Keep a Dealboard Full Without Adding Headcount

The job nobody wants to do by hand

A buy side team lives or dies on deal flow. Someone has to find businesses for sale, read each listing, throw out the junk, and put the good ones in front of the right person. Do it by hand and it eats a full day every week. Hire for it and you are paying a salary to run searches.

The team I built this for wanted the pipeline full without putting a person on search duty. That is a multiplier problem, not a hiring problem. The goal was simple: more good records landing in front of the right person, with no extra seat to pay for.

Why a scraper was the wrong tool

First instinct is to point a script at the listing sites and pull everything down. That breaks fast. A lot of the sources worth watching do not play well with automated tools. Some block them outright, some hide the real detail behind a live browser session, and some only surface their best results through a search box, not a clean page you can read straight off.

So instead of fighting that, I had an agent use the sites the way a person does. It opens a real session, runs a real search, and reads what comes back.

Why I have Claude drive ChatGPT Pro

Here is the part that surprises people. I run two AI tools together on purpose. Claude Cowork takes control of the desktop and runs the search inside ChatGPT Pro, because for this specific job ChatGPT Pro returned fuller and more current listings than anything else I tried. I wrote up how I pick between the two in Claude vs ChatGPT, and the same lesson held here: use the tool that wins the task, and let the other one run the workflow around it.

So Cowork is the operator. It opens the tool, runs the search with the company's buy criteria, and pulls back the raw listings. ChatGPT Pro is just the search step sitting inside that flow. Neither tool does the whole job. The win is in how they hand off to each other.

The workflow, start to finish

Step by step, here is what runs:

1. Cowork opens ChatGPT Pro on the desktop and runs the search using the company's buy box: industry, location, revenue range, asking price, and anything else they care about.
2. It collects the listings that come back into one place.
3. It validates each one. Stale posts, listings with no way to reach the seller, and anything outside the buy box get dropped.
4. It dedupes. The same business shows up across three or four sources with slightly different wording, so one real deal can look like four. Cowork collapses those into a single record.
5. It writes the clean records into the CRM dealboard and tags each one so the right person picks it up.

That last step is the whole point. The team opens their dealboard and the work is already sorted and assigned. Nobody searched. Nobody cleaned a spreadsheet. The validate and dedupe rules are wrapped as a repeatable skill so they run the same way every time. If that idea is new to you, start with what a Claude skill is.

What went wrong, and the fix

The first version skipped validation and dedupe and went straight from search to CRM. That was a mistake. The board filled up fast, but half of it was noise: the same listing four times, posts already under contract, listings with no way to reach the seller. The team stopped trusting the board, which is the worst thing that can happen to a pipeline.

The fix was to treat the cleanup as the real product, not an afterthought. Validate before anything lands. Dedupe on the business itself, not the listing text, because the wording changes from site to site. Only write a record once it is something a person can actually act on. A smaller board people trust beats a full board people ignore.

What this actually multiplies

This is the part I care about. One operator now covers what used to need a person on search duty. The same setup handles the admin work nobody enjoys: finding, reading, sorting, and routing. Set it once and it runs on a schedule.

The cost angle is the proof, not the pitch. A setup like this runs for less than a part time VA, and it does the boring part of the job every day without being asked. If you run solo and want a softer place to start, these five Claude skills are a good first build before something this involved.