The AI Business Case CFOs Actually Approve

01 –

The pilot graveyard is no longer a secret

Enterprise AI has settled into a predictable, expensive pattern: the high-profile announcement, the polished vendor demonstration, the rapid prototype and the small group of enthusiastic beta testers. Then reality intervenes. As teams attempt to move these pilots into everyday operations, they meet legacy systems, fragmented data, security constraints and workflows that were never redesigned for AI-assisted work.

The evidence reflects that friction. McKinsey reports that nearly two-thirds of organisations have not yet begun scaling AI across the enterprise. In addition, only 39% report any enterprise-level EBIT impact, and most of those respondents attribute less than 5% of EBIT to AI use.
spglobal

S&P Global Market Intelligence paints an even starker picture: the proportion of organisations abandoning most AI initiatives before production rose from 17% to 42%, while the average organisation scrapped 46% of proof-of-concept projects before they reached production.
github

What remains is a quiet accumulation of pilot debt: overlapping SaaS licences, abandoned prototypes, opaque data permissions, bespoke integrations that require manual maintenance and informal employee workarounds. Over time, the more corrosive consequence is cultural. Employees learn not to take the next executive AI announcement seriously.

---

02 –

The cost of inaction is not ‘falling behind’

Fear of “falling behind” competitors is a poor foundation for an investment case. Finance teams cannot price it, allocate an owner to it or track it against a baseline.

The real cost of inaction is more concrete: continuing to fund slow, manual and error-prone ways of working. It appears across payroll, external spend, service backlogs, customer churn, rework and delayed decisions. Because those costs are distributed across the organisation, they are easy to tolerate — until someone adds them up.

2.1 Persistent operating drag

Every day a business leaves a high-friction process untouched, it pays for that friction. The cost may appear as additional handling time, duplicated effort, unnecessary specialist input, avoidable overtime or a growing backlog.

For example, it may look like this:

Before proposing AI, establish the current baseline: transaction volume, handling time, error or rework rate, queue length, escalation volume and cost per case. Without that baseline, a claimed productivity gain is simply an impression. CFO-focused measurement frameworks similarly emphasise cycle time, cost-to-serve, quality, control and capacity — rather than tool usage or log-ins alone.

A credible proposal should then translate the expected benefit into categories finance can interrogate:

The distinction matters. Hours saved are not automatically money saved. The value becomes real only when capacity is used differently, external spend falls, throughput rises, quality improves or a commercial outcome moves.

2.2 Revenue lost to friction

Slow workflows are not merely inefficient; they can be commercially damaging.

A delayed response can weaken a sales opportunity. A slow onboarding process can postpone revenue. Poor access to reliable knowledge can turn a straightforward customer request into a repeat contact, escalation or cancellation. In each case, the business is already paying for friction — just not always in a budget line labelled “manual process”.

Therefore, the test for an AI investment is not whether it can generate an answer faster. It is whether a faster, more reliable workflow can move a business metric the board already values: conversion, retention, cost-to-serve, response time, revenue per employee, service-level attainment or time to revenue.

That is the difference between a technology demonstration and a business case.

---

03 –

Shadow AI is a cost centre in disguise

Blanket bans on AI tools do not eliminate AI use. More often, they displace it.

When employees lack a practical, approved way to use AI, they turn to consumer accounts, personal devices and improvised workarounds. The immediate concern is obvious: sensitive information may be copied into tools the organisation cannot assess, monitor or control. However, the deeper problem is inconsistency. Different teams adopt different models, prompts, quality checks and assumptions about what is permissible.

That creates shadow AI: activity that may be well intentioned and productive in the moment, but is invisible to the people responsible for data, security, legal risk and service quality.

The answer is not an exhaustive policy document that staff will never read. Instead, organisations need usable guardrails at the point of work: clear approved tools, simple data boundaries, role-based access, defined human accountability and an escalation route when something goes wrong.

For AI systems that process personal data, the governance model should also sit within existing data-protection and enterprise-risk practices. This includes documenting the purpose and data use of the system, applying appropriate security and oversight measures, and conducting a Data Protection Impact Assessment where the processing is likely to create a high risk to people’s rights and freedoms.

---

04 –

The hidden price of a stalled pilot

The visible cost of an AI pilot — the initial licence, subscription or API bill — is usually the smallest part of the investment. The larger cost sits around the technology: data preparation, integration, security review, workflow redesign, training, support and the time required to make a new process dependable.

When a pilot stalls, those costs do not disappear. They become sunk effort, fragmented systems and another unfinished change programme for employees to absorb.

S&P Global Market Intelligence found that organisations scrapped an average of 46% of AI proof-of-concept projects before they reached production. That makes a bounded pilot more than a technical trial: it is a capital-allocation decision that needs a pre-agreed end point.

Set the rule before the pilot begins: every initiative should have defined success measures, a named business owner, a spending ceiling and only three possible outcomes:

There is no fourth option: an indefinite extension labelled “learning”.

05-

The business case a CFO can approve

If the benefit cannot be measured, owned and realised, it is not a financial case yet.

A finance-ready AI proposal does not pitch abstract possibilities. It contrasts a loose, optimistic business case against a rigorous, outcome-driven operational plan:

This comparison forces decision-makers to evaluate three paths objectively: doing nothing, running a highly bounded, outcome-driven pilot, or committing to immediate scale based on proven evidence.

---

06-

Start with the workflow, not the model

A tool cannot fix a process nobody has documented.

If a business process is undocumented, poorly understood, and riddled with informal human workarounds, introducing AI will not make it efficient. It will simply automate and accelerate the chaos. The highest-leverage initial investment an organisation can make is often not software, but workflow preparation.

Five features of highly promising AI use cases:

---

07-

Adoption belongs in the financial model

If the projected financial value of an AI project depends on employees actually changing their daily habits, then user adoption is not a soft HR metric-it is a critical financial assumption. A professional business case must model the ramp-up period, required training overhead, and expected human fallback rates.

When presenting to finance, model exactly what percentage of the total workload is expected to shift to the automated process, how quickly that transition will occur, and what metrics will define “meaningful use” over simple tool login activity.

---

08-

A 90-day route out of pilot purgatory

To break the cycle of endless, unmeasured pilots, organisations should adopt a strict, time-bound 90-day execution framework:

---

09-

The CFO’s real choice

The question is not whether to invest in AI. It is which constraints justify intervention.

The fundamental question for any forward-looking leadership team is not “How can we implement AI?” but rather: “Which of our current operational constraints are expensive enough to justify a controlled, highly disciplined AI intervention?”

A successful, CFO-approved business case replaces the vague promise of automation with a clear financial trade-off. It states: we understand our current workflow costs, we know exactly what we will spend, we have mapped precisely how our processes will change, and we will stop immediately if the expected operational evidence does not appear within our 90-day window.

---