Scheduled Market Comp Tasks for Lenders: Six Markets, Every Monday, and a Failed First Test
By Alex Regueiro · October 25, 2026 · 5 min read
The first test of the AI deal screen came back with significantly wrong output, including incorrect budget amounts and stale comps. For a private lender, a screen that gets the budget wrong or leans on old comps is worse than no screen, because people trust the number on the page.
60-second check
Take the 2-minute AI readiness survey →
Talk it through
Book a call with the Reshape team →
The client is an executive at a private real estate lender. The deal screen is the part of our underwriting build the lender uses to check a deal quickly. If you run lending for investors, our page for mortgage loan officers and private lenders shows the wider work.
The situation
Comps are the part of underwriting that spoils fastest. A deal screen can have perfect scoring logic and still mislead if the sales it compares against are old. On this project the first test showed exactly that. The outputs were off in two ways: the budget amounts were wrong and the comps were stale.
The fix had two parts. Comps needed a refresh that does not depend on anyone remembering to run it, and the comp logic and budget parsing needed a review against real outputs.
What was broken
The failure was visible in two specific places:
- Budget amounts in the screen's output did not match the deal.
- Comps in the output were stale, so the comparison was against old market data.
- Taken together, the screen's output was significantly inaccurate on its first run.
What we built
We set the lender's deal screen up as a live artifact inside Claude Desktop and put the comp refresh on a calendar.
- Six market-specific scheduled tasks. Each task covers one market, so each market's comps can be refreshed and checked on their own.
- A fixed run time. All 6 tasks run every Monday at 8 AM EST, so the week starts with fresh comps in front of the lender.
- A review session for the first automated outputs. We booked a follow-up working session to look at what the scheduled tasks actually produce, and to fine-tune the comp logic and the budget parsing.
- A test assignment for the executive. The executive was asked to run the deal screen on a few deals and note every issue, so the fixes are driven by real deals and not by guesses.
What changed
No corrected output had been reported when this call ended, so we are not claiming the problems are solved. The change was the structure. Comps now refresh on a schedule that runs by itself every Monday, market by market, and the known errors went onto a defined fix list with a review session booked.
The first test did its job. It found two specific faults at the cost of one failed run.
Treat your first AI test as a bug report
If you are testing an AI tool on your own deals, run it on a few real files and write down every miss. Check the figures the tool read from your documents against the originals. Check the date on every comp. Then fix the data feed before you tune the scoring.
Also split scheduled jobs by market or product line. One job per market makes failures easy to find, and a bad run in one place does not hide in a combined report.
Frequently asked questions
Why would an AI deal screen show stale comps?
Usually because the market data it relies on was loaded once and never refreshed. In this build the answer was to schedule a refresh every Monday, so the comps carry a current date when a deal is screened.
Why run one scheduled task per market?
Each market has its own comps, so separate tasks keep each refresh independent. In this build there are 6 tasks, one for each market, which makes a failure in one market easy to spot.
What does budget parsing mean in an underwriting tool?
It is how the tool reads the budget figures out of a deal's documents. In this project the first test returned wrong budget amounts, so budget parsing went on the fix list alongside the comp logic for review once the first automated outputs came in.
Is a failed first test a bad sign for an AI build?
Not by itself. It is a normal and useful result if you catch it on test deals. Here the first test exposed wrong budgets and stale comps early, and the response was a scheduled refresh plus a review session to fix the logic.
If your market data or comps go stale between deals, book a 30-minute AI follow-up and reporting audit. We look at how your data refreshes today and what should run on a schedule.