KatalX.com Arrow right icon News & Views Arrow right icon Our Journey in GenAI

Our Journey in GenAI

In nearly every meeting with our customers, we’re faced with questions:

  • Are you exploring GenAI?
  • Does your product incorporate GenAI?
  • What are the applicable use cases for GenAI?

At KatalX, while we currently have two ongoing projects involving GenAI, our journey began with an analysis of how GenAI could generate new content from existing data for use in Supply Chain Management (SCM), with a particular focus on the Life Sciences sector.

Our analysis revealed that the utilization of GenAI is constrained by challenges related to explainability and model training. Below, we present a summary of our analysis along with suggestions for addressing these challenges.

We’ve classified the challenges into four groups:

Challenge #1:

  • Difficulty in identifying training data due to the complexity and opacity of SCM processes involving multiple parties, as well as the challenge of correlating data collected during business and logistical activities.


Challenge #2:

  • Dilemmas in explaining GenAI models’ compliance with legal frameworks and guidelines such as CFR21, GMP, GDP, or ISO27001.


Challenge #3:

  • Conundrums arising from the necessity to protect data privacy during logistical operations within the rapidly expanding market of Personalized Therapies. As these therapies are tailored to individual patients, it is crucial that logistic units (such as parcels and labels) do not reveal any patient-related data. While regulations like HIPAA (US), GDPR (EU), and LGPD (Brazil) have not traditionally been integrated into SCM practices, this is expected to change in the near future.


Challenge #4:

  • Difficulty in ensuring that GenAI produces accurate outcomes for supply chain operations. With personalized therapy costs often reaching tens or hundreds of thousands of dollars, using newly generated GenAI data for supply chain management products may pose significant risks.


Unsurprisingly, the remedies stem from underlying data analytics fundamentals:

  • the availability of high-fidelity data and
  • flexible framework that accommodates changing regulations, alongside applied data confidentiality practices.


Let’s review how each of these challenges can be addressed:

Remedy #1:

  • Implement a cointegration process for data in time, space, and business context to create repositories of high-fidelity data suitable for quantitative modeling. Cointegrated data dimensions also enable the reduction of required number of data observations to obtain meaningful output from GenAI models. Additionally, ensuring copyright protection for trained data allows for the use of derivative works from GenAI models.


Remedy #2:

Explore various approaches to explainability in the context of generative AI, such as

  • and preferred method Proxy Models. Proxy Models, in particular, are a common technique used in quantitative analysis in scenarios where regulators require understanding of decisions driven by black box models like LLM. Proxy Models necessitate the use of high-fidelity data.


Remedy #3:

  • Recognize that Data Confidentiality is transitioning into Supply Chain Management as a necessity rather than merely an option. Protecting data on a need-to-know basis through flexible configuration will determine the effectiveness of supply chain management platforms in personalized therapies.


Remedy #4:

  • Ensure the viability of GenAI outputs by applying feedback constraints anchored in financial, sustainability, and resiliency goals. Supply Chain Management may need to utilize a combination of GenAI and more traditional models to ensure alignment with the organization’s business goals.


While our journey to prove the value of generative AI continues, we’ve successfully applied most of the remedies listed above. We aim to share our successes to benefit supply chain management customers and ultimately patients themselves.

We encourage you to reach out to us to share your opinions on how our industry can further improve supply chains, and hopefully serve a long-term vision of fostering “antifragile” supply chains: supply chains that effectively reinforce themselves with shocks and disruptions.


Tom Z.,