Pharmacy technicians, pharmacists, and operations teams
THE PRODUCT VISION
Let automation remove repetition while making uncertainty and correction work visible.
An anonymized experience narrative. Demonstrations use reconstructed flows and invented inputs; internal metrics and artifacts are excluded.
01 / THE STORY
A use case worth solving.
RECONSTRUCTED SCENARIO
An incoming document is turned into a structured record. The extraction looks complete, but a downstream reviewer needs to correct it. If the dashboard counts only generated records, the automation looks productive while someone else absorbs the cost.
Context
In prescription data-entry automation, the important product question was how much useful work the whole process completed. A high output count could conceal downstream corrections. The strategy needed to connect extraction, review, exceptions, and the quality of the final result.
What I owned
I led automation strategy, roadmap, targets, and success-measure definition. I worked with technical and operational partners on a feedback loop that brought correction patterns back into prioritization. The human review path was part of the product, not something left outside the model’s success measure.
THE STARTING CHALLENGE
Activity at one step
Count what the automation produces and inspect downstream problems separately.
→
THE PRODUCT DIRECTION
Quality across the journey
Evaluate useful completion, human intervention, and corrections together.
02 / THE DECISIONS
The reasoning behind the product.
DECISION 01
Measure the entire workflow
Pair automation measures with correction and rework signals.
The tradeoff
Optimizing a local step can move work to another team without improving the overall experience.
DECISION 02
Make uncertainty actionable
Route unclear inputs to a human with enough context to resolve them.
The tradeoff
A silent fallback or an unexplained exception makes operators reconstruct what happened.
DECISION 03
Turn corrections into learning
Connect operator feedback to evaluation examples and product priorities.
The tradeoff
Resolving the same class of error repeatedly wastes both operational effort and a learning opportunity.
03 / EXPLORE THE FLOW
Follow a use case through the decisions.
All inputs in the interactive demonstration are synthetic. No model is making real prescription decisions.
Enable JavaScript to play the synthetic scenarios and inspect the decision flow.
04 / TEST THE PRODUCT THINKING
What would convince us it works?
A reconstructed evaluation plan showing the questions I would make explicit. These are evaluation criteria, not reported test results.
Use case
Expected behavior
Evidence to inspect
Clear input
A useful record reaches the next step without avoidable correction.
Correct completion; downstream rework; total effort.
Uncertain input
The operator receives a clear exception with the relevant context.
The reason is captured in a way that can inform a future test or improvement.
Repeat error patterns; feedback usefulness; correction handling.
05 / OUTCOME & LEARNING
What the work changed.
This work supported more effective automation and operational feedback while keeping people connected to uncertain cases. Exact performance figures are withheld. The product lesson is the shift from counting generated output to understanding the work that is genuinely completed.
WHAT I WOULD CARRY FORWARD
I would define the correction taxonomy early. A correction rate tells a team something is wrong; a useful set of reasons tells them what to change.