Roche developed a proprietary Investment Recommendation System designed to translate their deep institutional business know-how into a rigorous mathematical formulation. Roche built a specialised engine engineered to handle 70,000 daily permutations with a level of granularity standard systems cannot reach. The core of the solution is an Optimisation System, whose objective function is programmed to maximise the all‑in-spread above risk-free rates of the overall portfolio.

This includes interest rate differential arbitrage using projected interest rate curves and credit spreads based on live market data and historical transactions. The engine achieves this while simultaneously respecting a strict matrix of liquidity management constraints, such as counterparty limits, currency buckets and investment horizon targets.

The system generates actionable recommendations for investing and short-term funding by considering every facet of the investment decision. Inputs to the optimisation engine are, among others, a machine-learning based liquidity prediction model, which has also been built in‑house, information about counterparty risks, interest rate curves and FX swap costs for the major liquidity currencies. Furthermore, generative AI Large Language Models interpret the whole set of results and produce a curated, text-based explanation of the recommended actions.

Moving this optimisation logic into practice required deep domain knowledge and a technical data science team that turned this business idea into an enterprise-grade software solution. Under the guidance of Iñigo de la Maza, Digital Treasury Lead, who expertly coordinated the efforts between IT and treasury, the team achieved this award‑winning solution that is currently used on a daily basis. “It is an outstanding achievement that we developed this entirely in-house through close collaboration between treasury, data scientists and software engineers,” says Aramis Buscemi, Senior Treasury Manager and Treasury Front Office Lead of the initiative.