02 · AI portfolio strategy

Leaders were excited about GenAI. I steered that excitement toward the work it needed first.

The exciting part of seeing the whole portfolio was finding the need nobody even thought about. Sorting 279 requests by the job a model would perform revealed 176 needs AI alone could not solve—and gave leadership a clearer place to invest first.

176 of 279requests needed work AI alone couldn't do

My roleSole author
WhenFeb 2026
Scopethe firm-wide intake board, the architecture review board, the product-risk function
Outputwhat the firm wanted from AI, what still needed tech and data work, and how leaders should shape roadmaps and investment

Once intake sped up, I could see the whole portfolio instead of one request at a time. Everyone was writing GenAI roadmaps and every team wanted to know where to focus. The honest answer meant sorting the demand by what the model would be asked to do, not by what colleagues were requesting.

I'm the front door for the firm-wide intake board. It holds 344 opportunities today; 144 of them came in after I took over intake in August 2025, and I opened 68 of those. I also use the intake skill to write stories straight into other teams' projects. I did this analysis alone: in February I classified all 279 requests the board had collected by then, wrote up the themes, and took the finding to leadership. In one practice, I converged sixteen opportunities from four teams into two priorities.

Treat the outliers as the finding. 176 of the 279 requests didn't fit the four original task types: summarize, classify, verify, generate. Instead of stretching the taxonomy, I dug into what those requests really were and what they'd take. Five themes came out of it: data migration, platform access, foundational pipelines, governance, and connecting systems. The less visible data and platform work became the first investment decision.

I didn't ask leaders to cool off on GenAI. I showed them what their GenAI plans needed underneath, so the excitement pulled the data and platform work forward instead of competing with it.

AI got the attention. The portfolio showed us which data and platform foundations had to come first.What I told leadership
  1. every request279 demands on the intake board
  2. sort by tasksummarize, classify, verify, generate
  3. dig into what's left176 didn't fit; five themes underneath
  4. what gets fundedleadership's call, checked with the architecture review board

A lens, not a model: the same sort runs on any queue.

279 requests sorted by what the AI would have to do. 103 fit one of four task types: summarize, classify, verify, generate. 176 didn't, and they fell into five underlying needs: data migration, platform access, foundational pipelines, governance, and connecting systems. 279 requests everything on the board 103 fit one of the four AI task types summarizeclassifyverifygenerate 176 didn't fit: five needs underneath datamigrationplatformaccessfoundationalpipelinesgovernanceconnectingsystems the less visible data and platform work became the first investment decision
Sorted by what the AI would actually have to do, most requests weren't AI tasks at all. The five needs underneath set the order leaders funded things in.
Built withGenAI task taxonomyDemand analysisDFVArchitecture review board
176 of 279
requests didn't fit any of the four AI task types
16 → 2
opportunities from four teams, converged to two priorities and validated with the architecture review board
5
missing themes that refined the strategy instead of being forced into it

Senior stakeholders and decision-makers agreed with the findings. I looked at our whole tech system, including the architecture review board's and the product-risk function's queues, which confirmed this would have firm-wide impact.

Leadership does. They got a reusable lens for prioritizing builds and spotting reuse across practices, and they still use it.

  • Taxonomies are instruments, not truths. Date the snapshot, publish the definitions, and score the outlier by impact before you dismiss it.
  • Say the finding plainly. "Most of the demand isn't AI" is the sentence that moved funding. A chart didn't.