Takorin~3 min read

The factory tool had to say what to do next.

Solo work, three plant director interviews, and a pilot that has not reported yet.

  • Role: Product strategy and design
  • Timeline: Side project
  • Team: Solo
  • Impact: Not piloted yet
  • Platform: Web
Skip the static, live it

Overview

Three plant directors told me they never opened the dashboard they already owned, so I dropped the dashboard I had planned.

In food manufacturing, one point of overall equipment effectiveness, the share of scheduled time that is productive, can be worth six figures a year on one line. A bad shift reached directors as a report, after the moment to prevent it had passed.

Worked with: no one on the build; 3 plant directors in interviews.

What I owned

I owned

  • Primary research with three plant directors at top-four global food manufacturers
  • The product thesis, and throwing out the dashboard version after those interviews
  • The scoring model that ranks findings, including getting the ranking wrong first
  • Data-readiness tiers, so the product degrades on a thinly instrumented plant rather than failing
  • The entire interface, 22 screens

Research

Directors already owned the system and still did not trust it to tell them what to do.

I interviewed three plant directors at top-four global food manufacturers, expecting a feature wishlist. One said it plainly: it tells me what happened, not what to do about it. Each had a manufacturing execution system, and none opened it first.

All three feared the same thing: a tool that is confidently wrong, gets acted on, and the shift fails anyway. “What if it’s confidently wrong, I act on it, and the shift fails anyway.” Every one of them had been burned before.

Decision: one ranked queue

A ranked queue with the whole screen replaced a dashboard that nobody opened.

I rejected a delay-tracking dashboard and a ranked panel on the existing home screen. Both would repeat a failure the directors had described. One ranked entry point replaced five unranked alert sources.

Each finding carries a precedent from the plant’s own history, a countdown to when the window closes, and an owner. The risk is a wrong ranking that sends a director to the second issue while the first compounds.

The overview screen, showing four lines ranked by risk score with owners, SKU counts, and a shift briefing panel of decisions awaiting approval.
The shipped front door. One ranked list, an owner on every line, and the two decisions needing a signature before production starts.

Decision: show the uncertainty

A readiness score tells directors how much of the ranking to trust.

I rejected hiding low-confidence findings. For this audience, hidden uncertainty is how trust fails. Leading with a 64 is hard to sell, so each named gap has an effort estimate and what fixing it is worth. The score becomes a work queue, and the product degrades instead of failing on a thinly instrumented plant.

The Data Readiness screen, scoring the platform at 64 and naming the specific gaps reducing confidence, each with a readiness gain attached.
Readiness at 64, with the reasons named: the same ingredient under three different names across three systems. Resolving the supplier cluster is worth 18 points.

Redesign

A supplier module scores incoming material, the one risk a plant cannot instrument itself.

It tracks certificate status against the production schedule, delivery, and shelf life. Scores use certificate dates and delivery records, and suppliers can contest them against documents. Whether a visible score changes supplier behavior is untested.

A Knowledge tab captures the expertise of operators nearing retirement and rates the risk of losing it. Before this there was nowhere to put it.

The supplier module, showing a lot held for a missing certificate alongside a compliance scorecard ranking five suppliers.
Certificate status against the production schedule, and the scorecard the supplier sees. One supplier sits at 71 and provisional, 26 days left on its certificate.
The Knowledge tab, listing captured operator expertise with the person who holds it, how many times it has been applied, and its success rate.
Tacit expertise with a name and a hit rate on it. One operator’s amino nitrogen timing has been applied 112 times at 95%.

Scope

Twenty-two screens shipped, and one of four major review issues did not get fixed.

Two independent review passes found four critical and four major issues. Every critical one was fixed before sign-off. One operator workflow still takes three taps against a two-tap spec, because the fix needs a different input method than the screen supports. I logged it as design debt.

Results

The work was delivered as a pilot, and the effectiveness gain is still a hypothesis.

The design survived review, and one ranked entry point replaced five alert sources. The product is built to produce a 5-15% effectiveness gain on pilot lines and a shorter gap between a risk appearing and a response. Those are hypotheses for the pilot, not results.

Nobody argued the other side on whether a director would trust a machine-ordered queue

Interviews and a design review tested the decisions, but no stakeholder argued the other side on whether a director would trust a machine-ordered queue. The adoption risk is still mine to carry.

Sam Cusano