From a six-step manual drug lookup to a single automated research run

Client

Date

€2B+ global pharma and cosmetics packaging group

08/2026

The challenge

A €2B+ global packaging group supplies containment and drug-delivery systems to the pharmaceutical and cosmetics industries. Deciding which devices to build means knowing what a drug's formulation demands of the hardware. Its laboratory head ran that as a six-step manual sequence, per drug, and across the group every research question was a person reading documents one at a time.

The solution

FifthRow deployed a research automation layer across five functions: innovation, laboratory, medical affairs, digital health and commercial. A drug name, a device question or a market goes in; a structured, source-backed profile comes out, with the manual sequences the teams already ran encoded as reusable systems.

7,212

analyst-hours equivalent

4,266

verified sources, drug profile library

Building a drug-and-device profileOne run replaces six manual steps

Drug profile research99 runs, 4,266 verified sources

Who could run research16 people across five functions

Research systems in production77 systems, 86 research jobs

Analyst-hours equivalent7,212 hours saved; equivalent to just over 3 FTEs

01 — STRUCTURED PROFILE EXTRACTION

The deepest system on the account, and the one that replaced a named manual sequence. It takes a drug, or a supplied list of drugs, and returns the profile the device teams actually need.

Core Activities

Encoded the laboratory head's six-step manual sequence as a system: one input, one structured profile
Pulled the physical properties the device decision depends on from scientific publications and approval documentation
Built it to accept a supplied list, so a licensed commercial dataset runs in one batch
Ran it repeatedly across therapy areas assessed for device opportunity
Returned a comparable profile per drug, not a narrative, so runs stacked into a library

The Results

A structured, source-traceable drug-and-device profile library, assembled from 4,266 verified sources, that the device teams could query instead of rebuilding by hand each time a therapy area came up.

manual six-step lookups, one run

02 — CLINICAL AND REGULATORY RESEARCH

The questions that gate a device programme: what is in the pipeline, what has been approved, what patents are live and what regulatory pathway applies.

Core Activities

Built a pipeline analyser for what is in development, and a compiler for fast drug lookups
Connected directly to public clinical-trial registries: patient counts, durations and identifiers, not scraped prose
Built patent lookups on the public register, searchable by company, showing competitor positions
Covered device and formulation questions as teams raised them
Added trade and policy systems when supply-chain exposure became live

The Results

The clinical, regulatory and patent questions that used to sit with whoever knew the databases became something any team member could run, with the sources attached.

database questions anyone can run

03 — STAKEHOLDER AND CUSTOMER INPUT ANALYSIS

The group had qualitative material it was not using: an internal stakeholder survey, customer interviews, and unstructured patient discussion from public forums. This workstream turned all of it into structured findings.

Core Activities

Ran the internal stakeholder survey through a purpose-built system in one session, iterating rather than waiting on a report
Ran customer interview transcripts through structured insight and journey mapping that enriches or critiques findings, not summarises them
Mapped the patient journey, deriving device requirements from the treatment pathway rather than assumption
Built synthetic personas from public patient discussion, pressure-testing concepts before committing to primary research
Kept it inside the group's governed environment, not a general-purpose assistant

The Results

Qualitative material the group already owned was turned into structured, reusable findings in hours rather than sitting unanalysed, and the analysis could be re-run when the question changed.

1 day

unanalysed survey to re-runnable findings

04 — MARKET SIZING, WHITESPACE AND COMPETITOR SCANS

The commercial half of the question. Once a device opportunity looked technically viable, this workstream sized it and checked who was already there.

Core Activities

Built a company screen that filters a defined market to players meeting the group's criteria
Sized markets three ways – idea, domain, supplied dataset – matching method to what was known
Ran competitor analysis for pharmaceutical packaging, plus gap, similarity and emerging-trend scans field-wide
Ran whitespace and pain-point landscape analysis to find right to play, not just market size
Extended the same workflows to the cosmetics side and adjacent delivery-device markets

The Results

Sizing, competitor and whitespace answers produced in the same place as the technical research, so a device opportunity could be assessed commercially without commissioning a separate piece of work.

174

analyst-hours equivalent, per screen

What it proved

Sixteen people across five functions ran 308 research runs against 11,782 sources. Seventy-seven systems covered 86 research jobs, from profile extraction to patent lookups to whitespace analysis. The largest, a drug-and-device extraction encoded from a laboratory head's manual sequence, ran 99 times, roughly a third of the account, on 4,266 verified sources.

The group got the encoded version of research its people already did by hand. Across 308 runs the platform records 7,212 analyst-hours saved, the equivalent of just over three full-time analysts over the period. The larger return was routing, not speed: research stopped depending on whoever knew the databases, and the systems reaching real depth came from someone's written-down manual process – the deepest ran ninety-nine times across four months.

Results

Metric	Benchmark	Result	Improvement
Building one drug-and-device profile	Six manual steps per drug	One run, or a supplied batch	Six steps collapsed into one
People running own research	Routed through central innovation	16 people across five functions	Distributed rather than queued
Systems in production	Bought reports, manual desk research	77 systems, 86 distinct research jobs	Reusable library, not one-off answers
Analyst-hours saved	Manual desk research, document by document	7,212 hours across 308 runs	Equivalent of adding just over 3 full-time analysts

ENCODED METHOD

The system runs the group's own written-down research sequence, so the output follows how its specialists actually work rather than a generic template.

COMPARABLE PROFILES

Each drug returns as a structured profile rather than a narrative, and a supplied list runs in one pass, so results stack into a library.

REGISTRY-GROUNDED DATA

Trial and patent answers come back from the public registers themselves, carrying identifiers, counts and durations instead of prose gathered from search results.

GOVERNED ANALYSIS

Interviews, survey responses and patient discussion are analysed inside the group's own governed environment rather than pasted into a general-purpose assistant.

Recent case studies

See more proof

The intelligence function six business areas never had to staff

Client

$30B+ global electrification and automation group

Date

08/2026

596,505

analyst-hours equivalent, 150 FTEs

From one-off research requests to a standing new-business intelligence practice

Client

$500M+ global speciality chemicals manufacturer

Date

08/2026

382,406

analyst-hours equivalent, 155 FTEs

Matter More

Let's Talk