The problem
Export and regulatory operations ran on Excel, Notion, Trello and files spread across shared drives. For any given product in any given country, nobody could answer “where is this dossier” without opening several files and asking two people.
The real cost was not the searching. It was three things:
Certificates expired quietly. A GMP or CPP certificate lapses and you cannot sell in that market until it is renewed. That is revenue stopped, discovered late.
Every country moved its own goalposts. Requirements, tender windows and regulatory updates change per market, and someone had to track all of them by hand.
It could not scale. Adding a country or a product multiplied the coordination rather than adding to it. Growth was limited by how much a small team could hold in their heads.
What we built
A regulatory tracking platform, deployed on the group's own servers. No regulatory data leaves their infrastructure. Every registration is keyed on one thing: product x country x subsidiary. Open it and you see the submission status, the completion of each CTD module, which module is blocking, who owns it, and the deadline. That view did not exist anywhere before. Around it:
- Expiry alerts on every certificate. GMP, CPP, export approvals — each tracked with configurable warning thresholds, so a lapse is now something you schedule around rather than discover.
- A central document library. Group-level documents are uploaded once and referenced everywhere they are needed. Update the library and every linked dossier updates with it.
- AI monitoring of regulatory change and tenders. Agents watch the target health authorities and public tender publications, summarise what changed, and flag which open dossiers it affects, with a link to the official source.
- Document versioning and a formal review circuit, currently being rolled out.
How we got there
Before any of this we ran a full AI and automation assessment across their export and regulatory operations. We mapped every core activity, attached hours to it, and scored each opportunity on business value, feasibility, risk and whether anyone inside the organisation would actually own it. Two things came out of that scoring that shaped the build.
- 1
Start with the tracker, then let them look. The highest-scoring first move was not the most sophisticated one. It was making the pipeline visible. Once a team can see all their dossiers in one view, they identify their own next bottleneck far better than any external assessment can. Sequencing the visible thing first made everything after it easier to specify.
- 2
Some things are better done by people, for now. Partner financial assessment is a good example: most West African distributors are private companies without published accounts, so no automated credit score exists at any price. AI saves real time on collecting and formatting the file. The judgement stays with the team, and we said so rather than promising a score we could not deliver.
- 3
AI drafting of the scientific CTD modules is on the roadmap. It needs the underlying product data structured first, which is what the current phase is putting in place.
What it was built to do
These were the design targets the system was specified against:
- •Dossier status readable in under 10 seconds, without opening a file
- •A new dossier created with country requirements pre-loaded in about 10 minutes
- •Daily pipeline review in under 5 minutes
- •Market analysis in 15-30 minutes rather than several days
What’s next
Phase one is live. Document versioning and the formal review circuit are being rolled out now. AI-assisted drafting of the scientific modules follows, once the product data underneath is structured.