Killing the dashboard nobody used

This is the case study where the villain is my own product. Call intelligence worked: it transcribed, analysed, scored, and created tasks. It worked so well that it buried the practice. The summary page showed everything and therefore said nothing; the task list generated more items than any front desk could clear and gave them no priority.

The consequence was measurable and unforgiving. Practices logged in, felt behind, and logged in less. Adoption dropped, and churn conversations followed. A product can fail by doing too little; this one was failing by doing too much and leaving the triage to a human who had no time to triage.

Writing that down honestly — 'the amount of information we are showing is overwhelming', 'the summary page is pretty useless' — was the actual turning point. You cannot fix an adoption problem you are still describing as an education problem.

The reframe: the product's job is not to show a practice its data. It is to answer one question at 8am — what must we do right now, and what can wait until later today?

That question has a shape, and the shape became the page: an AI summary of how yesterday went, a To Action list split into Right Now and Today, and then two honest mirrors — To Improve and What's Working. Four sections, in that order, because that is the order a practice manager's morning actually runs in.

The hard part wasn't the interface, it was the rules. What earns 'Right Now'? I defined it precisely rather than by intuition: unreturned emergency and scheduling voicemails from the last 12 business hours, unreturned missed calls from new patients, new-patient opportunities from connected calls that haven't been actioned, and open non-scheduling emergency tasks. Everything else waits for 'Today'.

Business hours, not clock hours — a voicemail at 6pm Friday isn't urgent at 9am Saturday. Items age from Right Now to Today after 12 business hours, and close automatically after 36, marked Complete or Incomplete based on whether a connected call or message actually happened. The system closes its own loops so the list can be trusted to empty.

To Improve and What's Working are the same eight trackers measured against a benchmark — missed-call recovery, voicemail returns, first-call conversion for new and existing patients, opportunity follow-ups, average call score, task completion. Above the line it's celebrated; below it, it's work. And every benchmark is customisable per location, because a single-chair rural practice and a twelve-chair urban group do not share a definition of good.

Showing what's working was not decoration. A page that only ever tells a practice what they're failing at gets closed too — the same failure mode we were trying to escape, wearing better clothes.

The Action Center replaced the summary page as the product's front door. A practice opens it and hears or reads how yesterday went — calls by type and outcome, conversions, follow-ups, revenue opportunities, coaching signals — then works a short, self-clearing list, and sees where they sit against their own benchmarks.

Underneath it are two tables: action items, generated and closed by the rules above, and a daily summary stored in both audio and vector form so the last 7 and 30 days can be summarised without recomputing history. Non-task items can be assigned to owners — someone owns voicemail returns, someone owns missed-call follow-ups — so the list has names on it, not just counts.

This is the work I'm most willing to be judged on, because the problem was mine. I had shipped capability and called it value. The fix required admitting in writing that our summary page was useless, and then doing the unglamorous work of defining exactly what earns a practice's attention at 8am — in business hours, with closure rules, per location.

It also settled a habit I now apply everywhere: when adoption falls, the answer is almost never another feature. It is subtraction plus judgment — deciding, on the user's behalf, what matters today.