Share certificates, emails and pasted notes read by AI into the right case record
Seven Make scenarios turn the unstructured side of a case into columns on a monday.com board: scanned share certificates read by OpenAI vision and written back by GraphQL, then checked against a reference table; emails to the accounts inbox matched to their case; and free-text notes pasted onto an item split into the right fields.
Where this applies
Probate and estate firms, asset tracers, claims handlers and any case-based team whose evidence arrives as scans, emails and notes.
- certificate reads and lookups, no errors
- 561
- GraphQL write per file read
- 1
- case boards with a paste-and-split column
- 4
The problem
Recovering unclaimed shares and dividends for an estate is document work. The evidence is a scanned certificate, a registrar's statement, an email from a solicitor, notes from a phone call. Each holds a few facts that matter (company, number of shares, holder, reference), and each has to be typed into the case record before anyone can act on it.
Typing is slow and it is where errors enter: a transposed reference sends a case the wrong way. AI can read the documents, but a model that writes prose into a notes field has only moved the problem. The output has to land in the right columns on the right item, in a fixed shape, every time.
What we built
Vision on the certificate, JSON out
When certificates or statements are uploaded to a case item, an iterator takes each file in turn, downloads it and passes it to OpenAI's image analysis. The model is asked for the holding details as JSON, which is parsed before anything is written, so a reply that is not the expected shape never reaches the board.
One mutation for many columns
The parsed fields are written back with a raw GraphQL mutation rather than a chain of column updates, and an ignore handler means one unreadable file does not stop the others in the same upload. This scenario replaced an earlier one that converted emailed attachments to images first.
Checked against a reference table
A second scenario fires when a share item changes, finds the share on the board and looks it up in a Google Sheets reference table, enriching the item from it. A share that is not in the table is skipped rather than failed.
The accounts inbox, matched to its case
About every hour, a third scenario reads new mail in the accounts inbox. Two AI extraction steps pull out the case reference and the facts that matter; the reference finds the item on the case board, which is updated. An email it cannot match goes to a person by email instead.
Paste a block, get the fields
On each of four case boards, a scenario watches a free-text column. When someone pastes a block of notes into it, an AI extraction step pulls out the structured fields and a four-way router writes each one to its column, depending on what was found.
The outcome
From 18 September to 16 October 2026 the certificate reader ran 251 times and the reference lookup 310 times, both without an error, using 1,890 and 1,234 credits. The firm's team switched both off in mid-October.
The accounts-inbox matcher is live and ran 472 times without an error from 18 September to 21 October, using 472 credits: at one credit a run, nearly all of those runs were checks that found no new mail. The four note-splitting scenarios are switched on but had no runs in the window.
Where it stops
The model's reading is written straight to the board: nothing in these scenarios asks a person to confirm a certificate before its figures are used. The certificate reader and the reference lookup are switched off, and the note splitters have not run recently. The inbox matcher relies on the case reference being in the email. Four similar note-splitting scenarios on other boards are not described.
Built with
- Make.com
- monday.com
- OpenAI
- Make AI Toolkit
- Google Sheets
- Gmail
- GraphQL API
Patterns used
- Vision to strict JSON
- Raw GraphQL write-back
- Reference-table enrichment
- Match or hand to a person
Page copy drafted with AI assistance. Skofi is accountable for it. Editorial policy
Have something shaped like this?
If any part of the above resembles a problem you are living with, a short call is the fastest way to find out whether it is the same problem underneath.
What happens on the call
- 15 minutes, no deck
- We map the process on the call
- A written scope, or an honest no