Two of your AI initiatives might look the same on a slide. They are not the same thing, and they should not be governed the same way.
The single most useful distinction a board can make about its AI portfolio in 2026 is between embedded and bolted-on — because almost every other question downstream depends on it.
Two Different Products, Two Different Trajectories
Bolted-on AI is the AI feature that sits on top of an existing workflow: the meeting summariser, the document Q&A, the inbox triage assistant. It is mostly licensed, mostly optional, mostly unobtrusive. ROI is incremental and reversible.
Embedded AI is the AI that rewires the workflow itself: the underwriting engine that changes how credit officers spend their time; the clinical decision support that changes how a triage nurse routes patients; the agentic research workflow that compresses three roles into one. ROI is non-linear and the workflow does not snap back if you turn the AI off.
Why Conflating Them Is Expensive
Boards that treat embedded and bolted-on as the same thing make two predictable mistakes. They under-govern the embedded systems — applying procurement and assurance frameworks designed for SaaS to systems that are restructuring core operating processes. And they over-govern the bolted-on ones, asking for ROI cases on tools that cost less to deploy than the meeting it takes to evaluate them.
Both mistakes consume executive attention and slow real progress.
A Workable Decision Framework
Three questions sort almost every AI initiative cleanly. First: if we turned this off in three months, would the workflow snap back, or would the organisation have already restructured around it? Second: is the AI making decisions that previously required a regulated, accountable role? Third: does the AI change which functions you would hire for in the next two years?
Two or three yeses means embedded. Two or three nos means bolted-on. Govern accordingly: lighter, faster, more delegated for bolted-on; deeper, more cross-functional, board-visible for embedded.
What to do next
- Sort your AI portfolio into embedded vs bolted-on using the three-question test
- Build two governance tracks, not one — the same framework will fail both
- Reassign board attention to the embedded initiatives; delegate the rest
- Revisit the sort annually — initiatives migrate categories
If this resonates and you are leading an AI portfolio that is hard to prioritise or hard to govern coherently, Grant & Graham can help. We provide AI governance design, operating-model implications, and senior advisory for boards and executive teams overseeing enterprise AI portfolios across EMEA. Start a conversation.
