Teva and "Proof of Value": Choosing Which AI to Roll Out Across 30 Plants

On 4 November 2026, IIoT World reported on a panel at the Industrial AI Summit 2026, where Maya Schushan-Orgad of Teva Pharmaceuticals explained how the group selects, tests and scales artificial intelligence across its regulated manufacturing network. The panel also included Adam Procopio (Merck) and Kristen Sauter (Adlib Software); the moderator was Rick Franzosa (Tech-Clarity). The IIoT World article notes that AI tools were used to summarize the content.

The context: 30 plants and about 300 technologies

Teva evaluated about 300 technologies in a single year through its open innovation program, "Teva Rise". It runs 30 plants, and operations account for roughly 70% of its cost structure. Efficiency gains on the factory floor are therefore central to its "Pivot to Growth" strategy, the shift from generic drugs to biopharma.

The starting point is not the technology but the problems. Operational experts identified the main pain points first: predicting product quality risks earlier, improving asset reliability and making better-informed supply chain decisions.

From proof of concept to proof of value

A proof of concept asks whether the technology works. Teva's proof of value adds five questions:

• Does it work in our environment?
• Does it integrate and scale across a heterogeneous network of plants?
• Does it deliver measurable value against the identified problem?
• Can it operate in a regulated context?
• Can users trust the output and act on it?

Each validation is designed together with the vendor, with decision-flow mapping and traceability requirements built in from the start. Teams check whether the data exists at the site, whether operators can embed recommendations in existing workflows without heavy change management, and whether performance metrics translate into business value. A site lacking data maturity or system readiness is deferred, even when the vendor is a good one.

A good vendor is not enough: if the site is not ready, the project waits.

The gap between vendors

According to the panel, vendors with pharmaceutical experience already build in regulatory requirements, explainability and traceability. Some vendors from other industries had better technology, but Teva could not adopt it because compliance and trust features were missing.

Traceability beyond compliance

The data trails that satisfy regulators also feed predictive maintenance models, quality management systems and operational improvements. Traceability designed in from the start is therefore not just a regulatory cost: it becomes a reusable asset.

What food plant managers can learn

This case comes from pharma, not food or poultry, and should be read as a transferable example rather than a direct recipe. It remains useful for plant and operations directors who run many pilots and must decide which to scale. Some practical takeaways:

• start from priority operational problems, not from the technologies available;
• define evaluation criteria in advance that go beyond "it works": integration, measurable value, user trust;
• ask vendors for traceability and explainability of results already at the pilot stage;
• accept deferring sites that are not ready, rather than forcing a premature rollout;
• reuse traceability data for quality and maintenance.

In short, Teva's lesson is that scaling AI is first of all a selection problem: a few well-designed pilots, with clear criteria and built-in traceability, count for more than many isolated experiments.