A behavioral framework for automated recommendations

The Challenge

Engineers and operators in oil production make decisions every day that affect how wells run. Our client gave them two separate platforms: one for monitoring and diagnosing wells, and another for optimization recommendations. Built by teams working in silos, the two tools forced users to switch back and forth to finish a single task.

The tools were also dense. Operators spent hours in data-heavy dashboards, which caused fatigue. But those dashboards were familiar, and people relied on them. Any redesign had to lighten the load without breaking the routines users depended on.

Then there were the recommendations. They contained useful guidance for improving production, yet most users ignored them. They didn't trust them enough to act on them, and often didn't check them at all.

The goal was to bring both platforms together into one experience, ALLY, and turn automated recommendations into something engineers would actually use.

The Approach

Two constraints shaped the research: the platforms came from teams that rarely talked to each other, and we had no direct access to end users early on. So the approach had to answer two questions. How does the work actually happen across both tools? And how do people decide what to act on?

1. Discovery (2.5 weeks). I interviewed stakeholders and both product teams, ran desk research, and mapped an as-is journey for each platform. Seeing the journeys side by side showed exactly where a single task broke across two tools.
2. Learning from people close to users. Without direct user access, I worked with the client's internal engineers and operators, and with SMEs from Customer Experience and Support who trained users and took their support calls.
3. Archetypes. The client already had a lot of user knowledge, but it wasn't usable. I reorganized it into archetypes along two dimensions: leaders vs. practitioners, and day-to-day work vs. oversight. This showed which users needed analytics and which needed task-focused tools.
4. Alignment workshop (1 week, including planning). I designed the prioritization exercises and co-facilitated a co-creation workshop where both product teams agreed on priorities, KPIs, business needs, and a first version of the ideal journey.
5. Concept and design (4 weeks). We defined the information architecture and the ideal journey, then turned them into design concepts.
6. Validation. We ran usability testing with eight participants, including engineers and operators.

Throughout, I focused on how people read information, judge a recommendation, and decide what to do next. I then zoomed in on how engineers analyze a recommendation before acting on it. That behavioral lens is what shaped the framework.

My Contributions

Understanding the system

• Led the research: interviews with stakeholders and both product teams, desk research, and an as-is journey for each platform.
• Worked around the client's internal engineers and operators, and from the SMEs (Customer Experience and Support teams)
• Reorganized the client's existing user knowledge into archetypes that linked each type of user to the kind of tools they needed.
• Designed all the prioritization exercises and co-facilitated the workshop that brought two siloed product teams to a shared structure.
• Brought the Executive Design Director in at the moments that needed direction or approval, such as presenting findings and aligning the two teams.

Redesigning the experience

• Created the behavioral framework for recommendations and turned it into card components in several sizes.
• Developed the initial design concepts for the unified navigation, custom views, and recommendations, taking them close to final design using the design team's guidelines, ensuring the strategic intent and behavioral syntax were locked in.
• Handed the concepts to the design team, who refined and finalized the interfaces.

Outcome

The project produced a shared view of the full workflow and a redesigned scheduling experience ready for AI. The blueprint pointed to clear opportunities: automating repetitive tasks, reducing manual effort and back-and-forth between teams, and flagging dependencies before they disrupt a patient’s schedule. The redesigned screens and patterns turned those findings into an interface that’s simpler to navigate and built for schedulers and AI to work together.