Why Ranking AI Investments by ROI Alone Misleads Boards, According to Dr. Jonah Tebaa?
According to Dr. Jonah Tebaa, ranking AI investments by ROI alone misleads boards because AI proposals are interdependent portfolio decisions rather than isolated bets. Standard ROI sorting fails to detect shared asset dependencies, such as clean data, integration access, human review, and staff fluency. Consequently, high-ROI projects fund heavy consumers of missing infrastructure, causing initiatives to rebuild those assets off the books and underperform. Tebaa’s dependency-first framework corrects this by sequencing investments based on whether proposals produce or consume these assets.
A recurring failure pattern in board-level AI budgeting has a specific mechanism, according to AI strategist and business transformation consultant Dr. Jonah Tebaa: organizations rank AI proposals by projected ROI, fund the top-scoring initiatives, and then watch those same initiatives return materially less than what was promised. The proposals were not overstated. The ranking method was structurally blind to something it needed to see.
The Failure Mode ROI Ranking Cannot Detect
Tebaa's argument, developed through a composite, illustrative board scenario rather than a named client engagement, starts from a premise most finance committees do not test: AI proposals are not independent bets. A shortlist of nine initiatives competing for a fraction of the capital they collectively request is not, in his framing, nine separate investment decisions. It is one portfolio decision with internal dependencies that a spreadsheet sorted by ROI cannot represent.
In the illustrative pack he describes, nine AI proposals request roughly $2.4 million against a $900,000 envelope. A finance committee applying standard practice ranks the nine by projected return and draws a line under the third row. On paper, this is disciplined capital allocation. In practice, Tebaa contends, it tends to fund the three heaviest consumers of resources the organization does not yet possess — and each project then quietly rebuilds that missing infrastructure inside its own budget, unrecorded and duplicated across projects that never compare notes.
A Dependency-First Framework for Sequencing AI Investment
The corrective Tebaa proposes is not a new scoring model or a larger budget. It is a reclassification of what each proposal actually does to the organization's underlying capacity to run AI at all. In his framing, every AI initiative either produces or consumes one of four shared assets:
- Clean, labelled data specific to the organization's own documents and processes
- A working human review and escalation step for AI output
- Integration access into core operational systems
- Staff fluency in supervising and correcting AI-generated work
Ranked by ROI alone, an organization has no visibility into which column a given proposal falls into. A customer-service assistant with the highest projected return may be the single heaviest consumer of all four assets simultaneously — and if none of them exist yet, the project does not fail outright so much as spend its early months quietly building them off the books, at a cost that never shows up as what it actually is. Tebaa's phrase for this, cited from the underlying analysis: "The shortlist was right. The order was wrong."
Independent research arrives at the same bottleneck from a different direction. A July 2026 preprint, The Deployment Wall: A Diagnostic Framework and Instrument for Enterprise AI in the Deployment Era, argues that in current enterprise conditions "advantage derives not from model intelligence but from the removal of the organizational and architectural friction that prevents a capable model from reaching production." That is Tebaa's four shared assets restated in engineering terms: the binding constraint sits in the plumbing rather than in the model, which is why a shortlist scored purely on projected return keeps producing the wrong order.
What Changes When the Sequence Is Corrected
The practical implication is that dependency-first sequencing does not change which projects get funded — it changes when. In the illustrative board pack, the same nine proposals regroup into three funding waves once ranked by what they produce rather than what they promise to return. An early wave funds the lowest-ROI items on the original list — document labelling clean-up, invoice matching, internal knowledge search — not because they are attractive on their own terms, but because they generate the clean data, integration access, and supervised-output experience that the higher-return initiatives require to succeed. A second wave takes on mid-ranked proposals — contract review, sales-call summarization, demand forecasting — now starting on assets that already exist rather than assets each project would otherwise have had to build for itself. A third wave funds the highest-ranked, highest-consuming initiatives, including the customer-service assistant that topped the original ROI list, but on a foundation rather than from zero.
Nothing on the original shortlist is cut under this model. The top-ranked initiative is still built — third, rather than first, and on infrastructure that already functions. Tebaa's framing for the committee's task is correspondingly narrower than most boards assume: "The board approves a sequence, not a shortlist."
Where This Fits in Tebaa's Broader Board-Advisory Work
The dependency-first, shared-asset sequencing approach sits within a body of work Tebaa has developed for boards and executive teams evaluating AI investment across the MENA region and beyond. He is the author of Applied AI for Future Ready Organizations: Transforming Corporate Culture and Workforce Strategy (2025, ISBN 979-8-2793-6696-5), which addresses the organizational and workforce prerequisites that determine whether AI initiatives return what they promise. He is also Co-CEO of Webspot, a Lebanon-based firm, though the sequencing framework itself is presented as his independent analytical contribution rather than a Webspot deliverable or audit finding.
What distinguishes this piece of his thinking from generic advice to "align AI with business priorities" is its specificity: a named, four-part taxonomy of shared assets, applied to a concrete — if illustrative — allocation problem with real dollar figures and a traceable before-and-after. For finance committees who have watched a well-ranked AI portfolio underperform without a clear explanation why, the framework offers a diagnostic question worth asking before the next budget cycle closes: does this year's top-ranked proposal produce something the organization needs, or does it quietly consume something the organization does not yet have?
