AI applications are taking hold in companies: they take over recurring tasks in processes, support customer service, create documents, or facilitate document exchange in the supply chain. They are modern, and an individual prompt is inexpensive, but more and more companies are complaining about rapidly rising total costs. We view AI applications as a positive way to optimize processes, but we also see the need for every application to deliver a direct return on investment and for overall costs to be managed through company-wide AI governance.

Even though the cost per request is falling—especially for older AI models—total expenditure on AI applications is rising at many companies because more expensive, higher-performance models are being used and the number of requests per unit of time has increased. This is also primarily related to how AI applications are used. Until now, usage has mostly involved standard tasks such as classification, summarization, basic customer service, or translation. However, the number of complex applications is increasing, involving programming, long documents, and agents. Overall consumption within companies is growing, as more and more departments want—and are expected—to use AI applications.

Under new pricing structures, fixed base costs are falling, but we are seeing a tremendous increase in variable costs, sometimes hidden by the use of agents or automated workflows. In addition to pure licensing costs, companies need to consider the costs of integration, interfaces, programming, data preparation, API usage, hosting, databases, monitoring, and data protection. Many companies have established their own AI teams; these personnel costs must naturally be taken into account as well.

The two most important findings are:

1. Do not look only at pure AI costs; consider the costs of the entire business process.
2. Active license and user management, together with monitoring costs per transaction, saves a great deal of money without compromising quality.

We have developed a four-week program with and for our customers on the topic of AI cost management, so that costs can be brought under control and made predictable at an early stage.

ADCONIA four-week program for managing the costs of AI applications

Week 1: Establish transparency and cost management

Together with our customers, we create transparency around the current use of AI in the company. We categorize user groups, examine individual AI applications, and use this information to create a usage profile. By benchmarking against other companies, we also identify additional applications.

Week 2: Analyze and categorize consumption

Regardless of the user structure, we first determine the total costs of AI applications in the company. In addition to licensing costs, we also examine data security, server, personnel, and development costs. We then compare these with the user profiles, examine the AI models being used, and thus complete our current-state analysis.

Week 3: Introduce AI governance and optimize pricing models

Management and model gateways are crucial for future AI cost management. The management gateway defines the general rules for using AI in the company and determines upper limits for usage, budgets, limits, and available models. The model gateway provides technical control of these requirements and continuously reviews the licensing and hosting models. From a cost management perspective, there is a significant difference between a sovereign cloud, an external cloud, and an on-premises solution for AI applications.

Week 4: Introduce controlling and TCO assessments

Cost management only works when specified budgets and total cost of ownership (TCO) assessments are complied with and reviewed. This is particularly true for new and additional AI applications. As with an investment or software request, the benefits must be weighed against the additional costs.

Costs for AI applications rise gradually within companies. After the first successful pilots, everyone wants their own license, comparable to an Office package. However, actual usage varies significantly between users, also comparable to an Office package. One person writes only short texts, the next mainly analyzes simple spreadsheets, and the third uses extensive databases. Unlike Office packages, however, AI pricing models offer much greater flexibility to reduce costs without compromising quality.

Please feel free to contact us.