What breaks when the AI project list goes from 5 to 30
Evidence: Cambridge CCAF (2026), Gartner (2025), MIT NANDA (2025), and nCino and Wolf & Company surveys (2026). The failure modes and the response are my view, from work with lenders and credit firms. I am a partner of Strategy of Things, whose reference model this note draws on.
A mid-market lender used to run perhaps 5 technology projects a year. The AI candidate list now runs 20 to 40. The machinery for choosing, funding, and measuring them was built for the smaller number, and it breaks around return metrics and interdependencies.
Four things you'd see
- Every project has a business case. The portfolio shows no return. Each case is written against a baseline the other 29 projects are moving. Three or four outcomes for the whole program, measured together.
- The steering pack counts seats licensed and pilots launched. Activity reporting filled the space where return should be. Execution per project, return per portfolio.
- Funded projects stall 6 months after approval. They depend on work that lost the ranking. Dependencies mapped before ranking.
- The data layer and the document pipeline never get funded. Enablers have no P&L line of their own. Funded as infrastructure, they stop competing with earners.
Firms consistently confirm this. 76% of large financial institutions find it hard to measure the value of AI deployment (CCAF, 2026). 22% of organizations have defined, tracked, and communicated business impact metrics for most of their use cases (Gartner, 2025).
The enabler trap
Rank 30 projects by business case and the enablers tend to finish last. For example, the document pipeline, the data layer, the model registry, the entity resolution: none of them show up as revenue or cost saved. The projects that do show up quietly assume this work already exists.
So you fund only the winners. Six months later, several are likely blocked by work that never got funded.
Running the list
Everything above happens above the projects. Strategy of Things places that work in the top zone of its 9-Layer Enterprise AI Reference Model, the one covering governance, adoption, and strategy (Strategy of Things, 2026). Lower layers enable the ones above them, which gives the enabler and earner split a published structure.
So running the list is a capability, and it carries the same build, buy, or partner question as the projects do. A mid-market lender rarely builds that one.
The effect differs by firm: how many projects run against the same operations, and whether anyone owns the list.
Open question: which enabler has been proposed twice and lost twice?
Email me to discuss.
Based on a longer analysis.
From the work on AI project sourcing for lending and credit.