The board has decided on AI. The use cases are identified, expectations are high, the contracts with the external provider are ready to sign. Then comes the budget discussion. Each division points to its own constraints. The CFO demands an ROI proof for a technology that has not yet been implemented and for which no reliable empirical data exists within the company. The initiative lands on the agenda for the next budget round.
AI investments fail in many organizations not due to technology and not due to lack of will. They fail due to funding logic. Those who treat AI as additional budget compete with everything. Those who treat it as reallocation finance it from what it makes obsolete.
A managing director I advised resolved this blockage with a question: “Which activities do we maintain today that we want to eliminate in two years?” The answer was the actual business case. Three manual processing workflows, for which the company maintained a total of five full-time positions, could be fundamentally transformed by two AI applications. The investment was no longer a question of additional resources, but a simple calculation: Where does the counter-financing come from, and when does it pay off? This is a different discussion than “Who approves the budget?” It is a leadership discussion.
Clayton Christensen described the principle of efficiency innovation: technologies that make existing processes cheaper or faster generate their actual value not through what they add, but through what they replace. AI in most corporate applications today is precisely that: an efficiency innovation. Its financing potential lies in what it makes obsolete.
The first step is making them visible. Many organizations know roughly that an AI application saves time, but they do not measure it systematically. What is not measured cannot be strategically redirected. Instead, it dissolves into the general perception of capacity: employees have slightly more buffer, workflows become somewhat more comfortable, but the savings are not realized.
Effective counter-financing begins by quantifying the efficiency gain before approval, measurably and conservatively. Not only in the economic assessment, but as an operational commitment: the application should reduce manual effort in this process by X percent. Executives must measure and report this saving in the next KPI review. The second step is redirection: the freed capacity must not flow uncontrolled into general operations, but must be strategically managed. In some cases, this means shifting to more value-creating activities, in others a capacity reduction. What is crucial is that the use of capacity is anchored in the business case before the system goes live.
| Apparent Savings | Actual Savings |
|---|---|
| Time is saved, capacity remains unchanged | Freed capacity is strategically redirected |
| No control, no measurement | Savings are quantified and measurable before approval |
| Dissipates into general operations | Flows into financing the next application |
| Business case remains abstract | Counter-financing is concrete and verifiable |
A division head I supported during the implementation of an AI-powered document management system experienced a common pattern: the new system was running productively, but the legacy system remained operational. The reason was structural: the legacy system had technical dependencies on two other systems whose migration required time and resources. So both systems ran in parallel. Twelve months after rollout, the costs of parallel operation exceeded the original operating costs of the legacy system. The expected savings effect existed on paper, but was not visible in ongoing cost accounting.
This is not an isolated case. Legacy systems remain operational after an AI implementation for several reasons, all of which are legitimate: technical dependencies that prevent immediate decommissioning. Depreciation on legacy systems that make continued operation appear more cost-effective in the first year. Regulatory retention requirements that force transition periods. These are not excuses, but structural obstacles that belong in the business case and must be evaluated accordingly.
The consequence: the decommissioning strategy must be planned before approval, with a concrete date, clear responsibility, and realistically calculated migration costs. Those who postpone shutdown to later bear the risk of permanent operation. And making the decision under planning conditions is fundamentally more cost-effective than under the pressure of an already running system.
The third lever is prioritization. Most organizations distribute AI resources according to a logic that has more to do with political balance than with effectiveness: each division that has submitted a convincing use case receives a portion of the available resources. The result is a multitude of small pilots, all too small to prove their value, and collectively too costly to truly scale any of them.
A strict filter criterion helps: Which applications achieve positive savings within twelve months? Every organization should establish a binding ranking criterion for its AI initiatives, whether payback period, return on investment, or another metric that the CFO and business units jointly consider appropriate. Not as a guarantee, but as an expectation when allocating resources. Initiatives that fail to meet the chosen criterion are treated as lower priority, not because they are strategically wrong, but because short-term savings should co-finance the more investment-intensive initiatives. Concentrating on two or three consistently funded applications almost always beats a dozen pilots, because only consistent scaling generates actual value.
This is also where the real danger of the valley of death lies: the phase between a successful pilot and productive operation. Those who have distributed their resources across many pilots ultimately have no capacity left for the last mile that turns a functioning test into a value-generating system. The counter-financing logic reverses this: first, consistently take one application to production maturity and verifiable savings, then use it to finance the next. This is prioritization as a financing strategy: not distribution to all, but concentration on what is effective.
First: For how many AI applications in your division have you defined a concrete, measurable saving that was agreed upon before approval and verified after six months?
Second: Which legacy systems continue to run in parallel to an already implemented AI application without a decommissioning strategy in place?
Third: If you had to concentrate your available resources on the two applications that would achieve verifiable savings within twelve months, which would they be?
AI finances itself where it is consistently deployed: from the reduction of what it replaces, and from the savings it generates. This requires executives to actively measure these savings, redirect them strategically, and not let them dissipate uncontrolled into general operations. And it requires that the decommissioning strategy for legacy systems is part of the business case.
Financing AI as a special project means permanently fighting for resources. Successively refinancing AI investments from successful AI implementations means treating them as what they fundamentally are: an efficiency innovation.
Reducing Costs Without Weakening the Organization – The logic of intelligent reallocation applies to AI financing just as it does to any other cost decision.
From Pilot to Production – Why the expensive last mile determines the success of an AI investment.
All Insights can be found in the overview.
Proven tools and models for self-application are available under Solutions.
If you want to take these thoughts further for your company, a no-obligation initial conversation is worthwhile.
