In recent months, Forrester has explored how AI-powered innovation is reshaping experimentation, content creation, and product development across industries.

In this context, we recently had the chance to sit down with Sonia Fife, Global Leader, Consumer Package Goods, Strategic Industries GTM at Google Cloud; Lauren Milne, Chief Strategy & Growth Officer at APPLY; and Justin Thomas, Chief Digital Officer at Aptar to discuss their key insights about AI-driven innovation, from experimentation workflow reinvention at scale.

Our discussion focused on three questions relevant to any innovation leader looking to scale AI:

  1. What determines whether AI initiatives scale or stall?
  2. Which methods help organizations design for adoption, not just capability?
  3. How should teams approach data readiness without slowing momentum?

Usability Is As Important As Model Capability

When it comes to adoption at scale, user experience is equally important as the power of the underlying AI model. It turns out that no matter how promising the outcome, users will abandon the process if the path to get there remains cumbersome.

  • Ease of model interaction. In one example of AI-assisted content generation, the project team observed that output quality varied significantly across users, even when underlying models and data sources were unchanged. Varying user ability to express intent through prompts created the risk of inconsistent brand execution and an over-reliance on a small group of highly skilled users. Both limited scale.
  • Expressing intent beyond the prompt. To address this, the team shifted from prompt exposure to prompt abstraction. Complex prompting logic was embedded directly into predefined workflows aligned to common tasks such as content creation, localization, and refinement. Users interacted through constrained controls rather than free-text prompts.
  • Simplicity drives scale, speed, and consistency. This design change produced three measurable effects: Faster onboarding and reduced training dependency. More consistent outputs across roles and regions. And higher repeat usage beyond creative specialists.

Co-Creation Is Central To Scalable Usability

Generative AI creates value through workflow reinvention – not task automation alone. Successful teams redesigned technology, human roles, and work sequences together.

  • Joint problem definition and solution design. Rather than deploying generic tools, they identified high-value domains and jointly redesigned how work should happen, grounded in stakeholder empathy and real usage observation.
  • Unlearning what usability used to look like. When asked about editing AI-created visual designs for example, the group stressed that the user base in the redefined workflow changed completely. Recreating established features of traditional digital image creation software was not the path to go. Instead, they redesigned the interfaces and features jointly with their target users.
  • Observing outcomes and redesigning model interaction. Initially, their content editing process was an open-ended chat experience, similar to most standard LLM However, they observed consistent elements that users wanted to change, which led to the creation of a simpler user experience and functionality with one-click changes instead of requiring chat prompts. By taking this approach, they found they could execute 50 to 60 times the prompting of an individual user, allowing non-savvy users to reach the point of value faster.

Minimum Viable Data Beats Enterprise-Wide Readiness

The discussion challenged the assumption that generative AI requires uniformly clean, enterprise-wide data. Teams that waited for 12–18‑month data cleansing programs lost momentum.

  • Minimum viable output to prove value. To get an MVP for on-brand content creation assembled quickly, early pilots focused on a limited number of brands and regions. The team focused only on the essential data that would enable marketing teams to produce on-brand visual and written content more quickly and at lower marginal cost. Consequently, the team prioritized data sources with direct impact on output accuracy and brand risk.
  • Minimum viable data to get results. Rather than boiling the data ocean, the team focused on the data assets that made a real difference in model output. Together with Google, they implemented a RAG-based architecture that grounded generation in a select number of curated internal data assets, such as brand guidelines, product data, and regional constraints.
  • Expanding and refining as you scale. Gaps were addressed incrementally as usage expanded. For example, getting the right data to improve the visualization of product specificity in AI generated videos proved significantly more valuable than chasing the next level of granularity of financial data.

What’s Next For AI-Driven Innovation?

  • From experimentation to end-to-end innovation orchestration. Organizations are moving beyond isolated AI use cases toward agentic systems that span the full innovation lifecycle, from trend sensing to portfolio prioritization and scenario testing. Unified agentic data foundations with contextual persistence and multi modal integration become critical enablers; as well as goal oriented planning and tightly integrated security and governance.
  • Continuous idea testing: Rather than maximizing output volume, teams increasingly use AI to assess desirability, feasibility, and viability early, including through synthetic personas grounded in customer data, at all steps of the innovation process.
  • Expanding the sensory frontier. Model capabilities are rapidly advancing into domains such as taste, scent, and material properties, opening new opportunities in consumer products, beauty, and fashion
  • Agentic Innovation Flows: All parties agreed that the next step in AI-powered innovation is moving from phase-specific solutions to agentic innovation lifecycle orchestration. Google highlighted their recent work with Oxford University an agentic AI co-scientist They see such models applied in various domains going forward, including holistic innovation lifecycle management.

Key Takeaways

1. Design Agentic AI systems around user workflows, not prompt expertise
2. Co-create interfaces and processes with target users to drive adoption
3. Prove value with minimum viable data—do not boil the data ocean

About Aptar

Aptar is a global leader in drug delivery, dosing and protection technologies, and consumer product dispensing. Aptar partners with the world’s top healthcare and consumer brands to deliver medicines and create exceptional user experiences. Serving diverse markets, from pharmaceutical to beauty to food and beverage, Aptar combines market expertise with proprietary design, engineering and science to develop innovative solutions that help improve lives worldwide. Headquartered in Crystal Lake, Illinois, Aptar employs 14,000 dedicated people across 20 countries.

About APPLY

APPLY is a global Agentic Customer Experience (ACx) partner for ambitious brands across CPG, retail, sports, and media. Apply has worked with brands like Arc’teryx, NFL, Lululemon, and Kraft Heinz.