Skip to content
Husain Bootwala
07In discovery

Pharmacovigilance Productivity AI

Mapping a regulated industry's working day before committing to a product

The adverse-event case lifecycle as five columns: intake and triage by an intake associate, case processing by a drug safety associate, quality control, medical review by a physician, and regulatory submission. Two dashed arrows run backwards: a rework loop from QC to processing, and a loop from the archive back to intake when new information reopens a closed case. A highlighted band explains that Day 0 is company awareness, that a fatal unlisted case is due in seven calendar days and other serious unlisted cases in fifteen, and that late seriousness upgrades recalculate the clock from the original date.
Diagram · The life of a case: five pairs of hands, two rework loops, and a submission clock
Channel
07
Period
2026 —
Domain
Software
Role
Sole researcher: domain modelling, process flowcharts, opportunity scoring, competitive teardown

Stack

  • Domain research
  • Process modelling
  • Regulatory analysis
  • Opportunity scoring
  • Agentic research pipelines
  1. A case flows through an automatic assessment box straight to submission, with the human sign-off link cutCaseAuto-assessAI decidesSubmitHuman sign-offJudgement removedCompliance can't sign it off
    trapThe obvious AI product automates the judgement regulators insist stays human.
  2. Five case stages, each still done by a person, with AI opportunities pinned beneath: 11, 9, 6, 6 and 7Owned by a personIntake11Process9QC6Review6Submit7AI pinned to a step+33 on other flowcharts
    insteadMap the day first, then pin AI beside steps people still own.
AI opportunities scored
72
Role flowcharts
10
Roles timed hour by hour
4
Directions chosen
0

How the discovery ran

  1. A case passes intake, processing, QC, review and submission, with a rework loop, while a clock runs from Day 0 to day 7 or day 15IntakeProcessQCReviewSubmitReworkLegal clock, daysDay 0715FatalSeriousDeadline
    ProblemSerious unlisted cases run against a 7- or 15-day legal clock.
  2. A master case lifecycle fans out to ten role flowcharts, with a separate hour-by-hour day in the life of four roles00Case lifecycle01Intake02Process03Coding04Review05QC06Submit07Literature08Signal09Aggregate10QPPV11Day in the life: 4 roles, by the hour
    mapOne master lifecycle, ten role flowcharts, and four roles timed hour by hour.
  3. Four case-processing steps, each pinned to an AI opportunity with its score out of 20: 14, 13, 16 and 17Flowchart 02/20Read sourcesAI-02-0114Enter fieldsAI-02-0213Code eventsAI-02-0316Self-QCAI-02-0717Pain · Freq · Open · Feasibility
    pinEach opportunity pinned to a step and scored on four axes.

What the map shows

  1. Strip plot of all 72 scored AI opportunities by flowchart; 21 score 16 or more out of 20, and 7 are already contested by vendors16+1113151701020304050607080910Score /20ContestedFlowchart
    Result72 scored. 21 score 16+ of 20. Vendors already ship 7.
  2. A five-phase gate at phase 0, with the MVP locked at phase 3 and none of the 6 to 10 practitioner interviews donePhase gateDirection: open0Understand1Validate2Choose3Scope MVP4BuildNowPractitioner interviews0 / 6–10
    nextPractitioner interviews come next. The MVP stays locked until then.

Why it’s here unfinished

No product and no code. Anyone can generate an app now; the scarce skill is knowing which one is worth building. In a regulated industry, a wrong guess is a compliance exposure, not just a wasted sprint.

Read the engineering detail

Why this project is here unfinished

This project has no product, no code and no launch date, and it is the one I would most like to be asked about.

Anyone can generate a working application now. The scarce skill is knowing which application is worth generating. In a regulated industry, getting that wrong costs more than a wasted sprint; it creates a compliance exposure.

So this repository holds the part of the work that usually gets skipped.

The trap

Pharmacovigilance is the discipline of detecting and assessing adverse effects of medicines. Safety teams process individual case safety reports under legally binding submission deadlines, the volume is high, and the deadlines do not move.

The obvious AI product, then, is automated case processing. It is also close to the worst thing to build. Case assessment is where regulators require demonstrable human judgement, and a vendor promising to remove it is selling something a compliance officer cannot sign off. A product built that way would be unbuyable however well it worked.

What I did instead

I modelled the day first: a master model of the whole case lifecycle on one page, ten role flowcharts from intake through medical review to regulatory submission, literature, signal, aggregate reporting and compliance oversight, and an hour-by-hour day in the life of four roles.

Then every opportunity had to be anchored to a step. Seventy-two candidate AI interventions, each scored on pain, frequency, openness and feasibility, and each required to point at a specific box on a specific flowchart. Anything that cannot be located in someone’s working day is a feature idea rather than an opportunity.

The pattern that came out: incumbents have automated the task, but not the accountability around it, such as proving it was done, catching it before it fails, and explaining why it went wrong. Seven of the 72 are already contested by shipping vendor features, narrative drafting among them.

Vendor numbers are kept apart from observed fact. Incumbents publish efficiency figures; those are recorded as claims, in a dated file, until something independent supports them.

Where it stands

The direction is still undecided, and I am comfortable with that. The models come from published literature and vendor material, and desk research has a ceiling: they need correcting by people who do this work every day. Practitioner interviews are the next step, and the MVP folder stays locked until they have happened.

Picking a direction now would have felt like progress. It would also have been the most expensive mistake on the table.

Media

A two-column diagram. The left column lists the steps a case processor performs: read every source document, enter the fields, code in MedDRA and WHODrug, complete seriousness and expectedness, write the narrative, draft the follow-up query, self-QC. The right column pairs each step with the registered AI opportunity that attaches to it, by identifier. The narrative row is the exception: it is highlighted, and its right-hand cell says drafting can be assisted but the step stays owned by a human.
Diagram · A case processor's day, with every AI opportunity anchored to a step
The top six rows of the register's shortlist table, verbatim: AI-04-02, AI-02-07, AI-08-02, AI-06-06, AI-05-05 and AI-03-02. Each row gives the opportunity, the flowchart it is anchored to by number, its pain, frequency, openness and feasibility scores out of five, a composite of 16 or 17 out of 20, a regulatory burden of M or L, and "the catch", the honest objection to it. The last row reads: no platform vendor can build this, the knowledge is per-company.
Screen capture · Top of the shortlist: six of the 72 candidates, each scored and anchored to a diagram

How this was built

The spec sheet above lists the stack. This section covers how the work was run, which a screenshot cannot show you.

Where AI did the work

  • Ran research as a pipeline with a fixed output contract: every finding lands in a register with a score and a link back to the step in the process model it attaches to. An opportunity that cannot be pointed at a box on a diagram is not admitted.
  • Used separate passes for building the domain model and for challenging it, because an agent that researches and validates in the same breath will confirm whatever it already wrote.

The discipline around it

  • Observed fact and vendor claim are stored separately, with a dated validation file tracking which is which. Every efficiency percentage a software vendor publishes is treated as marketing until independently checked.
  • Point-in-time research is dated and never edited in place; evergreen domain understanding is updated in place. Mixing the two is how a knowledge base rots without anyone noticing.
  • The repository states in writing that nothing in it is a commitment. Without that line, a well-written research document starts getting treated as a decision.

What stayed human

  • The product direction is marked undecided in writing, and the MVP folder stays locked until that changes. AI can enumerate and score options; it cannot be allowed to pick, because the choice is where the judgement sits.
  • Desk research has a stated ceiling. These models are built from published literature and vendor material and are marked as unvalidated until real practitioners correct them.

Process artifacts in the repo

flowcharts/00–11
Master case lifecycle, ten role flowcharts, and an hour-by-hour day in the life of four roles
AI_OPPORTUNITY_REGISTER.md
72 candidates, scored, each anchored to a process step
VALIDATION-2026-07.md
Dated ledger of what is verified versus what is vendor claim
DECISION_LOG.md
Open questions, with "undecided" recorded as a state in its own right