Enrichment That Doesn't Poison the CRM
Clean and de-dupe the list before you spend a credit, keep the raw value beside the enriched one, and never overwrite a human's field.
What it is for
The order of operations for enriching a scraped lead list with Bardeen and pushing it into a CRM without corrupting it. Use it whenever you are about to enrich more than a handful of rows, especially if the destination is a shared CRM other people trust. It keeps credits from being spent on junk rows, keeps machine-written values distinguishable from human ones, and forces you to measure the error rate on ten known records before it reaches a thousand unknown ones.
The skill
Enrich last, not first. Every bad row you remove beforehand is credits you keep and a bad record the CRM never sees. Order of operations: 1. Collect the raw list with the source URL on every row. 2. Drop rows with no company domain and no profile URL. They cannot be enriched, only guessed at. 3. Check the remaining rows against the CRM by domain or email. Accounts you already have do not need re-enriching. 4. Only now spend anything on enrichment. Rules for writing back: - Put enriched values in their own columns and keep the raw scraped value beside them. When a record looks wrong you need to see which layer invented it. - Never overwrite a field a human filled in. Write to an empty field, or to a parallel "suggested" field somebody reviews. - Stamp every enriched row with its source and the date. A contact detail with no provenance is a rumour. - Send anything the enrichment flagged as uncertain to a review list, not straight into the CRM. Prove the error rate before you scale: - Take ten rows whose answers you already know and enrich only those. - Count how many come back wrong. That number is your error rate for the other nine hundred, and you will not get a better estimate later. - Push those ten to the CRM and open one record by hand. The failure this prevents: a thousand contacts with plausible, unverifiable email addresses, a bounce rate that gets your sending domain flagged, and no way to tell which records were machine-written.
How to install it
1. Open the tool and start a new conversation, project or template. 2. Paste the text below wherever the tool keeps standing instructions — usually called system prompt, instructions, or persona. 3. Save it so it applies to every run rather than a single message.