Funding the AI initiatives that really matter

In many companies with 80 to 800 employees, the AI budget is set in one of two unsatisfactory ways. Either a general "innovation" envelope is opened and people draw on it for experiments whose outcome nobody tracks. Or every request is declined for lack of a business case, because the promised gains look too vague to a CFO. In both situations, money does not reach the places where it matters.

We think the issue is less the amount than the unit of funding. One does not fund "AI". One funds the improvement of a process or a decision, with an owner, a baseline and a time horizon. AI is then one possible lever among others, next to simplification, clearer roles or better configuration of the ERP.

Three kinds of value, three ways to judge them

An initiative creates value in three ways, and it helps not to add them up in a single table.

Capacity. Hours are freed, for example in the monthly close or in report production. This is the most quoted value and the most fragile: an hour freed is not a franc saved. It becomes a financial gain only if it avoids a hire, removes external spend or is redeployed to a task with an identified value. We detail this in our piece on the augmented CFO and a faster close.

Cash. The gain is a flow or a stock of liquidity: shorter collection times, better-calibrated inventory, less reliance on credit lines. It is usually a one-off release of cash plus a recurring effect on financing costs. This is the logic behind continuous cash flow forecasting.

Risk and quality. Fewer invoicing errors, fewer compliance lapses, fewer decisions taken on stale figures. The value is real but probabilistic; it should be judged on a prudently estimated cost of avoided incidents, not on a promise.

A worked example, with explicit assumptions

Take a fictitious Swiss industrial company: 150 employees, CHF 60 million in revenue, a fully loaded cost of CHF 800 per day for a finance profile, and a financing cost of 4%. Two initiatives compete for funding.

Initiative A: automate the monthly reporting pack. Assumption: two people save three days a month, so 72 days a year, or about CHF 58,000 of capacity. First-year cost: CHF 40,000 for implementation, CHF 16,000 of internal time (20 days) and CHF 14,000 for licences and running costs, about CHF 70,000 in total. If that time is not redeployed, the financial return is nil in year one, even if the team feels relieved. If redeployment is decided in advance (commercial analysis, margin tracking), the initiative can be justified, but over a two-year horizon.

Initiative B: prioritize customer collection actions. Assumption: average collection time moves from 55 to 52 days. One day of revenue is about CHF 164,000 (60 million divided by 365), so three days release about CHF 490,000 of cash, once, and save roughly CHF 20,000 a year in financing costs. For a comparable cost, in the order of CHF 50,000, the effect on liquidity is of a different scale.

None of these figures is a market benchmark: they are assumptions, to be replaced with your own. The value of the exercise lies elsewhere. It forces you to name the kind of gain, who receives it (who gets the cash, who gets the time) and the condition for it to materialize. A company under liquidity pressure will rank B ahead of A; a company struggling to hire finance talent may reason differently.

Fund in tranches, not by project

Once priorities are set, the funding mechanism matters as much as the amount. We recommend four short tranches, released one after the other.

Tranche 0: frame and measure

Two to three weeks to describe the process as it actually runs, measure the starting point (duration, error rate, delay) and name the person accountable for the result. Without a baseline, no gain can ever be demonstrated. The cost is small, and a fair share of ideas often stop here, which is a good outcome.

Tranche 1: pilot on real data

Six to ten weeks, narrow scope, real users. The aim is not a technical demonstration but an answer to one question: does the gap measured in tranche 0 narrow? Stop criteria are written before the start.

Tranches 2 and 3: deployment, then steering

Deployment is funded only if the pilot reaches the agreed threshold. Part of the budget stays reserved for tracking results after go-live, because this is where gains erode when nobody answers for them. An executive dashboard showing the starting indicator next to the current one is often enough to sustain that discipline, as in this automated executive dashboard use case.

The costs that are forgotten

Business cases almost always underestimate the same set of costs, which are better provisioned from day one:

• data quality and preparation, often the heaviest part of the work;
• integration with existing systems (ERP, banking, CRM);
• change management: training, adjusted roles, managers' time;
• running costs: licences, supervision, maintenance, rework after version changes;
• compliance. In Switzerland, the revised FADP (nLPD) has been in force since 1 September 2023 and applies as soon as a tool processes personal data. For companies operating in the European Union, the EU AI Act, in force since 1 August 2024, applies in phases: prohibitions in February 2025, obligations for general-purpose AI models in August 2025, most obligations in August 2026. These topics, covered in our Data, AI & compliance expertise, weigh little when built into the framing and a great deal when discovered at the end of a project.

As an order of magnitude, it is prudent to assume that non-tool costs amount to at least as much as the tool itself. This is a starting point to adjust, not a rule.

What this changes for the executive team

For a CEO or CFO, three concrete decisions follow.

First, limit the portfolio. Two or three initiatives carried through to a result are worth more than ten pilots with no follow-up. The attention of the executive team and of the organization is the truly scarce resource.

Second, require a single sentence for every request: "We want to move this indicator from this value to that value, under this person's responsibility, by this date." If the sentence cannot be written, the initiative is not ready.

Third, accept stopping. A tranche that is not released is not a failure but the application of a rule set in advance. That is what makes the retained initiatives credible to teams and to the board alike, in Zurich as in Chicago.

Our approach, from strategy to deployment and then to steering results, follows this logic across our Finance & performance expertise. If you want to know which initiatives deserve funding in your organization, and which are really a matter of better organization, our diagnostic is a good place to start.