The problem this solves
Product data ages faster than company data. A price that was right last quarter is wrong now, a line that was in stock is on back order, and a retailer that carried it has dropped it. Anyone who sells into retail, builds comparison content, or watches a competitor catalogue is re-checking the same facts by hand on a schedule, and the checking is the job rather than the thinking.
None of this lives in a business database. Company records describe the company, the stack it runs and the people who work there. What something costs today, whether you can buy it, and who lists it are facts on a handful of retail pages, and they are different pages for every product. That is why product columns tend to be the ones left empty on an otherwise enriched sheet.
The reason those columns matter is that they are the ones a person will question. Somebody reads a figure, disagrees, and the whole sheet loses authority unless the number can be traced. A price without a source is an assertion, and an assertion is worth arguing about.
How the mission runs
- Name the columns as questions. Each column you want is written as the question it answers: what is the current list price, is it in stock, which retailers carry it. The Enrichment Agent fills a schema you define, so the columns you get back are the columns you asked for rather than whatever a vendor happens to publish.
- Research every row against the open web. Each product on list <name> is researched in turn, reading the manufacturer page and the retail pages that carry it. Rows are handled together rather than one after another, which is what keeps a long list to a sensible wait instead of a multiplied one.
- Record the answer with its source. Every value arrives with the page it came from, an extract of the wording that supports it, and a confidence. That is the evidence policy the whole platform runs on: the question comes from you, the answer comes with its receipt.
- Leave a blank where nothing is published. When a price is not public or a product has no traceable stockist, the cell stays empty and the confidence says so. An honest gap is workable, and it is far cheaper than a plausible number somebody has to disprove later.
- Export the enriched list. The finished list goes to Google Sheets with the research columns and their source links alongside the original rows, so the sheet can be read, sorted and challenged without going back to the mission.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
The original list with one column per question, each cell carrying a value, a confidence and a link to the page it came from, exported to Google Sheets and kept on the list itself so the next run updates rather than rebuilds.
Make it yours
- Ask for a competitor price alongside your own to turn the same run into a positioning sheet.
- Add a question about shipping terms, warranty length or bundle contents when the purchase decision turns on something other than headline price.
- Run it weekly on a schedule and keep the previous values, so a price change becomes a signal rather than a surprise.
- Point it at a list of company domains instead of products to research pricing models rather than individual prices.
Frequently asked questions
How current are the prices?
They are read from live pages at the moment the mission runs, so they are as current as the page. The source link on each cell tells you exactly which page, which is what lets you judge whether a figure is a list price or a promotion.
What if two retailers disagree on price?
The answer records what was found and where, so a spread between stockists shows up rather than being flattened into one number. Asking for the stockists column alongside the price is what makes that visible.
Does this work for products without a public price?
It returns an honest blank with a low confidence. For categories sold by quotation, the useful questions are usually about availability, distribution and packaging rather than price, and those can be asked in the same run.
How is this billed?
Research is priced per row of the list, and the run reports what it spent. Rows that already carry an answer are skipped on a repeat run, so keeping a list current costs less than building it the first time.