In the previous unit, you identified which moment that matters deserves AI attention first — using Iris McQuillan-Grace's criteria of highest friction, lowest adoption, and inability of manual effort to close the gap. But knowing where to focus isn't the same as knowing what to build. As Iris walked through in the session, Oliver didn't jump from "performance management is broken" to "let's build a bot." They zoomed in first — breaking the moment into discrete journey steps and building personas for each employee segment involved — so the intervention they designed actually matched the problem they'd diagnosed.

You'll recall that Iris described a deliberate layering process: first, identify the moments that matter; then, break each one into the specific "steps each employee is taking in those moments that matter"; and finally, build employee personas asking "which employee segments are involved in the steps of the moment that matter." At Oliver, this meant mapping performance management not as a single monolithic process, but as a sequence of discrete interactions — each experienced differently by new managers needing basic guidance, established leaders navigating nuance, ICs anxious about receiving formal feedback for the first time, and HRBPs stretched across all of them. This layering is what prevents you from building a one-size-fits-all solution that technically works but practically misses. A new manager asking "how do I write constructive feedback?" and a senior leader navigating a difficult performance conversation have fundamentally different needs, anxieties, and readiness levels. Without personas, you'd design for an average employee who doesn't exist.

The journey mapping and persona work didn't just clarify what to build — they surfaced why only AI could solve this particular problem. As Iris described, Oliver's engagement data, one-on-one interviews, and focus groups all confirmed that feedback was necessary, but the L&D team's training was "only getting like 40% of our employees engaging." Meanwhile, HRBPs were "coming to us very nicely with shivs in their teeth" because they were "spending hours coaching to behavior for new managers, established managers, ICs" who knew formal feedback collection was coming and "were very anxious about it."

The critical insight was that even unlimited budget wouldn't have solved this. As Iris put it, "I can't bring 60 coaches into the business" — and even with a blank check, "I wasn't going to be able to integrate them into the culture, build the relationships, get them up to speed." This is where the human-in-the-loop decision becomes concrete: scaled feedback coaching — helping hundreds of managers draft better written feedback — is exactly the kind of repeatable, high-volume need where AI outperforms headcount. But the complex, relationship-dependent work HRBPs were already doing — navigating escalations, coaching through emotionally charged conversations — is where human judgment stays essential. The data told Oliver both things simultaneously, and the journey map made it visible.