Donor messages were buried across two inboxes. Not anymore.
Client
Palmera
Year
2026
The problem
Palmera's supporters write in on Instagram and WhatsApp, and every message was read and answered by hand. Donations, broken recurring gifts, tax receipts and offers to sponsor a family sat in the same pile as thank-yous and story reactions. The same questions were answered again and again, and nobody could see who had asked what, or whether anyone had replied. "Realised I haven't responded in a while!" Palmera told us, after checking Instagram.
Our process
The inbox, audited
We exported ten years of Instagram and WhatsApp: 6,238 messages from supporters.
The end result
The inbox runs in real time. Every Instagram and WhatsApp message is pushed to one Google Sheet the moment it arrives, with no exports or waiting for a daily sync, and within a minute it's sorted by what the person wants and how urgent it is. Common questions arrive with a template ready to approve. Anything about money, a bereavement or whether the charity is genuine always goes to a person, by a fixed rule rather than the AI's judgement. Emoji and story reactions are logged and left alone. The founder now opens the sheet to just the messages that need her: 90 out of 630 this year. It all runs on Palmera's own accounts, so they own every part of it.
From 6,238 messages to 22 needs
We didn't skim the archive, we studied it. We read ten years of Instagram and WhatsApp without deciding the groups in advance, so new kinds of request could show up instead of being lumped into "other". Similar messages were grouped into 22 types, every message was sorted into one, and a second, independent check made sure the sorting held up. What we found: almost half the inbox was emoji and story reactions, while donations, broken recurring gifts and sponsor-a-family requests sat in the same pile as "thank you".
What a machine should never answer
We ranked the 22 types by volume and risk, and drew the line with Palmera. Bereavements, payment problems and questions about whether the charity is genuine always reach a person, by a fixed rule. Common questions get one of 11 templates built from real message groups, and only when the AI is sure.
The right route, not the obvious one
We tested every way in: each platform's API, their native inbox tools, and the messaging provider Palmera already paid for. The direct Instagram route would have tied the charity to a registration it could never undo, so we found an already-approved route instead. Everything lands in Google Sheets, because "some of our staff struggle with the many different platforms that we use", said Palmera.
Built, tested, handed over
Before launch we replayed real messages through the rules and fixed every miss, including two donor messages an early version would have filed as spam. Then we launched approve-first, so nothing sends until the team ticks it. It runs on Palmera's own accounts with a written guide, so they own every part of it.
They loved it so much, they came back for seconds.
Next up was month-end reporting. Palmera's field teams track their activities across fifteen project spreadsheets, and every month someone checked them by hand and built the next month's sheet. We automated the check and the rollover, so the team spends the month on the work, not the spreadsheets.
Capabilities
Research synthesis
Thematic analysis
Platform research
API integration
Triage design
AI classification
Template design
Workflow automation
Google Sheets
