“Let’s start by brainstorming what AI could do for us.” The moment that conversation begins, the project is already three months behind schedule. Before widening the candidate list, the work of cutting candidates needs to come first.
CAIO’s intake receives several dozen “we are evaluating AI adoption” inquiries each month. In most of them, the opening shape is the same: five to ten AI-use ideas that have come up from operating teams, with the question, “which of these should we start with?”
You should not answer that question with a priority matrix. Most of the candidates lined up are themes that should not even be on the priority discussion at all. The issue is not that ROI is low. Putting them on the discussion at all carries a structure that delays the organization’s decision-making.
How to think about “cut criteria”
The standard axes for evaluating engagement themes are return on investment (ROI) and execution feasibility. Those are valid, but for the early-stage judgment at mid-cap and small-cap firms they are insufficient. We add one more.
THE THIRD AXIS — Decision Cost
How many stakeholder agreements does discussing this theme require — among executives, operating leads, IT, legal, and information security? The higher this is, the more the theme stalls the entire project regardless of its own ROI.
When evaluated on these three axes, themes where ROI and feasibility are moderate but decision cost is extreme are candidates to cut first, however attractive they look. Three patterns I see repeatedly are below.
The three candidates to cut first
01 — Company-wide AI chatbot deployment
“Give every employee access to AI.” On the surface, easy to start and large in apparent impact. In reality, agreement is needed from information security, IT, HR, legal, and every business unit, and the discussion expands into handling of personal information, confidential information, and competitor data. The highest consensus-building cost of any theme.
Decision Cost: extreme · ROI: medium
02 — AI enhancement of executive dashboards
A proposal to add AI features to existing BI tools. This theme reaches into “the existing executive decision process itself.” Discussion extends to how the executive committee runs, how KPIs are defined, how responsibilities are split across departments. Mid-discussion, the AI conversation has expanded into a management-reform conversation.
Decision Cost: extreme · ROI: unknown
03 — Automatic summarization of sales-meeting notes
The theme looks small, but the decision to upload customer conversations as audio data to the cloud requires the information-security committee. Then customer-consent acquisition flow, recording retention period, linkage to performance evaluation — the agenda keeps growing.
Decision Cost: high · ROI: small
The shared “trap” structure
What these three candidates have in common is that they are themes that cut horizontally across existing organizational boundaries. Horizontal themes are individually correct and produce ROI in part-optimization terms. But as the first move, they involve too many stakeholders to advance.
The rule for producing early AI-adoption wins at mid-cap and small-cap firms is to start from themes that complete inside three or fewer stakeholders. Once stakeholder count reaches four or more, the decision cycle stretches from weekly to monthly and the entire project enters a “looks stalled” state.
The criterion for the first theme is not ROI. It is “what can be started without asking permission.”
— Frank Wang, Founder, CAIO
What to look at instead
Setting horizontal themes aside, the candidate list shrinks substantially. What remains are themes that complete inside a single department, on a single process, with one or two decision-makers. They look unexciting; they are the right starting point.
For example:
- Invoice classification automation in accounting (the accounting head can decide alone)
- Proposal-template generation for a specific sales team (within the sales head’s authority to start)
- FAQ auto-generation for customer support (completes within the CS lead’s scope)
These are not characterized by being “small.” They are characterized by short decision paths. Companies that succeed with one of this type see decision speed on the next theme rise two- or three-fold by perception. The organization is building “judgment muscle” for AI adoption.
Closing — Do the dropping work first
In early-stage AI adoption judgment, the most valuable work is not generating candidates but cutting candidates. Apply “decision cost” before ROI or feasibility, and the candidate list shrinks to a third or less.
Start from what remains, build a working track record in the first ninety days, and raise the organization’s judgment muscle. Then come back to the horizontal themes you “cut” today. That is the realistic order at mid-cap and small-cap firms.
This article composes anonymized and abstracted material from multiple consultation cases. It is not the case of any specific company.
About the author
Frank Wang — Founder, CAIO
Operator-led AI adoption advisory for Japanese SMB and mid-cap companies. Adapts 15 years of enterprise DX implementation across Japan, US, Europe, and Asia for the SMB context. AI-native delivery — judgment design through implementation. Trilingual in Japanese, English, and Mandarin.