AI Monetization: Business Models and Strategies

Lisa Ernst · 17.11.2025 · Technik · 12 min

Companies are investing billions in generative AI, yet few are seeing measurable returns. This article explores why most AI projects fail to deliver a return on investment (ROI) and what the successful five percent are doing differently.

Source: Proprietary Visualization

Introduction

Approximately 95 percent of surveyed companies see no measurable profit yet from their gen AI initiatives, despite an estimated 30 to 40 billion US dollars flowing into such projects worldwide. At the same time, MIT describes a small group of around five percent that achieves millions in savings or new revenue with the same technologies – a kind of gen AI elite. The crucial question behind this: What are these few doing differently – and how can more companies truly earn money with artificial intelligence instead of just collecting expensive pilot projects?

When we discuss how companies can make money with artificial intelligence, it's fundamentally about returns: Does an AI project yield more measurable benefit than the money, time, and risk invested in it? IBM distinguishes between hard ROI – directly measurable in euros, francs, or dollars – and soft ROI, such as better decisions, less frustration within the team, or more satisfied customers.

Generative AI refers to models that create content: text, code, images, audio, or entire dialogues. They are embedded in tools like chatbots, copilots, or content engines and are intended to accelerate processes or enable new products. Many companies today run precisely these kinds of applications: document summaries, automated emails, marketing texts, software prototypes.

Agentic AI goes a step further . Instead of just reacting to individual commands, a system acts like a digital employee: It pursues goals, plans steps, coordinates multiple AI models and systems, and operates largely autonomously. For example, an agent can review customer inquiries, search internal databases, initiate workflows in ERP and CRM systems, and ultimately complete a process entirely – including follow-ups, error corrections, and documentation.

Crucially for real ROI, these systems don't just deliver 'cool demos' but concretely connect to business metrics: lower process costs, higher conversion rates, more revenue per customer, fewer downtimes, or significantly shorter cycle times in complex processes.

Current State & Challenges

The MIT analysis, which many current reports rely on, describes a significant divide: In recent years, companies worldwide have invested around 30 to 40 billion US dollars in generative AI , yet about 95 percent of the surveyed firms report no measurable profit increases or cost reductions from these projects. Only about five percent of the organizations studied claim that integrated gen AI pilots have already delivered 'millions in value', for instance, through saved expenses or new revenue.

TechRadar summarizes this situation pointedly : Almost all gen AI pilot projects in companies fail – with the consequence that many models deliver far less than marketing promises. According to this report, 95 percent of the surveyed companies have seen only 'very little' or no effect of their LLMs on operational business. Tom’s Hardware cites the same MIT finding with the note that most implementations have 'no measurable impact on the profit and loss statement' – primarily because they are poorly integrated into processes.

Other studies also present a mixed picture. One IBM-Analyse unter C-Level-Führungskräften concludes that only about a quarter of AI initiatives have delivered the expected ROI and only 16 percent have actually been scaled – most projects get stuck in the pilot phase. While Gartner expects over 80 percent of companies to use Generative AI by 2026, it anticipates that only about 20 percent will cleanly measure ROI.

On the other hand, individual corporations show that AI and agent programs can be very profitable – if they deeply impact the business model. IBM reports that the company-wide use of AI and automation since the beginning of 2023 is expected to result in overall productivity gains of around 4.5 billion US dollars and contributed to a free cash flow of 12.7 billion US dollars in 2024. Specific measurable effects underpin this: millions of saved working hours, greatly accelerated HR and IT processes, and high automation rates in customer service.

Another much-discussed example is Salesforce. CEO Marc Benioff reports that the company reduced its customer support staff from around 9,000 to 5,000 employees , while AI agents via the Agentforce platform handled about 1.5 million customer interactions – with similar satisfaction scores as human support teams. At the same time, a Recherche von Business Insider shows that less than half of the 12,500 Agentforce customers actually pay for the product and less than two percent have more than 50 Agentforce conversations per week – indicating that many installations lag far behind ambitious ROI promises.

In short: Most companies are experimenting, a few are already counting concrete millions – and a large gap exists between them.

Motives & Context

Why are companies continuing to invest so much money in generative AI despite disheartening interim results? One motive is simply the magnitude of the bets already placed: The eleven largest cloud providers alone are expected to invest nearly 400 billion US dollars in infrastructure by 2025 , driven primarily by the computational demands of large language models. Those building so much capacity need customers – and thus convincing stories about future productivity jumps.

For many executives, AI is also a strategic signal to investors: We are not missing the next wave. Media reports about supposed productivity revolutions increase this pressure, even if one's own numbers don't yet align. MIT speaks here of a „GenAI Divide“ between a few 'winners' who have consistently restructured processes and a large majority that is content with loosely coupled pilots.

Furthermore, there is the dynamic around Agentic AI. Analysts like Gartner see it as one of the most important technology trends of the coming years, but simultaneously warn against 'agent washing', i.e., cases where classic automation solutions are marketed as 'agents' without possessing the necessary autonomy. Reuters reports that over 40 percent of Agentic AI projects are likely to be discontinued by the end of 2027 – partly due to rising costs and unclear business benefits.

At the same time, these same analysts promise that agents can bring enormous efficiency gains in the long run. Gartner estimates that by 2029, Agentic AI could autonomously handle up to 80 percent of regular customer service inquiries and potentially reduce operational costs by about 30 percent. This type of forecast fuels investments, even if the actual implementation is still in its infancy.

Source: wohlstandsnavigator.net

Artificial Intelligence as a Driver for Financial Value Creation.

There are also legitimate concerns: The President of the Signal Foundation, Meredith Whittaker, warns, for example, that agents that independently perform tasks will almost inevitably need to access sensitive data such as contacts, payment information, or calendars – and that this involves enormous data protection and security risks. Therefore, anyone wanting to derive ROI from Agentic AI must not only restructure processes but also rethink governance, security, and compliance.

Ultimately, three forces collide: the economic interest of providers and cloud operators, the innovation and competitive pressure within companies – and legitimate concerns regarding security, quality, and the world of work. Deciding within this tension when an agent truly pays off is significantly more complicated than with a classic software upgrade.

The panel 'AI ROI in Practice: What Leading Enterprises Get Right' clearly demonstrates how large companies deal with these very tensions internally and which metrics they use for true ROI.

Facts & Claims

It is proven that a large majority of companies have not yet seen direct financial benefit from their gen AI investments. The MIT-Auswertung mit rund 95 Prozent „Null-ROI“-Projekten is picked up by several independent media outlets and consistently described. IBM confirms with its own data that only a small portion of AI initiatives deliver the expected ROI and that scaling beyond pilot projects is the exception. At the same time, concrete case studies – for example at IBM itself or in individual service areas of Salesforce – show that with consistently restructured processes and agent architectures, significant savings and productivity gains can indeed be achieved.

It is unclear how representative the currently available figures are for all industries and company sizes. The MIT-Studie basiert auf einer begrenzten Stichprobe of specialists and managers and primarily measures short- to medium-term effects; long-term innovations or indirect competitive advantages are only partially captured. Gartner points out that while many companies are introducing gen AI tools, they have hardly any reliable baselines, data quality, or KPI systems to strictly calculate ROI. Studies like Kanerika's also highlight that intangible effects in particular – such as faster innovation or better customer loyalty – are difficult to map with classical ROI models.

Statements like 'Companies that don't introduce agents now will soon be out of the market' or 'Gen AI automatically replaces a large part of the workforce' are false or at least misleading. The MIT-Analyse findet keine Hinweise auf massenhafte KI-bedingte Entlassungen , but rather gradual effects like not refilling certain positions. At the same time, the Salesforce example shows that aggressive cost reductions through agents can trigger significant internal tensions and acceptance problems – and that investors are now very skeptical of AI promises if revenue figures do not follow. Likewise misleading is the image of Agentic AI as a "finished miracle machine": Gartner expects that over 40 percent of Agentic AI projects will be abandoned in the coming years , precisely because the business value remains unclear.

Reactions & Counterarguments

Technology providers and large platforms often present a very optimistic picture. Marc Benioff speaks of agents as a new form of work and promises that Agentforce not only reduces costs but also taps into previously unaddressed leads in the six- to seven-figure range. At the same time, Business Insider, wie Analysten vielen dieser Aussagen derzeit „null Glaubwürdigkeit“ beimessen, documents as long as reliable figures on revenue growth, margins, and adoption are missing.

Gartner and other market researchers hold a much more ambivalent position. On the one hand, they see Agentic AI as a key technology that could autonomously handle a majority of standard customer inquiries by 2029. On the other hand, they warn of inflated expectations, insufficient data quality, and misincentives that could lead to costly project failures.

Source: gruender.de

AI Consulting: Human Expertise Meets Artificial Intelligence in a Trillion-Dollar Market.

A third group – for example, IBM or specialized integrators – tries to focus more on concrete use cases and reliable metrics. IBM emphasizes in several contributions that successful Agentic AI projects usually start with clearly defined cost reductions, require clean baselines and process analyses, and only aim for new revenue streams and business models in a second step.

Critical voices – for example, from civil society organizations – remind us that agent architectures can also create new surveillance and dependency risks. Meredith Whittaker von Signal warnt davor, to equip agents with far-reaching access rights to personal and business-critical data when it remains unclear how exactly these systems work and which cloud providers have access in the background. This makes it clear: It's not just about ROI, but also about the price companies are willing to pay for possible efficiency gains – financially, organizationally, and socially.

Practical Implications & Open Questions

If you are considering how your company can make money with artificial intelligence, the most important realization is: ROI rarely arises from simply implementing a tool, but from consistently redesigning processes. Studies show that successful projects often start where there are clearly measurable process costs, lead times, or failure rates – for example, in document processing, customer service, or internal IT. IBM empfiehlt, vor jedem Projekt eine Prozesszerlegung zu machen: How long does the process take today, what does it cost, where are the bottlenecks – and which key figures should concretely change with AI? Secondly, you need a clear baseline, otherwise, you cannot later prove that agents are actually saving time, money, or risks. Thirdly, it is worthwhile to consider hard and soft effects separately: direct savings and additional revenue on the one hand, satisfaction, innovation capability, and resilience on the other.

For Agentic AI, additionally applies: Without a clean data basis and robust process design, the promise of 'digital employees' quickly becomes an expensive, unpredictable system. TechRadar bringt es mit „Garbage in, Agentic out“ auf den Punkt and points to cases where poor data quality makes autonomous systems practically unusable. At the same time, companies like IBM show that agents in clearly designed workflows – for example, in HR services or for IT tickets – can combine very high automation rates with measurable savings.

Practically, this means for you and your team: Before any agent or gen AI project, ask a few simple, hard questions. Which key figure should change – and by how much? Which data does the system use, and how good is it really? What does the process look like afterward, including escalations to humans? And how is it documented what the agent decided? Frameworks from IBM, Gartner, and specialized consultancies provide checklists and metrics that you can adapt.

The presentation 'Agentic AI ROI: From Automation to Decisions' shows with concrete examples how companies use agents to actually achieve measurable financial effects and new decision-making qualities from automated tasks.

Despite all the numbers, central points remain open. There are still few independent, cross-industry meta-studies that evaluate the actual ROI of gen AI and Agentic AI projects over several years and economic cycles. The MIT spricht zwar von einem „GenAI Divide“ between a few very successful projects and a large mass of experiments that have little impact, but how this gap will develop long-term is unclear. It is also open whether the current investment wave will prove to be an over-investment or a basis for sustainable productivity leaps.

A second open point is the measurability of soft ROI. How much is it worth if product teams can test prototypes faster thanks to AI, marketing teams find better target groups, or executives make decisions based on broader data? IBM betont, dass solche Effekte langfristig entscheidend sein können, but fit poorly into traditional ROI formulas. Consultancies like Kanerika try to establish combined KPI systems for this that combine financial, productive, customer-related, and risk-related indicators – an area where much is likely to change in the coming years.

Finally, the question remains how regulation, accounting rules, and supervision will deal with AI investments in the future. There are still no uniform standards on how companies should disclose AI projects in annual reports, sustainability reports, or risk reports and make their economic viability transparent. For you, this means: Your own internal standard for measurement, documentation, and governance will likely be as important in the coming years as the choice of models or platforms.

The current status can be soberly summarized as follows: Most companies are still looking for a sustainable business model for Generative AI and Agentic AI, while a small minority is already achieving clear financial effects. The data suggest that ROI does not arise from the technology itself, but from the way companies consistently rethink processes, organization, and data basis around this technology.

If you really want to make money with artificial intelligence, a sober look is worthwhile: start small, measure clearly, radically align with concrete business goals – and understand agents not as magical beings, but as sophisticated building blocks of a new process design. This increases the chance of moving from the large group of experimenting companies to the small group of those who are already drawing real, verifiable returns from AI today.