Airtable Feedback Base Rebuild for a Review Platform

Failure mode 01 Schema drift made records unreliable

The base's schema had accumulated inconsistencies that made filtering, reporting, and automation unreliable. Fields that should have been standardized contained free-form variations, related records were not consistently linked, and the overall structure no longer reflected the logical model of the feedback workflow. The team could not trust that a given query or view would return the complete and correct set of records.

Failure mode 02 The AI Summary field had stopped working

An AI Summary field — designed to automatically generate a concise summary of each feedback record — had failed. The field was either returning errors, empty values, or summaries that did not correspond to the record content. Because the team had built review workflows around this field, its failure created a manual gap: someone had to read and summarize feedback by hand, which defeated the purpose of the automation.

Failure mode 03 The public intake form produced inconsistent records

The public-facing intake form was the primary way new feedback entered the system. However, the form was generating records with inconsistent field values, missing links, and formatting problems. The team had to clean up incoming records manually before they could be processed, adding friction to the intake pipeline and reducing confidence in the data.

Decision 01 Rebuild the schema in place

The first decision was to rebuild the base's schema in place rather than create a new base and migrate records. This preserved record IDs, historical links, and automation references that would have broken in a migration. Octacer mapped every table, field, and relationship, identified where the schema had drifted, and corrected the structure field by field. This approach carried more risk than a clean rebuild because the base remained live throughout. Each schema change had to be validated against existing records and the team's active workflows before it was applied.

“Fix the structure without moving the data.”

Decision 02 Fix the AI Summary field at its root

The AI Summary field was repaired by correcting its configuration and the data feeding it. The field's input parameters were redefined to pull the correct record content, and the output format was standardized. The underlying data inconsistencies that had caused the field to fail were also addressed as part of the schema corrections, so the field received clean, consistent input.

“Make the automation trustworthy again.”

Decision 03 Repair the public intake form

The public intake form was reconfigured so that new submissions produced consistent records on the first pass. Field mappings were corrected, required fields were aligned with the schema, and the form's output was validated against the corrected structure. The goal was to eliminate the manual cleanup step that had become part of the team's routine.

“Structure the intake so records arrive correct.”