The call went well, the agent captured everything, and the Google Sheet row arrived with half its columns empty. This is one of the most common problems when teams first connect AI calls to a sheet, and it is almost never a data problem. It is a naming problem. In one deployment, a hospital group's intake calls collected treatment, department and arrival time correctly, but those three columns stayed blank until the names were fixed. Here is how to find and fix it.
Connect your agent to a sheet: Try Edesy free — Rs 50 free credit, no demo needed.
How the data travels
- During or after the call, the agent's data fields are filled: for example
patient_name,diagnosis,specialty,arrival_time. - When the call ends, a workflow receives those fields.
- The sheet step maps fields to columns: "put
patient_namein column B", and so on.
If step 3 asks for a field name that step 1 never produced, the column is empty. Nothing errors. The row simply arrives incomplete. (Capturing missed calls rather than AI call results? See how to send missed call leads to Google Sheets.)
What went wrong in one deployment
The agent's instructions told it to capture:
| Field the agent produced | Field the sheet mapping expected |
|---|---|
diagnosis |
reason_for_visit |
specialty |
department |
arrival_time |
appointment_time |
Both lists described the same information in different words. The name and summary columns filled correctly because their names matched; the other three stayed blank. Aligning the mapping to the agent's field names fixed every new row.
The fix, step by step
- Make a test call and open it in call history.
- Read the captured field names exactly as they appear, including underscores and case.
- Open the sheet mapping in your workflow and compare each entry.
- Change the mapping (or the agent's field names) so both use the same names.
- Make another test call and confirm every column fills.
Prevent it next time
- Keep one field list. Write the field names once and use them in the agent's instructions, the data field definitions and the mapping.
- Friendly headers are fine. The column header can say "Department" while the mapping uses
specialty. Only the mapping has to match. - Re-test after every edit. Changing a prompt can rename a field without anyone noticing.
- Watch for silent gaps. Check the sheet daily in the first week; an all-blank column is your signal. See catch silent CRM and webhook failures.
- Define fields for every agent. A second agent with no data fields sends nothing useful.