How I research AI-assisted workflows in complex B2B products

The product team was exploring how AI could support media planners in one of the most complex parts of their workflow: deciding where budget should go, which advertising channels to recommend, and how to justify those decisions.

At first, the opportunity space was broad. AI could potentially support targeting research, budget allocation, property recommendations, campaign generation, workflow automation, or reporting. But the team needed clarity on which use case would create the most meaningful value for planners and should be prioritized first.

This research helped turn a broad AI concept into a clearer product direction grounded in real user pain points.

Prioritizing AI around real workflow friction

To make the AI opportunity actionable, I grounded the research in existing survey data and planning workflow pain points.

The survey showed that the most time-consuming and complex tasks in media planning were concentrated around researching and determining targeting parameters, determining the best mix of advertising channels, allocating budget effectively, managing collaboration challenges, and handling task complexity across planning workflows.

Rather than starting with “What AI feature can we build?”, I focused the research around a more useful question: where are planners experiencing the most decision complexity, and where could AI meaningfully reduce effort or increase confidence?

1. Start with the workflow, not the AI. The goal was not to force AI into the product. The goal was to understand where the planning workflow created the most friction and where recommendation support could be genuinely useful.
2. Separate automation from decision support. Some tasks could be streamlined, but others required planner judgment. I wanted to understand where AI should automate, where it should recommend, and where it should help planners compare options and reason through trade-offs.
3. Prioritize trust and explainability. Because planners are accountable for their recommendations, any AI output needed to be understandable, adjustable, and defensible. A recommendation without rationale would not be enough.

My role was to translate research signals into product direction

As the lead UX researcher, I was responsible for turning survey findings and workflow pain points into product direction for AI-assisted planning.

My work included synthesizing survey responses, identifying the highest-friction planning tasks, comparing pain points across time consumption and task complexity, translating findings into potential AI use cases, prioritizing which AI direction best fit user needs, and communicating recommendations to product and design partners.

The work supported product planning conversations around where AI should be introduced first and how it could create value for media planners.

Part 1: Synthesizing the problem space

The first step was to understand which parts of media planning were creating the most friction. Survey responses showed that planners consistently struggled with tasks that were both time-consuming and complex. The strongest patterns were around targeting research, channel mix decisions, and budget allocation.

This was important because these tasks were not isolated. They were connected parts of the same planning challenge. Planners needed to understand the audience, choose the right channels, distribute budget effectively, and explain the logic behind their recommendations.

The research helped clarify that the opportunity was not just about making campaign setup faster. It was about supporting the decision-making process behind the media plan.

Part 2: Mapping time consumption and task complexity

After identifying the major pain points, I compared them across two dimensions: time consumption and task complexity. This helped separate tasks that were simply tedious from tasks that carried deeper decision complexity.

The analysis showed that targeting research was consistently one of the most complex tasks, determining the best mix of advertising channels appeared repeatedly as a high-complexity planning task, and allocating budget effectively was a major time-consuming task when connected to channel strategy.

This mapping gave the team a clearer way to evaluate AI opportunities. The strongest use cases were not necessarily the flashiest ones. They were the ones that directly addressed high-friction planning decisions.

Part 3: Translating findings into AI use case prioritization

Once the pain points were mapped, I translated them into possible AI product directions and evaluated how well each one fit the user pain points.

• Budget allocation system: This had the strongest fit. Budget allocation was one of the top time-consuming tasks. A system that could recommend how to distribute budget across channels based on campaign goals, predicted performance, and channel fit would directly address a major planner pain point.
• Property recommendation system: This had medium-to-high fit. A property recommendation system could support more granular optimization within selected channels. It would be valuable as a secondary layer, especially if connected to the budget allocation system.
• Campaign generator: This had medium fit. A campaign generator could streamline campaign creation, but it did not directly address the highest-friction planning tasks identified in the survey. It was better positioned as a complementary future capability after the core planning pain points were addressed.

This prioritization helped the team avoid jumping too quickly into a general AI campaign generator and instead focus on the use cases most connected to planner needs.

The research helped the team move from a broad AI opportunity to a clearer product development priority. Before the research, the AI opportunity could have gone in many directions: campaign generation, targeting support, recommendations, workflow automation, or budget planning.

After the analysis, the team had a stronger rationale for prioritizing AI-assisted budget allocation and channel mix recommendation.