Protocol Deviation Analyst
Finds whether the sites are failing or the protocol is
- 9hard rules
- Things it must never do, enforced in every run
- 5handover rules
- Conditions where it stops and gives the work to a person
- 10output fields
- A structured result, not a paragraph you have to re-read
- ≤7steps per run
- Bounded work: at most 5 tool calls, enforced by the runtime
Read-only by design. This agent has no path to sending, spending, publishing or committing anything. Where that limit is the product, removing it would remove the reason to trust the output.
Overview
A clinical study generates deviations continuously: a visit outside its window, a procedure out of order, a sample handled incorrectly, an eligibility criterion applied loosely. Each is recorded, classified by whoever recorded it, and filed. Monitoring reviews them one at a time, and the standard report is a count by site.
A count by site answers almost nothing. Sites differ in enrolment, population and reporting honesty — and the site with the most deviations is frequently the one reporting properly, while the site with none is the one that worries an inspector.
The information is in the shape. Deviations concentrating on one protocol section point at a protocol that is hard to follow rather than careless sites, and if that is true, amending it fixes every site at once. Deviations clustering shortly after a site activates are a training problem with a known fix. Deviations rising after an amendment mean it was not absorbed. And the same deviation recurring after a corrective action closed means the action did not work — a finding about the response rather than the conduct.
This is deliberately narrow. It is an operational quality analysis: it makes no clinical or safety judgement, never assesses a participant, and never classifies anything as reportable. Those are regulated decisions with named accountability.
Benefits
The protocol fixed instead of the sites blamed
When the pattern says it is the protocol.
Training aimed where it works
At activation, where the cluster actually is.
Corrective actions that hold
Because recurrence after closure is finally visible.
What it does
Reads the protocol, not just the sites
Deviations concentrated on one section mean the protocol is hard to follow. Amending it fixes every site at once.
Normalises for enrolment and exposure
A raw count by site rewards the site that under-reports and penalises the one being honest.
Finds the post-activation cluster
Deviations concentrated in a site's first weeks are a training problem with a known fix.
Tests whether amendments landed
Deviations rising after an amendment mean it was issued and not absorbed.
Checks corrective actions actually worked
The same deviation recurring after an action closed is a finding about the response.
Makes no clinical or regulatory judgement
It never assesses safety, never assesses a participant, and never classifies anything as reportable. Those decisions carry named accountability.
What this changes
What this changes about your week
- It takes over
- Reading deviation patterns across a study rather than one at a time.
- Done today by
- A count by site, which rewards under-reporting.
- How often
- Reviewed individually; patterns never.
- Typical manual effort
- Normalising by exposure, then grouping by protocol section, activation date and amendment.
Why Pharma & Life Sciences in particular. Deviations concentrated on one protocol section mean the protocol is hard to follow — amending it fixes every site at once, where retraining nineteen sites would not.
Runs unattended
Started by you or by an event, and it finishes on its own. Nothing waits for someone to be at a desk.
The same standard every time
The two-hundredth item is held to the bar the first one was. Consistency is the part people cannot sustain.
It cannot act on its own
Protocol Deviation Analyst has no path to sending, spending or committing. That limit is why its output is safe to act on.
How it runs
This agent runs server-side through the PROMIVO runtime. Each run is logged step by step and every tool call is permission-checked before it executes.
- 1InputValidated against the agent input schema
- 2Modelclaude-opus-5 with automatic fallback
- 3Toolssearch_knowledge, extract_structured_data
- 4PermissionsRuns READ actions automatically; anything higher waits for approval
- 5OutputStructured result matching the output schema
- 6LoggingTokens, duration and every step recorded
Demo
Demo dataIllustrative sample output, abridged.
Input
{
"sites": [],
"study": "Study 4412",
"period": "18 months",
"deviations": [],
"minimumSample": 10
}Output
{
"escalate": true,
"disclaimer": "An operational analysis of the deviation log. No clinical, medical or safety judgement has been made; no deviation has been classified or assessed for reportability; no participant has been assessed or identified; no investigator or coordinator is named; no statement is made about compliance; and this is not advice on regulatory reporting.",
"normalisedBasis": "Deviations per 100 participant-months, so sites are comparable on exposure rather than on raw count.",
"amendmentEffects": [
"Deviations on Section 4 rose at 7 of 22 sites in the 6 weeks after Amendment 3. The amendment was issued; at those sites it does not appear to have been absorbed."
],
"escalationReason": "One protocol section accounts for deviations at 19 of 22 sites, and a deviation has recurred four times after its corrective action was closed.",
"byProtocolSection": [
{
"section": "Section 7.3 — visit window for the secondary assessment",
"deviations": 214,
"sitesAffected": 19,
"interpretation": "19 of 22 sites deviate on the same section. That is evidence about the protocol rather than about the sites — the window is 48 hours and requires a fasting sample. Amending it would address every site at once; retraining 19 sites would not."
}
],
"lowReportingSites": [
"Two sites report at less than a fifth of the study median per participant-month. Reported as a data quality observation, not as good performance — a site reporting nothing is the one an inspector asks about."
],
"tooThinToConclude": [
"Four sites have under 10 participant-months. No pattern is offered for them."
],
"postActivationClusters": [
"Three sites show 61% of their deviations in the first 8 weeks after activation. That is a training and start-up finding with a known fix, and it disappears afterwards."
],
"recurrenceAfterClosure": [
"The sample handling deviation at one site has recurred 4 times since its corrective action was closed in March. That is a finding about the action, not about the conduct."
]
}Requirements
Integrations
No integrations required.
Before you start
- The deviation log with dates, sites, protocol sections and classifications
- The protocol and its amendments, in a knowledge base
- Site activation dates and enrolment by site
- Corrective actions with closure dates
- Requires the All-Access plan or higher
Use cases
Periodic quality review
The shape of deviations, not the count.
Before a monitoring visit
Which patterns to look at on site.
After an amendment
Whether it was absorbed or only issued.
Study close preparation
Recurring patterns an inspector would follow.
Pricing
All-Access
Billed monthly through your PROMIVO subscription. Cancel at any time.
Runs consume your plan allowance for agent executions and tokens. See plan limits
FAQ
Does it classify deviations as major or reportable?
No. Classification and reportability are regulated decisions with named accountability, and an agent making them would be doing something it cannot be responsible for. It reports patterns across the classifications you already assigned.
Does it assess participant safety?
Never. It makes no clinical or safety judgement of any kind. Where a pattern touches safety it stops and escalates without characterising it.
Does it see participant data?
It works from the deviation log, which is operational. It reproduces no participant identifier and reports nothing about any individual participant.
Reviews
Reviews open once customers have run this agent.
Need this agent to work differently?
Tell us what to change and our team will scope a customised version for your business.
Customize this agent