Case study: Gedeon Richter's Regulatory Intelligence Triage, and What Four Clicks Measure
Clarivate's case study reports one number: more than four clicks saved per document. The rest of the value, and all of the legal risk, sits in what happens after the alert lands.
EMA's PRAC recommendations on signals from the meeting of 31 August to 3 September 2026 carry an expected publication date of 28 September 2026. One asks every holder of an oxacillin product to file a variation within two months of publication, adding a warning on a severe skin reaction called DRESS. The document says PRAC timelines apply to generic products as well as originators. The clock runs from publication.
That is the problem regulatory intelligence software exists to solve. Clarivate puts the daily flow of regulatory and industry documents in the thousands, and only a handful start a deadline for your products. Good regulatory change monitoring is judged by what reaches the right owner in time, not by how much it reads. Clarivate's case study of Gedeon Richter, dated 2024, is a useful test of what such a tool does and what its vendor can show.
- Gedeon Richter, headquartered in Budapest, was one of the earliest subscribers to Clarivate's Regulatory Intelligence Tracking Application. Before it, staff checked each authority's site and emailed downloads to experts.
- The case study's only measured benefit is more than four clicks saved per document, across hundreds of documents a day. It counts interface steps, not hours, missed changes or the quality of impact assessments.
- In the EU the monitoring duty binds: Article 23(3) of Directive 2001/83/EC and Article 11(1)(f) of Implementing Regulation 520/2012 tie product information to what EMA publishes.
- A PRAC recommendation adopted on 3 September 2026 gives oxacillin holders, generics included, two months from publication to file a variation. Missing the alert does not stop the clock.
- Legal status is the hard part. On 12 October 2026 the new EU pharmaceutical directive is a Council first-reading position, not law, and FDA guidance binds nobody under 21 CFR 10.115(d).
The situation: thousands of documents a day, and email as the filing system
According to Clarivate, Gedeon Richter is a major pharmaceutical company headquartered in Budapest. It has a direct presence in Western Europe, mainland China, Latin America and Australia. Its portfolio spans women's healthcare, the central nervous system and cardiovascular medicine, and it is active in biosimilars.
The case study describes the starting point plainly. Thousands of regulatory and industry documents are released worldwide every day. Gedeon Richter teams spent hours triaging them by hand to find impacts and pass them on. Staff went to the sites of EMA, FDA and other authorities one by one, downloaded updates and emailed them to subject matter experts.
Clarivate lists what went wrong. Each region had its own storage, so a global picture was hard to build. Only the people on an email knew about a change. Nobody else could search for it unless someone built and maintained a store. People saved local copies, so the risk of working from an out-of-date version grew.
None of those failures is about reading. They are failures of distribution and record-keeping. An email reaches whoever was on the list, not whoever owns the affected product. The page gives no headcount, no count of sources and no measured hours.
What Clarivate deployed
Gedeon Richter adopted the Clarivate Regulatory Intelligence Tracking Application, which Clarivate describes as a companion to its Cortellis Regulatory Intelligence product. The case study lists six components:
| Problem the case study names | Component | What it changes |
|---|---|---|
| Staff visiting each authority site | AI that automates daily processing of new regulations and legislation | Collection |
| Regional silos | Dashboard filtered by region and topic | One view |
| Only email recipients know | Notifications and alerts to experts | Reach |
| No searchable record | Global searchable storage | Retrieval |
| Local, outdated copies | One store as the source for all departments | Version control |
| Manual newsletters | Automated alerts and newsletters | Distribution |
Three later features matter more than the list. Experts can now subscribe themselves to notifications an administrator has set up, which removes manual assignment. Auto-status mapping rules publish items from pre-publication triage to the daily updates in one step. And auto-prioritisation rules turn reviewers' answers to set questions into custom priorities, such as a flag for business impact.
Read together, these describe the real design. A reviewer triages the day's documents and answers structured questions. Rules turn the answers into a priority and publish the item. Experts who subscribed get the alert, and their impact assessments are recorded in the same tool. The AI sits at the front, processing new publications. The case study does not say what the AI does in detail, which model it uses or how accurate it is.
What it achieved, and what four clicks measure
The outcomes Clarivate reports are qualitative. Gedeon Richter has more visibility into regulatory changes and saves time it used to spend on manual triage. It reinvests that time in actions and impact assessments, and it can see which experts were assigned and which assessments are done. After many years of use, the company is expanding its use of the application. Its regulatory intelligence expert, Kinga Pallay, is quoted saying: "We are now able to efficiently triage and manage our regulatory alerts".
There is one number. Auto-status mapping saves more than four clicks per document, and Clarivate multiplies that across the hundreds of documents triaged each day. That is a count of interface steps, reported by the vendor. It converts to no stated time. It says nothing about the three things that decide whether monitoring works:
- Recall. What share of the changes that mattered for your products reached anyone?
- Latency. How many days passed between publication and the owner knowing?
- Outcome. Were the deadlines those changes started actually met?
A contrasting case in our library carries a time claim instead. Genpact built a generative AI monitoring system for a global consumer healthcare company in a six-week pilot. It scrapes sources such as FDA and DailyMed, translates them into English, and feeds dashboards and filtered newsletters. The company's experts tuned the prompts and checked the output. Genpact reports report creation falling from weeks to hours.
| Gedeon Richter | Consumer healthcare company | |
|---|---|---|
| Vendor | Clarivate | Genpact |
| Scope | Regulatory change triage and impact assessment | Regulatory and competitor intelligence reports |
| AI described | Daily processing of new publications, unspecified | Large language models, OCR, machine-learning tagging, translation |
| Human role | Reviewers set priority through set questions; experts assess impact | Experts tuned prompts and validated output |
| Number | More than four clicks per document | Report creation from weeks to hours |
| Measured by | Vendor, no baseline stated | Vendor, no baseline stated |
| Not shown | Time saved, recall, latency | Accuracy, recall, the baseline itself |
Genpact's number is closer to something a budget holder cares about. It also measures report writing, not change detection, and mixes competitor news with regulation. Neither page tells you whether anything important was missed.
What makes monitoring work: the triage rule, not the feed
Collecting publications is now the cheap part. The work is deciding, for each item, what its legal status is and whose product it touches. Gedeon Richter's auto-prioritisation rules point the right way: priority comes from structured answers, not from whoever happened to read the item.
Legal status deserves its own question, because the same week can bring items of very different weight. Position current at 12 October 2026:
| Item | Status | What it starts |
|---|---|---|
| PRAC signal recommendations, adopted 31 August to 3 September 2026 | Recommendation; the duty to keep product information current binds | Variation within two months of publication for oxacillin holders |
| New EU directive repealing Directive 2001/83/EC | Council first-reading position, listed for adoption on 28 and 29 September 2026; Parliament still to approve | Planning only |
| E2B(R3) format for postmarketing ICSRs to FDA | Required from 1 October 2026 under 21 CFR 314.80(g) | A system change already overdue if missed |
| FDA guidance documents | Not binding, under 21 CFR 10.115(d) | A decision on whether to follow, or how else to comply |
| FDA complete response letters on openFDA | Published agency correspondence; no obligation | Precedent for questions you may face |
A tool that files all five under "new EU/US regulatory update" has collected them and understood none of them.
The second rule is routing to a named person. Self-service subscriptions put the choice with the expert, which keeps lists current. They also mean an expert who never subscribed gets nothing. In regulatory data generally, named ownership predicts performance better than technology does, as our review of the 2026 RIM benchmark found. Monitoring is no different. Every product and every source needs an owner who is accountable for acting on what arrives.
The third use is less obvious: answering health authority queries. FDA now publishes complete response letters in an openFDA database, including letters on applications it did not approve. Reading what reviewers asked other sponsors about similar products is cheap preparation for your own questions. Our piece on health authority query management argues for measuring your own query effort before automating. A precedent library routed to the person drafting the response is one of the few inputs that shortens that effort without touching the submission itself.
The rules it had to satisfy
Position current at 12 October 2026.
In the EU, the monitoring duty binds. Article 23(3) of Directive 2001/83/EC requires the holder to keep product information up to date with current scientific knowledge. That includes the conclusions and recommendations published on the European medicines web-portal. Article 11(1)(f) of Implementing Regulation 520/2012 requires quality system procedures for that update, based on "continuous monitoring by the marketing authorisation holder" of the portal. The PRAC document itself reminds holders of Article 23(3), adds Article 16(3) of Regulation 726/2004 for centrally authorised products, and says EMA's website currently acts as the portal.
The two-month timeline is a PRAC recommendation, not a legal text. For nationally authorised products, the document says national authorities oversee adherence. In practice, missing a recommendation is a direct route to breaching the binding duty. If you want the wider monitoring picture on the safety side, our note on EudraVigilance signal duties after Regulation 2025/1466 covers it.
What does not bind is just as important. The reform that will replace Directive 2001/83/EC is not law on 12 October 2026. The Council's item note of 21 September 2026 records the provisional agreement and puts the Council's first-reading position up for adoption. The Parliament still has to approve it at second reading. In the US, 21 CFR 10.115(d) says guidance documents do not bind the public or FDA. An alternative approach must still comply with the statutes and regulations.
The tool itself faces lighter rules. No regulation requires regulatory intelligence software. FDA's January 2025 draft guidance on AI credibility excludes AI used for operational efficiencies that do not affect patient safety, drug quality or study reliability, as our analysis of the framework's scope explains. Monitoring sits inside that carve-out. The impact assessment does not always. Once it drives a label change or a manufacturing change, it becomes a record your quality system relies on. Treat the assessment, not the alert, as the controlled document.
What this means in practice
These steps run in lead-time order, longest first.
- List the clocks you owe. Write down every publication type that starts a deadline for your products, starting with PRAC signal recommendations and Federal Register notices. Give each an owner and an internal response time shorter than the external one.
- Map products to sources. For each active substance, list the authorities and portals that matter. Include generics. The oxacillin recommendation shows they are in scope by default.
- Measure a baseline. Take last quarter's PRAC recommendations and check which reached the right owner, and how many days it took. That gives you recall and latency, the two numbers no vendor page above reports.
- Write the triage questions. Copy the pattern behind Gedeon Richter's auto-prioritisation: is it binding, which product does it affect, is there a deadline, who owns it? Make the answers drive priority and routing.
- Then pilot the tool or the AI. Test classification on your own past items, with recall and precision scored against what your experts decided. Ask vendors for recall on your products, not clicks saved.
- Add the precedent library last. Once routing works, feed complete response letters and assessment reports to the people drafting health authority responses.
Prototype this in two weeks
A monitoring prototype on a month of your own regulatory alerts. It first measures how long each relevant item took to reach its owner, then sorts new publications by legal status and affected product, routes each to a named owner, and starts a deadline wherever one applies.
Pharmasight puts pharma subject-matter experts next to the engineers, so the prototype is built on a sample of your own material and checked against the rules it will have to pass, from the first day.
Questions people ask about this
- What does regulatory intelligence software do in pharma?
- It collects new publications from health authorities and legislators every day, sorts them by region and topic, routes them to subject matter experts and stores the outcome. In Clarivate's case study of Gedeon Richter, AI handles the daily processing of new regulations, reviewers set priority by answering set questions, and experts record impact assessments in the same tool. The legal duty to act stays with the company.
- Is there a legal requirement to monitor regulatory changes?
- In the EU, partly yes. Article 23(3) of Directive 2001/83/EC requires marketing authorisation holders to keep product information up to date, including recommendations published on the European medicines web-portal. Article 11(1)(f) of Implementing Regulation 520/2012 requires procedures for doing so based on continuous monitoring of that portal. No rule requires a particular tool.
- Do FDA guidance documents bind a company?
- No. Under 21 CFR 10.115(d), guidance documents do not establish legally enforceable rights or responsibilities and bind neither the public nor FDA. A company may take another approach, but it must still comply with the statutes and regulations. Requirements that do bind come from statute, regulation, and notices issued under a regulation, such as the E2B(R3) format required from 1 October 2026.
- How does regulatory intelligence help with health authority queries?
- It is the precedent library. FDA now publishes complete response letters, including for applications it did not approve, in a public openFDA database. EMA publishes PRAC recommendations every month. Reading what regulators asked other sponsors about similar products is the cheapest preparation for your own questions, provided the material reaches the person drafting the response.