Most leadership teams have more data than ever, and an AI assistant that will summarise it in seconds. What is still rare is a decision you can defend: one whose sources, assumptions and risks survive a hard question in the boardroom.
Decision intelligence promises exactly that [1], but it only delivers when the model is built for one decision, in one market. A model meant for every decision ends up precise about none.
Should we bid for this lot?
Take one of the most consequential recurring decisions in European pharma. Tenders and rebate contracts decide which manufacturers supply a medicine, often for years. For every lot, generics, biosimilar and originator companies face the same questions: is it worth bidding, at what price, can we supply it, and what happens if we win, or lose?
A sound answer draws on four kinds of evidence:
The tender: buyer or purchasing group, rules, single or multiple winners, award criteria, lot structure, duration and volume commitments.
The market: prior awards, price benchmarks and the likely field of competitors.
The company: bid price against benchmarks, supply capacity, delivery record and eligibility.
The risks: a lot won at the wrong price, a lot lost that mattered, capacity tied up that is needed elsewhere.
Public data tells only part of the story. Award notices show who won, not the losing bids, the real prices or a company's own costs and capacity. The decision lives where public and private evidence meet.
Faced with forty relevant lots and capacity for twelve, the better question is not which lots we can manage, but which are worth it, with a reason beside every lot chosen and every lot left out.
One Europe, many markets
There is no single European tender market. German statutory health insurers award rebate contracts through exclusive and open-house tenders, and an exclusive contract often has more than one winner: in our analysis of award notices, around half or more of exclusive lots named several. Dutch insurers each set their own preference policy [2]; Sweden names a lowest-priced "product of the month", one month at a time [3]. A model trained on one country misleads in the next. We build country by country, starting with Germany.
What we are building
We are developing a research-led model for tender decisions that combines computational modelling, statistical learning and simulation. We know of no published, validated model that estimates the chance of winning at the level of an individual lot, so this is genuine research, not configuration.
The work runs in three strands. A governed evidence base: every public tender and award notice reconciled, versioned and traced to its source, with company names resolved into groups. Learning from history: models that learn how buyers, rules and competitor fields have shaped past outcomes, and estimate for each lot the chance of winning, what it would take and the risk carried, with honest uncertainty where the evidence is thin. Validation: predictions are tested against award outcomes held back from the model, and probed on invented company profiles before any client data is involved. Inside a client's environment, its own data stays there.
The tender manager remains in charge of every decision. The model's job is to make that judgement sharper, faster and easier to defend.
If this is a question your team is working on, we would be glad to compare notes.
References
Gartner, Market Guide for Decision Intelligence Platforms, 18 July 2024 (accessed 09.11.2026)..
Tweede Kamer, Aanhangsel Handelingen 2025–2026, nr. 2952: answers of the Minister of Health, Welfare and Sport to questions by Bushoff on the sale of non-preferred medicines, received 7 September 2026.
TLV (Dental and Pharmaceutical Benefits Agency, Sweden), "Product of the month" (accessed 09.11.2026).
Method
The figure for Germany comes from Latengram's own analysis of TED award notices for medicines published by German statutory health insurers, January 2024 to October 2026. The tender example is illustrative.
