top of page
image (19) copy.webp
AIforCFO - Home
image (19) copy.webp

AIFORCFO FRAMEWORK · REVISED EDITION 1.1

The Four Gates of AI ROI™: How CFOs Decide Which AI Initiatives to Fund

AI activity is easy to count. A return needs a measured workflow outcome, a credible route to benefit and the full cost of delivering it. The Four Gates turn that into four connected questions.

By Alankar Joshi, AIforCFO · Updated 13 October 2026

16-page white paper, free · Worked case and worksheet · Sources linked

EXECUTIVE SUMMARY

Three Decisions to Make Explicit Before You Fund AI.

Licences issued, prompts written and faster drafts show that people can use a tool. An organisational return needs more: a workflow that changes, a benefit that can be evidenced and a cost that includes the whole operating change.

01

Define the benefit before approving the spend.

Hours released, cash saved, additional contribution and better controls require different evidence.

02

Match evidence to the next decision.

A discovery exercise, a controlled pilot and production scale require different levels of assurance.

03

Resolve every material blocking condition.

A strong benefit forecast cannot compensate for an unapproved data route or an unreliable output.

The Four Gates of AI ROI™ is a CFO decision framework by Alankar Joshi at AIforCFO for deciding whether a finance AI initiative deserves investment. It asks four connected questions: Direction (is this a worthwhile problem, and is AI suitable?), Economics (is the expected benefit worth the full cost?), Architecture (can the workflow operate reliably within approved boundaries?) and People (can people deliver and sustain the operating change?). Direction comes first; the other three are assessed together as the design develops.

What are the Four Gates of AI ROI™?

Released capacity is not a cash saving

A salary-based value for hours released is not a cash saving. Capacity can still support more output, avoid future hiring or replace external work. Claim those effects only when the mechanism is evidenced, and do not count the same hours twice. Keep the P&L (profit and loss), cash flow and capacity measures distinct. UK HM Treasury guidance draws the same line between cash-releasing and non-cash-releasing benefits. HM Treasury, Government Efficiency Framework

EXECUTIVE SUMMARY

See Where Your Next AI Initiative Stands at Each Gate.

Apply the Four Gates to your team →

METHOD AND LIMITS

How the Framework Was Developed, and What It Does Not Claim.

AIforCFO’s earlier reference analysis, The AI ROI Gap, organised the subject into ten themes. The Four Gates compress those themes into Direction, Economics, Architecture and People, informed by practitioner judgement and editorial case analysis.

This edition adds public finance research, benefits-measurement guidance and AI risk-management material to support specific observations. It does not report an original survey, independently verified client results or a test of the framework’s predictive performance.

The framework complements existing capital allocation, benefits-management and risk processes. Its main application is generative AI in finance workflows; predictive models and systems that take actions need additional evaluation and controls.

QUESTIONS

Four Gates of AI ROI: Common Questions.

  • The Four Gates of AI ROI™ is a CFO decision framework by Alankar Joshi at AIforCFO for deciding whether a finance AI initiative deserves investment. It asks four connected questions: Direction (is this a worthwhile problem, and is AI suitable?), Economics (is the expected benefit worth the full cost?), Architecture (can the workflow operate reliably within approved boundaries?) and People (can people deliver and sustain the operating change?). Direction comes first; the other three are assessed together as the design develops.

  • No. Start with Direction by defining the workflow and the decision. Assess Economics, Architecture and People together as the design develops, and reopen any gate when its assumptions change. The diagram identifies four lenses; it does not prescribe four implementation phases.

  • PASS: the evidence meets the criteria for this stage. REVIEW: an uncertainty remains, with a recorded test, owner, deadline and permitted activity. FAIL: a condition is not met in the current design, so redesign or stop that activity. MISSING: evidence has not been gathered, and the gap may still block the activity.

  • No. A salary-based value for hours released is not a cash saving. Capacity can support more output, avoid future hiring or replace external work, but claim those effects only when the mechanism is evidenced, and do not count the same hours twice.

  • No. Passing supports a decision under stated conditions. Four passes also do not make an initiative the best use of capital; compare eligible options on the same horizon and benefit definitions.

  • It applies established investment discipline to AI-specific operating questions: variable output quality, review and correction effort, usage-sensitive costs, data access and changing model behaviour. Use it alongside your existing investment and risk approvals.

  • Start with one initiative and the worksheet in the white paper. For a personal first view, take the free readiness check. For a company-specific assessment reviewed by Alankar Joshi, see the AIforCFO Diagnostic.

image (19) copy.webp

Apply the Four Gates to your own finance team.

The AIforCFO Diagnostic is a fixed-fee, company-specific assessment reviewed by Alankar Joshi. It shows where your next AI initiative stands at each gate, and what evidence the next decision needs.

alankar-joshi.webp

​ABOUT THE AUTHOR

Alankar Joshi leads AIforCFO, a practitioner-led AI education and toolkit business for corporate finance leaders, and teaches in the Singapore Management University CFO Programme. More about Alankar

bottom of page