Situation

One angry message, sent at the wrong moment, can end up read aloud in front of a judge.

My client runs a mediation platform for people in high-conflict situations, such as divorced parents co-parenting a child while under a restraining order, where communication is legally required but often difficult. Every message and video call is monitored and moderated to keep communication safe, structured, and civil. It is a lower-cost alternative to traditional mediation involving legal professionals.

That framing sets the stakes for everything the platform does. A message is not just a message. It is evidence that can be used in court against you.

The platform is built on Knack, a no-code tool for rapid custom business apps.

Complication

The product worked but the operation behind it was dependent upon people doing everything by hand.

On the last day of every month, someone backed up the system one table at a time. Manually. They had done it that way for over ten years.

Compliance documents were hand-crafted, one PDF per user, across the entire database.

Every completed review ended with someone remembering to send an email.

Every message was read start to finish with no flagging of any kind, so a reviewer gave the same attention to a routine scheduling note as to a veiled threat.

Booking a paid monitored call was the longest chain of all. A parent filled out a form to request a date and time. They were redirected to an external site to pay. Someone received that payment, then manually checked whether the slot was actually free, manually created an event in Calendly, manually created a Google Calendar event for the invite, and finally updated the database to mark the payment as received and the call scheduled. If the slot turned out to be taken, they replied to ask the parent for another time and the chain restarted.

None of this was anyone’s fault. It is just what happens when a business grows steadily and each new need gets solved by a person rather than a system.

The problem is that manual processes pose risks: they fail on the day someone is sick, or busy, or reading the two hundredth message during a shift.

Investigation

The obvious approach would have been to automate the most painful task first and work down the list to ease the burden off the client.

I went the other way, and started with what was the riskiest to get wrong.

A skipped backup stays invisible until the month it is needed. An unflagged threat reaches the person it was aimed at. Those failures do not announce themselves the way a slow afternoon does.

So backups and moderation came first, ahead of the tasks people complained about more.

Resolution

Moderation was the clearest case for automation and the clearest case against automating the decision. Each message runs through a word and phrase check first, which grades what it finds by how serious it is and how confident the match is. Confidence is important because people start disguising words once they know someone is watching them. Once those checks are completed, the findings go to an AI pass with recent message history for context, and come back with a recommendation and a suggested revision drawn from the platform’s own guidelines. Every decision that reaches a parent was made by a person. How the flagging works, and why the final decision stays human.

Compliance documents moved from hand-crafted, one per user, to generated on demand, with the formatting a legal record actually requires. Court-ready transcript generation.

Backups went from a monthly manual ritual to a scheduled job with offsite copies, and the restore path was automated alongside it. That mattered more than it sounds: the platform prunes records to control cost, so deletion is only safe while the way back is reliable. Making deletion safe on a database that bills by the record.

Consultation booking collapsed from that nine-step chain into a flow the parent completes themselves. Availability, payment, and calendar invites now settle without anyone relaying information between three systems by hand. Wiring scheduling into the platform’s own billing rules.

Notifications stopped depending on someone remembering, and gained something manual sending never had: they respect each recipient’s local hours, correctly, across daylight saving transitions. Timezone-aware scheduling.

Results

• Nobody backs up the database by hand. A ten-year monthly ritual is now a scheduled job with offsite copies.
• Nobody hand-crafts compliance PDFs. Documents generate on demand, formatted for a court.
• Nobody remembers to send the email. Review completion triggers it.
• Nobody reads every message at equal attention. Reviewers still read all of them, with the likely problems already surfaced and a suggested revision attached.
• Nobody relays a booking between a form, a payment provider, and two calendars. The parent books, pays, and receives the invite without a staff member in the middle.

The reviewers, administrators, and support staff still do the work that needs judgment. What they stopped doing is the work that only needed someone to remember.

Back to Case Studies