CRO data management without the re-keying: every deliverable read, normalised, checked against its contract and cited
Every contract-lab deliverable becomes clean, cited records in the warehouse within hours, with nothing typed in by hand.
See one case, screen by screen ↓Every contract lab sends results its own way. One CRO drops a final report PDF and a results workbook on SFTP, another posts an instrument export to its portal, a third emails an Excel file. Half-lives arrive in minutes or hours, clearance in mL/min/g, permeability in nm/s, protein binding as % bound — and compounds come back as “NB-4471”, “Cmpd 16” or a tube label with two digits swapped.
So a data steward re-keys every value into the house schema, converts the units and matches each name to the registry, while a scientist reads the report to check the controls and replicates against the statement of work. A CRO can mark a plate “Pass” against its own SOP range when the contract does not accept it — and unless someone opens the contract, that value goes straight into the SAR view the designers rely on.
Estimated split for a typical in-vitro ADME deliverable, by hand and with the solution, in working days.
Six agents classify, extract, normalise and check every deliverable against its contract; the study monitor decides each exception.
Halden Biosciences delivers microsomal stability study HB-26-0417 — a report, a results workbook, 23 compounds in human and mouse liver microsomes. Here is what happens next, screen by screen, in the working solution.
Priya opens the solution and sees every CRO package this week: 38 deliverables read, normalised and checked against their contracts, 30 already in the warehouse, 6 waiting for her decision — each with its evidence. Two packages are arriving now, from Quillon and Larkspur.
“38 CRO deliverables this week — read, normalised and checked against their contracts.”
Larkspur Screening drops plate set 12 on SFTP and Priya presses Process now. The classifier names study LS-0932 and contract NWB-SOW-050 §1; the extractor reads 4 plates, 1,408 compound wells and 4 plate Z′ values; every compound matches the library manifest. Z′ is 0.72–0.82 on every plate against a contracted minimum of 0.5, and 6 hits sit above the 60% threshold. No exceptions — it joins the next warehouse load without anyone touching it.
“No exceptions — joins the next warehouse load.”
D-2611 opens with the source on the left and the records on the right: 96 values built by 4 agents, mean confidence 98, 46 CLint values converted from mL/min/g to µL/min/mg. “Cmpd 16” is matched to NWB-4486 through row 16 of the shipped manifest, and “> 60 min” stays a qualified value. Priya clicks the verapamil control on mouse plate 2, and a line runs to Table 2 on page 6 of the report.
“Outside the contracted 3–15 min — the CRO marked it ‘Pass’ against its own SOP range (3–20 min).”
Three exceptions, each with its evidence. The verapamil half-life on MLM plate 2 is 17.9 min; Schedule 4.2.3 of NWB-SOW-031 accepts 3–15 min in mouse microsomes. Page 11 of the report notes that plate 2 used microsome lot ML-2291 after a second freeze–thaw cycle. The exception affects 22 MLM values across 11 compounds.
Suggested: “Hold the plate 2 values and ask Halden to repeat the plate with a fresh microsome lot.” The agents propose — a named person decides.
NWB-4490 was not incubated — a precipitate formed at 1 µM in phosphate buffer — so Priya records it as not tested, and chemists will see why it is missing. NWB-4488 replicates in human microsomes were 17.6 and 27.6 min, a CV of 31% against the contracted 20%; it loads with a “high variability” flag beside the value.
For the plate 2 control, the query writer drafts a note to Dr. Ilse Varga, Study Director at Halden, with Dr. Lena Ortiz in copy. It cites page 6, Table 2 and contract §4.2.3, asks for MLM plate 2 to be repeated with a fresh microsome lot, and says which 22 values are held. Nothing is sent until Priya approves.
“Drafted by the CRO query writer — sent through the Halden portal after you approve.”
With every exception decided, Priya loads the study. The loader checks 96 records against house schema v4, writes 74 values with house units, qualifiers and flags, keeps 22 out for the CRO’s reply, and writes 74 citations back to the page or cell. The Aster program’s SAR view refreshes, so designers see the new values now.
“Nothing re-keyed — every value carries its source.”
The study’s activity trail shows each step in order: the classifier matching the package to HB-26-0417, the extractor reading 96 values from Table 2, Table 3 and the Summary sheet, the normaliser’s conversions, the checker’s three exceptions, Priya’s decisions, the query approved and sent, and the load of 74 values with 22 held.
The dashboard shows values loaded, clean first pass, receipt to loaded, on-time delivery by CRO and where checks fail. It names the pattern: Sorrel’s in-vivo PK misses most often — 3 of the last 8 rat studies ran past 20 working days. The CRO scorecard goes to Omar each week.
The same study, seen by the five people who handle it — what their week looked like, and what it looks like now.
Six agents classify, extract, normalise and check every deliverable against its contract; the study monitor decides each exception.
Reads every file in a CRO package — report PDF, results workbook, instrument export, cover email — and identifies the CRO, study, assay and contract schedule.
Turns report tables, workbook cells and instrument exports into result records in the house schema, with a page or cell for every value.
Converts units by approved rules and matches compound names to the registry through the manifest shipped for the study.
Checks controls, replicates, recovery, integrity markers, time points, compounds delivered and turnaround against the contract, and raises an exception with the evidence for every failed criterion.
Loads decided records to the research data warehouse with house units, qualifiers, flags, holds and the citation to the source.
Drafts a short, precise query to the CRO study director, citing the report page or workbook cell and the contract clause, and saying which values are held.
Accepts, holds or queries. The agents propose; a named person decides.
The study monitor, the steward and the scientists can ask in plain words — or tell it what to change. Every answer links to the page or cell it rests on.
What is blocking the Halden stability load?
Three decisions are open on HB-26-0417. The verapamil control on MLM plate 2 ran at 17.9 min; the contract allows 3–15 min. Halden marked it “Pass” against its own SOP and notes a second freeze–thaw of the microsome lot — 22 values wait on this. NWB-4488 replicates differ, CV 31%. NWB-4490 was not tested — solubility. The other 71 values are clean.
Which CRO fails quality checks most often?
Over the last 8 weeks, Halden Biosciences at 6.1%, then Corrib Labs at 4.2% and Quillon Bioanalytical at 3.8%; Larkspur Screening is lowest at 1.6%. Most of Halden’s failures are controls — its SOP accepts a wider verapamil range in mouse microsomes than the contract.
Show every result for NWB-4471
Across the CRO deliverables, in house units: HLM CLint 28.8 µL/min/mg, MLM CLint 64.5 µL/min/mg, Caco-2 Papp A→B 8.4 × 10⁻⁶ cm/s, efflux ratio 1.3, IRAK4 IC50 3.2 nM, hERG IC50 > 30 µM, fraction unbound 0.042, IL-6 release IC50 38 nM, CYP3A4 IC50 14 µM. Every value opens the exact page or cell it came from.
Add a rule: remind the CRO when a deliverable is 2 days late
Done — the rule “Send the CRO a reminder when a deliverable is 2 working days past the contracted turnaround” is in Settings. Today it would remind Corrib Labs about CL-7719 (day 12 of 10). Every reminder is on the audit trail.
13 screens from the working solution, on its sample data. Pick one to see it large.
Who decides, which quality rules the agents apply and where deliverables come from are all settings.
| Setting | Default | Choose from |
|---|---|---|
| Hold values when a control is outside the contracted range | On | On · Off — even when the CRO marks it “Pass” against its own SOP |
| Flag replicate variability above | 20% CV | Up or down in steps of 1% |
| Minimum recovery for permeability | 70% | Up or down in steps of 5% |
| Accept a compound alias when the manifest row matches | On | On · Off — otherwise the data steward decides |
| Keep qualified values as reported | On | On · Off — “> 60 min”, “< LLOQ” never become plain numbers |
| Draft a CRO query for every held value | On | On · Off — you approve before it is sent |
| Who decides compound identity | Sam Patel | Any named person on the team |
| Unit rules | 6 rules | e.g. nm/s → 10⁻⁶ cm/s (× 0.1) · Add a rule |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its fictional sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate, before measurement. People, CROs, compounds and studies named on this page are characters and examples in the working solution.
It is the work of getting results from contract research organizations — reports, results workbooks and instrument exports — into your own research data systems in a consistent, trusted form. CRO Data Intake does it with agents that classify each package, extract every value into your house schema, normalise units and compound IDs, check the data against the contract, and load it with a citation to its source.
The classifier reads every file in a package — report PDF, Excel workbook, CSV instrument export and cover email — and identifies the CRO, study, assay and contract schedule. The extractor reads tables with their page or cell coordinates, so a value from a PDF table and a value from a workbook cell are both traced to their source.
Units are converted only by approved rules you keep in Settings — for example mL/min/g protein to µL/min/mg, nm/s to 10⁻⁶ cm/s, or % bound to fraction unbound. Compound names are matched to the registry through the manifest shipped for the study; a name that does not match goes to the data steward with the most likely match and the reason.
Yes. The study-spec checker tests every acceptance criterion in the statement of work — controls, replicate variability, recovery, integrity markers, time points, compounds delivered and turnaround. When the CRO’s own SOP is looser than the contract, the contract range wins.
Yes. The agents propose a decision for every exception; the study monitor or data steward decides. A study loads only when every exception is decided, held values stay out of the warehouse, and no query goes to a CRO until a person approves it.
They stay as reported. “> 60 min” or “< LLOQ” never become plain numbers, and zero is never written for a value below the quantification limit.
It loads GLP study values as a convenience copy for research use, labelled as such. The signed report and the test facility’s archived dataset stay the record, and every value cites the signed report page.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your statements of work, house result schema, unit rules and a sample of past CRO deliverables — and it goes live once every quality gate has passed. We will walk you through it on your own deliverables first.
Ask the literature in plain words and get a sourced answer in seconds.
1-min filmPipeline WatchEvery rival trial, readout and approval lands in a sourced brief for each programme, every Monday.
1-min filmPreclinical ReportsRaw study data becomes a draft report in hours, with every number traced to its source.
1-min filmWe’ll run CRO Data Intake on a sample of your own CRO deliverables and contracts.