Create a flow
Choosing the folder, the schedule, the fields to extract, and the entity the results land on.
A flow is created from Integration Flows in the console with New flow, which walks through four steps. Creating, editing and running a flow is limited to owners and admins; other members can see flows and review what they produce.
Folder
Choose one of your connected storage accounts, then browse to the folder you want watched. The picker starts at the drives that account can reach, so you can drill into My Drive, a shared drive, or a SharePoint document library and pin any folder along the way.
A flow watches exactly one folder, and the same folder cannot be watched twice through the same connection, so a second flow over the same folder is rejected.
Schedule
Name the flow, choose how often it should run, and set the sample size.
Frequency can be manual only, daily, every three days, or weekly. Sample size is how many files a sample run looks at, between 1 and 50, and it has no effect on scheduled runs.
What to extract
Define the fields you want read out of each document. Each field has a label, a type, and optionally a description and examples.
The description and examples are worth the effort: they are what the extraction reads to decide what a field means, so "the candidate's current employer, not their previous one" produces noticeably better results than a bare label. Types beyond plain text also constrain the answer, so a date field comes back in a single consistent format rather than however the document happened to phrase it.
Where it goes
Map each extracted field onto a column of an entity, and choose which column identifies the record.
That identifying column is how a flow decides whether it is updating an existing record or creating a new one, so it should be something genuinely unique such as an email address or contract number.
This step is optional. Skip it and the flow still runs, but its results park in the Review tab marked as needing field mapping, and cannot be approved until a mapping exists.
Nothing runs until you activate it
A new flow is created as a draft. Drafts never run on a schedule, whatever frequency they were given. Open the flow's Settings tab and choose Activate flow to put it on its schedule, and Pause flow to take it off again.
You can also run a flow by hand at any time, draft or not, from the flow's header:
- Run sample reads the first few files in the folder, up to the sample size, ignoring what it has processed before. This is the one to use while you are still tuning the fields, because you can re-run it against the same documents as often as you like.
- Run now performs a full run, reading everything new or changed since the last one, up to 200 files.
Only one run happens at a time per flow, so starting a run while another is still going has no effect.
Changing a flow later
Everything except the folder and the connection can be edited from the flow's Settings tab. Editing the extraction fields changes what future runs read, and does not revisit documents already processed. Clearing the entity mapping parks future extractions instead of discarding them.
Deleting a flow stops it and removes it from the list. Entity records it already wrote are kept, since they are ordinary records once written.
Where this is managed
Flows are created and edited under Integration Flows in the console. Entity types and their columns are configured under Entities and records. What happens to a flow's results is covered in Review and approve.