A platform for commercial and medical affairs

Decision-grade analytics for life sciences, in days, not months.

OBI-HUI is the platform pharma commercial and medical affairs teams use to defend their next decision. Eight purpose-built components, specialist agents grounded in named datasets, and an audit log that traces every figure and statement to its source — across two firewalled tenants, on one architecture.

Request a 30-minute walkthrough → Read the methodology brief

Built with forecasting and commercial-insights leads from six top-20 pharma companies

Profiled in [Industry publication] [Industry publication] [Analyst note]

Built for regulated workflows

SOC 2 Type II in progress HIPAA-aligned controls EU & US data residency 21 CFR Part 11 alignment Customer data is never used for training

The work today

The same three failure modes — across commercial and medical affairs.

Whether the deliverable is a launch forecast, a digital twin, a publication, or an AdBoard synthesis, the pain is structurally the same. OBI-HUI was built directly against it.

"It's an eight-week Excel build that breaks in two days."

Fourteen linked tabs, a brittle assumption layer, and one analyst who knows where the bodies are buried. When the deal team needs a sensitivity, it starts over.

"The consultants own the work — and the next invoice."

Forecast, digital twin, publication plan, AdBoard synthesis — it leaves the room when they do. You pay again for every refresh, every revision, every scenario. Institutional knowledge never lands in-house.

"Medical content gets stuck behind eight reviewers and a folder no one can find."

Every publication, MIR response, and AdBoard summary loops through MLR with no shared workspace, no provenance trail, and no way to surface the same insight the next time it's needed.

What OBI-HUI does differently

Three things, done with rigor — not twelve features.

The product sits between your data and your decision. It is opinionated about what to make easy: visual modeling, transparent AI, and provenance on every figure.

01 · Visual workbench

A canvas, not a spreadsheet

Every component on the platform follows the same shell — blocks you can drag, edit, swap, and A/B without rebuilding the model. Whether you're forecasting a launch, modelling a digital twin, or running an AdBoard simulation, the canvas is the same.

IN Incidence sourcer Done

SE Segmentation Running

AM Analog matcher Done

AU Audit trail writer Idle

02 · Specialist agents

Named agents, each with a job and a source

No "AI" black box. Each agent has a defined role — sourcing, structuring, synthesising, drafting — a named dataset it reads from, and a human reviewer who signs off before the output enters the model or the document.

14:02 SOURCE SEER 2024 · cohort N=236,740

14:03 DERIVE PD-L1 high = 28.4% (cohort)

14:05 REVIEW Approved by B. Diop

14:08 COMMIT Scenario A · v1.4

03 · Audit log

Every figure and statement traces to source

Click any number in a model — or any sentence in a medical document — and walk the chain: which agent produced it, which dataset or paper it came from, which assumption or claim was overridden, who signed off, and when. Built for the review you can defend.

What's in the platform

Eight apps. Two tenants. One architecture.

Five commercial apps and three behind the firewall in Medical Affairs. Each is a workbench in its own right — and each follows the same shell, the same agent pattern, the same audit log.

Commercial 5 apps · Commercial tenant

Forecasting

A visual revenue-forecasting workbench: patient-flow canvas → segmentation → uptake → price → 10-year trajectory.

Digital Twins

Synthetic patient cohorts grounded in RWD, used to stress-test trial designs and commercial assumptions.

Qual Research

Transcripts to theme / quote / sentiment trees, with every claim traceable to the speaker turn.

Modular RWD

Compose longitudinal patient cohorts from claims and EMR sources with auditable inclusion criteria.

Whitespace

Map competitor portfolios across indication, mechanism, and segment to surface unaddressed pockets.

Medical Affairs 3 apps · Separate tenant Firewalled

Medical Repository

Pulls primary literature, trials, and conference proceedings into a structured evidence base with named citations.

AdBoard Simulator

Simulate KOL panel responses to a discussion guide, using each KOL's published views and conference history.

Document Generator

Drafts MIR responses, MSL briefings, publications, and congress summaries — every sentence linked to a source.

Inside one of the workbenches

A single screen replaces fourteen Excel tabs and a deck.

Every component on the platform follows the same shell: scenario, assumptions, agents, audit trail. Below, the Forecasting workspace running an NSCLC US 10-year forecast — one of eight components, shown here for depth.

1 2 3 4

workbench / NSCLC · US · 2026–2035 · scenario A · v1.4 Saved

NSCLC US revenue forecast

10-year horizon · last run 14 min ago · 4 agents active

Peak revenue

$3.2B

▲ 8.2% vs. prior · 2031

Cumulative

$18.7B

2026–2035

Peak treated

142K

patient-years

Confidence

High

86% inputs validated

Revenue trajectory · 3 scenarios

Last sync · 14 min ago

Base case Upside Downside

Five commercial workspaces Forecasting, Digital Twins, Qual Research, Modular RWD, Whitespace. Medical Affairs runs in a separate, firewalled tenant — surfaced at the bottom of the sidebar.

Versioned, signed scenarios Every forecast is named, versioned, and frozen on commit. The breadcrumb shows exactly which artefact the user is looking at.

KPIs trace to source Click any figure to walk the chain back to the agent that produced it, the dataset it came from, and the reviewer who approved it.

Inflection points labelled inline Peaks, plateaux, and divergences are annotated in the chart itself. No exporting to PowerPoint just to mark a number.

How the AI works

Specialist agents, mapped to apps.

Each app on the platform is powered by named, scoped, reviewable agents — never a chatbot. Below, six representative agents: three commercial-side and three behind the firewall in Medical Affairs. Each anchors one of the eight apps, with a named dataset, a structured output, and a human reviewer who signs off.

Uptake modeler

Powers Forecasting Commercial

Selects analog launches matched on indication, mechanism, and market context, and fits an uptake curve with named assumptions and sensitivity bands.

Grounded in: IQVIA launch panel, FDA approvals, internal analog library
Outputs: Y1–Y5 share curve with sensitivity bands and the analog set cited
Reviewed by: Forecasting lead

Cohort builder

Powers Modular RWD Commercial

Assembles longitudinal patient-level cohorts from licensed claims and EMR sources, with named inclusion criteria and an auditable derivation chain.

Grounded in: Flatiron, Komodo, IQVIA claims, Symphony
Outputs: cohort.parquet with patient-month rows and full derivation lineage
Reviewed by: RWD analyst

Theme synthesizer

Powers Qual Research Commercial

Ingests interview and focus-group transcripts and extracts themes, quotes, and sentiment trees — every claim traceable back to the exact speaker turn.

Grounded in: Customer transcripts, persona library, prior research
Outputs: theme_tree.json with quotes, timestamps, and speaker attribution
Reviewed by: Insights lead

Evidence synthesizer

Powers Medical Repository Medical Affairs

Pulls primary literature, registrational trials, and conference proceedings on a clinical question, and structures the evidence with named citations and strength-of-evidence ratings.

Grounded in: PubMed, ClinicalTrials.gov, conference proceedings, internal medical literature
Outputs: evidence_synthesis.md with linked citations and strength-of-evidence ratings
Reviewed by: Medical writer

AdBoard orchestrator

Powers AdBoard Simulator Medical Affairs

Simulates KOL panel responses to a discussion guide, using each KOL's published views and conference history to produce a synthesised transcript with per-speaker attribution and confidence.

Grounded in: Published literature, conference recordings, prior AdBoard transcripts (with consent)
Outputs: simulated_adboard.md with per-KOL attributions and confidence ranges
Reviewed by: Medical Affairs lead

Document drafter

Powers Document Generator Medical Affairs

Drafts MIR responses, MSL briefings, publication first-drafts, and congress summaries from a brief plus the Medical Repository's structured evidence — every sentence linked to a source.

Grounded in: Medical Repository, customer brief, brand voice guide
Outputs: draft.docx with sentence-level citations back to source
Reviewed by: Medical writer + MLR

Methodology & data foundation

What we ground on — and what we don't do.

Two columns, written plainly. If a buyer's data-science team asks "what are you doing under the hood," this is the answer.

What we ground on DO

Primary epidemiology SEER, GLOBOCAN, national registries — refreshed quarterly, cited by row.
Licensed real-world data Flatiron, Komodo, Symphony, IQVIA — vendor of record disclosed per derivation.
Customer-owned datasets Your contracts, your analogs, your historical models — used only inside your tenant.
Frontier LLMs for structuring, not for facts Models extract, normalize, and structure. They do not invent figures. Every output is grounded.

What we don't do DON'T

Train on your data Customer data is never used to train shared models. Contractual and architectural.
Output un-cited figures If an agent cannot produce a source row, it returns an open question, not a number.
Hide assumptions in a black box Every assumption is named, editable, and reviewable by the user who owns the scenario.
Replace your forecasting team The agents do the sourcing and structuring. The judgment — and the sign-off — stays with your people.

Workflow integration

Slots into the stack your team already runs.

OBI-HUI is not a replacement for Excel, your BI stack, or your CFO model. It sits between them and makes the connection auditable.

Tenant architecture

Medical Affairs runs behind a firewall. By design.

Promotional and non-promotional work cannot share data, agents, prompts, or audit trails. The separation is contractual, technical, and visible in the product — not a policy memo.

Commercial tenant

For forecasting, RWD, qual research, and commercial insights

Components: Forecasting, Digital Twins, Qual Research, Modular RWD, Whitespace
Data sources: IQVIAFlatironSymphonySEER internal analog library, congress data
Agents: Uptake modeler, twin simulator, cohort builder, theme synthesizer, whitespace mapper, audit writer
Outputs: CFO model, deal-team brief, LRP review, scenario library
Sign-in: SSO via your commercial domain — provisioned by commercial admin

Firewall
No automatic
data crossover

Medical Affairs tenant

For evidence, KOL strategy, and medical communications

Components: Medical Repository, AdBoard Simulator, Document Generator
Data sources: PubMedClinicalTrials.govregistries congress proceedings, internal medical literature, de-identified RWD
Agents: Evidence synthesizer, KOL mapper, AdBoard simulator, document drafter, citation verifier
Outputs: MSL briefings, AdBoard transcripts & insights, medical publications, MIR responses, congress materials
Sign-in: SSO via your medical domain — provisioned by medical admin only

Separate cloud tenants Different VPCs, different encryption keys, different audit logs. No shared storage.

No shared embeddings or prompts Agent context is scoped per tenant. A commercial run cannot leak into a medical scenario.

Personnel access by SSO domain Commercial users cannot authenticate into the medical tenant. Joint roles require two sign-ins.

Aligned to industry codes PhRMA Code, EFPIA Code, 21 CFR Part 11 audit. Annual attestation available to your compliance lead.

Outcomes

What teams using OBI-HUI report.

Three representative results from pilot engagements. Numbers shown are illustrative ranges from in-flight programs — final figures shared under NDA on request.

8×

faster forecast cycle vs. the spreadsheet build · Commercial

The thing that used to be a quarter is now a week. The deal team stopped waiting on us.

— Forecasting lead, top-10 pharma (anonymized)

4×

more AdBoards run per quarter at a fraction of the cost · Medical Affairs

We can pressure-test a clinical question against simulated KOLs before we ever convene a real panel.

— Head of Medical Affairs, oncology biotech

100%

of figures and statements traceable to source in audit · Platform

For the first time the review didn't end with "let's get back to you on that number" — or that citation.

— Head of finance & head of MLR, specialty pharma

Illustrative ranges. Real figures from in-flight engagements shared under NDA — ask in the walkthrough.

Governance & trust

What your security and compliance teams will ask.

A short, honest answer to each. Full security pack and architecture diagram available on request.

Security & compliance

SOC 2 Type II audit in progress
HIPAA-aligned technical controls
21 CFR Part 11 alignment for audit trail
EU and US data residency options
Pen-tested annually by a CREST-accredited firm

Data & IP

Per-customer tenant; no co-mingling
Customer data never trains shared models
Customer owns all outputs, exports, and IP
Zero-retention LLM configuration by default
Bring-your-own-key encryption for sensitive cohorts

AI governance

Every agent has a published model card
No agent output enters the model without human review
Citations required on every generated figure
Confidence and uncertainty surfaced — never just point estimates
Aligned with EU AI Act high-risk system controls

A 30-minute walkthrough with the team that built it.

Benjamin Diop, forecasting lead. Bring a real launch question — we'll model it live and leave you the scenario.

Request a walkthrough →