Take a case from the agreed sources, check the validity elements, find duplicates, code the verbatim terms and draft the narrative — the clock runs from day zero and a safety physician assesses it.
Take the case in from the agreed sources — the mailbox, portal, partner, literature or site channels in scope.
02
Stamp the moment the organisation first held the minimum information, and count day zero from that point.
Reason
03
Search the safety database on patient, product, event and dates, and list the cases this one may already be.
04
Code the reporter's verbatim terms to MedDRA, keeping the words as written beside every term selected.
05
Check the four validity elements — reporter, patient, suspect product, event — and name the one that is missing.
Decide
06
Mark what only a safety physician rules on: causality, seriousness, expectedness and whether the case is reportable.
07
Hold a thin, contradictory or unmatched case for the case processing lead rather than filling the gap itself.
Out
08
Draft the narrative in the order your convention sets, and draft the follow-up questions that would close the gaps.
09
Retain the source document, the receipt date, the duplicate search run, each coded term and every reviewer change.
→Product statement
The agent takes in, searches, codes and drafts. Causality, seriousness, expectedness, whether a case is valid and whether it is submitted stay with a qualified safety physician and your case processing lead.
Example workflow
One case, end to end
AgentHuman
1Case receivedWhichever of the agreed sources is in scope — mailbox, portal, partner exchange, literature screening or a study site
2Day zero stampedThe moment anyone in the organisation first held the minimum information, recorded before anything else runs
3Duplicate searchedThe safety database searched on patient, product, event and dates, so a follow-up joins the case it belongs to
4Coded and draftedVerbatim terms coded to MedDRA beside the reporter's own words, then the narrative and the follow-up questions
No human action required
Stages 1 to 4 run without a person in the loop — receipt, the duplicate search, the coding and the draft finish before a reviewer opens anything. The clock is running through all four.
5DecisionSplits on the validity elements and coding confidence
Elements present, terms matched
Reaches the physician's queue ready to assess.
Element missing or coding unclear
Held with the gap named, and the clock still running.
Safety physician and case lead
The physician assesses causality, seriousness and expectedness; the case processing lead settles what the case is and whether it goes.
Physician assesses · Correct the coding · Ask for follow-up
Assessed — handed back▼
6Case handed to reviewWritten to the safety database only where write access and policy allow; the assessment fields stay empty
7Outcome evaluatedCoding agreement, duplicate recall, narrative edits, follow-up yield and the clock spent before a physician opened it
Corrections
Every term the physician recodes is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Determining causality, seriousness or expectedness.
Deciding a case is not valid.
Deciding a case is not reportable.
Submitting an ICSR to an authority or partner.
Automation boundaryAgent acts unaided
✓Take the case in, stamp day zero and check the four elements.
✓Search the safety database and list the cases this one may already be.
✓Code the reporter's verbatim terms to MedDRA.
✓Draft the narrative and the follow-up questions.
Write actions run only inside the approval boundaries agreed during implementation. ICSR submission is not among them.
Closing a case or nullifying a submitted one.
Contacting a patient or reporter on the company's behalf.
Downgrading a consumer's report to a non-reaction.
Changing coding conventions or duplicate-match rules.
Example output
One case, annotated
Everything the agent proposes stays attached to the source document it was read from.
Case-processing output · single caseIllustrative example
Reporter said
Source
Day zero
Element missing
Confidence
Seriousness
Rash, then felt faint
Call centre, consumer
First receipt
Identifiable patient
92%
Not set by the agent
As receivedThe reporter's own words, the channel they came in on and the moment the organisation first held them.
Evidence usedCall record, verbatimNo database match foundConsumer, not confirmed
Why it is heldOne of the four elements is not there. The agent names which one, and the clock runs on.
ActionPhysician assessesCorrect the codingAsk for follow-up
What the score decidesConfidence says how much coding a reviewer re-reads, not whether the case is serious.
Value
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value landsValue 01 – 04
Every case receivedCall centre, literature, partner or study site
03Intake & coding
Work to your own conventions
Use the coding conventions, duplicate rules, narrative order and follow-up forms your safety system already runs on.
01Approved path
Give the physician back the clock
Receipt, the duplicate search, the coding and the draft are done before the queue opens, so the assessment starts nearer day zero than the deadline.
02Human review
Send up what will not resolve
A missing validity element, an unmatched duplicate candidate or a verbatim term that will not code reaches the case processing lead as a named gap.
04Build an evidence trail
Retain the source document, the receipt date, the duplicate search run, each coded term against its verbatim, the confidence and every physician correction — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the safety database
Each control wraps the one inside it. A case clears every layer before a reviewer opens it, and a change to any rule inside them goes through your change control.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeReturn case processing to your team if evaluations or production signals degrade.Roll back
L5TraceabilityRecord the source, the search run, each term against its verbatim, the action and the reviewer changes.Record
L4Day-zero clockDay zero is read from the source's own receipt evidence; where it cannot be, the case is held.Stamp
L3Physician gateCausality, seriousness, expectedness, validity and submission stay with named people.Gate
L2Duplicate searchCases are searched against the database first; candidates are listed, never merged, and a search that cannot run holds the case.Search
L1Validity and verbatimEach coded term is bound to the wording it came from, and a missing element is named.Bind
Model coreCase draft proposed — elements checked, duplicate candidates, coded terms, narrative and confidence
L1 – L2Keep the case tied to what was reported
L3Decides what only a physician may settle
L4 – L5Stamp day zero from the source, and record the trail
L6Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole case — not only the terms that were coded.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the case a safety physician opens
Depth of coverage ▼
E1Coding agreementDo the selected terms match what your own coders select from the same verbatim?
E2Validity-element recallWas a missing element found, and was a present one ever called missing?
E3Duplicate detectionDid the search find the existing case, and how many false matches did it put up?
E4Narrative faithfulnessDoes the narrative state what the source says, in the order your convention sets?
E5Clock accountingWas day zero right, and how much of the window had gone before review?
E6Slice evaluationHow does case quality change across sources, languages and populations?
Floor — the report a named person submits and answers for
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle. Under-reporting and over-reporting are both harms, and both start on this map.
Agent lifecycleDirection of processing →
01 · Intake / receipt2 modes
PV-01
Day zero recorded late
The clock starts at data entry, not at first receipt.
PV-02
Case missed at intake
A report in a shared mailbox is never opened as a case.
Stage receivesThe source document, the channel and the day-zero date
02 · Duplicate search1 mode
PV-03
Duplicate entered as new
A follow-up becomes a second case, and the aggregate moves with it.
Stage searchesThe safety database for the same patient and event
03 · Coding1 mode
PV-04
Verbatim coded too general
A term one level up drops the case below the seriousness threshold.
Stage codesVerbatim terms to MedDRA, with the wording kept
04 · Draft / write2 modes
PV-05
Narrative outruns the source
An inference no reporter made is written into the case.
PV-06
Follow-up question never put
The gap that would complete the case is drafted and left.
Stage draftsOnly where write access and approval policy allow it
05 · Change / Version1 mode
PV-07
Silent convention change
A coding-convention update shifts term selection unnoticed.
Stage tracksModel, prompt, convention and match-rule changes
Sev-1 · the case reports late, or not at allSev-2 · the case and the aggregate carry itSev-3 · the case stays thin, review grows
Both harms sit in one number here: a term pushed too general, and a case held that needed no holding. It counts cases re-coded, re-matched or sent back before a physician could assess them, cut by source.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Non-English source documents
5.9%
3.8×
Review
Consumer reports, no confirmation
4.5%
2.9×
Review
Literature cases, abstract only
2.8%
1.8×
Watch
Routine follow-up on open cases
1.1%
0.7×
Normal
Bar: reworked-case lift vs. routine follow-up baseline · scale 0–4.0× · tick at the 2.0× threshold2 of 4 slices over threshold
Evidence-linked improvement
One case corrected by hand teaches nothing on its own
What a physician recoded or rewrote points at a convention, a match rule or a narrative order. That is where the change goes, before the next case.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Recoding, rematching or send-backs rise in one case source.
02Diagnose
The reporter's own words are the reference; the case is read to them.
03Improve
The convention, match rule or narrative order changes with a named approver.
04Verify
Held-out cases from that source are processed again, the failures among them.
05Learn
The case is kept as a test, and the convention carries what the physician wrote.
Learn → DetectThe return edge. A change here is a change to a validated safety system — assessed, approved by a named person and recorded before any case is processed against it.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, intake and the clock, duplicates and coding, evaluation, review workflow, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Workflow discovery and automation-boundary definition.
02Day-zero rules and clock accounting.
03Safety-database and intake assessment.
04Case intake from the agreed sources.
05Duplicate search and match rules.
06MedDRA coding to your own conventions.
07Validity-element checks and gap naming.
08Narrative and follow-up drafting templates.
09Coding-agreement and duplicate-recall evaluation.
10Physician and case-lead review workflow.
11Safety-database write-back and audit trail.
12Observability, deployment and Agent Care handover.
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallelFinal scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
Capability✓ in scope · — not at this tierPilotOne product, one sourceProductionProduction safety-database linkAdvancedMulti-product / multi-source
Introduced at Pilot
Case intake and day-zero stamping✓✓✓
Duplicate search and candidate list✓✓✓
MedDRA coding with the verbatim kept✓✓✓
Validity-element check and gap naming✓✓✓
Narrative and follow-up drafts✓✓✓
Assessment by a safety physician✓✓✓
Traceability and baseline evaluation✓✓✓
Introduced at Production
Additional intake channels and sources—✓✓
Safety-database write-back and observability—✓✓
Introduced at Advanced
Partner exchange and multi-language intake——✓
Multi-product and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on products and sources in scope, safety-database and intake integrations, case volume, coding conventions, review controls and deployment requirements.
Separate from buildBuild pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bringWe build with it
01Your day-zero rule and who counts as first receipt→Day-zero rules and clock accounting, agreed with QAWeek 1
02The sources cases actually reach you from→Case intake from the agreed sourcesWeek 2
03Access to the safety database and intake channels→Safety-database and intake-channel assessmentWeek 2
04Cases you know are duplicates, and pairs you decided were not→Duplicate search and match rulesWeek 3
05Your coding conventions and the terms your coders argue about→MedDRA coding to your own conventionsWeek 3
06Cases that went wrong, including the ones that went late→Coding-agreement and duplicate-recall evaluationWeek 4
07Named safety physicians and a case processing lead→Physician and case-lead review workflow, then supervised casesWeeks 5–6
Nothing else is requiredDeployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Week 5 carries both the coding-agreement evaluation and the first cases worked under supervision.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Day zero, the boundary and what only a physician settlesW2Intake on the first sources and safety-database accessW3Duplicate search, MedDRA coding and the drafting templatesW4Coding agreement, duplicate recall and clock accountingW5Supervised cases, write-back and targeted correctionsW6Physicians assess from agent-built cases, then Agent Care
Reading the bandEach bar covers the weeks its work is named in, and no more. Day zero is settled in week 1, and the clock figures reported later are only as honest as that definition.
At the end of W6A safety physician has assessed cases built this way and can tell you whether the clock moved. Conventions, match rules and clock reporting go to Agent Care.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Biotechnology AI agent
Build a case-processing agent around the safety system you already run.
Show us one product, the sources cases reach you from and how a case gets to your safety physician today. Ask your team when day zero was for the last five cases, and see how many different answers come back. That number, not the coding, decides whether the rest of this is worth building for you.
