data entry step fell from 2 hours 30 minutes to 20 minutes. That returns 2 hours 10 minutes per report. A whole report takes about 16 hours on average. So the saving is 13.5% of every report.

Capacity grows by more than the time saved, because the base shrinks. The same appraisers now need about 13 hours 50 minutes per report, which lifts the network's capacity by about 16%. In the terms of the RACE triangle, that is more scope on the same headcount.

Because appraisers are the constraint and demand is not, freed hours become more appraisals rather than idle time. That is why the 16% reads as a revenue effect and not a cost saving. It is First Line's estimate, and the client has not yet confirmed it in revenue terms.

The first point, not the last

Classic automation ships a feature, and the effect stops there. Here the workspace stays on the desk after the feature ships, and the appraisers keep using it for other work. One of them has already built a financial classifier for their own work. First Line wrote none of it.

That makes the 13.5% the first point on a curve. The method behind it has four stages:

1. Get onto the desk. An agentic workspace on every desk, with the training to use it.
2. Work from the desk. Forward deployed specialists sit beside the most expert and AI-fluent people in each line of the business. New practices start there.
3. Share what works. What people built for themselves is put into a shape that travels, checked against policy, and adopted across the network.
4. Promote what gets used. Usage picks the candidates, and the business case decides. What wins becomes a platform service, and the local workaround retires.

At this client, stage one is done, and appraisers building their own tools is already happening. Pairing with the strongest experts, sharing and promoting are the plan, not a result. The short form: teams invent, the organization adopts, the platform follows.

Why this is a RACE Programming case

The delivery side follows the framework. The extraction was built by a Pit Wall of two, an AI Product and a Forward Deployed Engineer, with the agent writing the code.

The client side applies the same stance to the client's own people. Appraisers get capability on their desk, the platform keeps the gates, and the plan is for the next platform feature to come from observed use rather than from whoever asks loudest.