AI Doesn’t Know Your Margins. That’s Why It Isn’t Making You Money

$100 dollar bills spilling out of suitcase. MORE LIKE THIS.. in lightboxes below.

getty

The question about whether to adopt AI is done, but the question of how to get money back from it is just now beginning. The Writer’s 2026 Enterprise AI Study reported that 59% of organizations are spending an average of $1 million per year or more on AI. Only 29% reported see significant returns. In PwC’s 2026 Global CEO Survey, only 12% of CEOs indicated they were seeing both increased revenues and cost reductions from their use of AI. That may be due to a missing application layer for businesses when it comes to using AI. While most companies have access to essentially the same types of tools to build an AI application layer, the results vary greatly.

The Application Layer Argument

Foundation models know language. They do not know your margins, customers or the business outcome you need. It is this missing AI business context that drives much of the disappointment people experience when trying to get a return on investment from their artificial intelligence initiatives.

AppLayerAI is a venture studio founded by Furkat Kasimov. It has one idea. The missing piece is not a smarter model. It is an application layer. This layer is the software and context between general model access and a specific business problem. It provides the model with data, rules and limitations so it will generate results that are reliable enough for others to verify.

“The biggest misconception in enterprise AI is that a more powerful model automatically produces better business outcomes. In reality, even the most advanced AI struggles when it doesn’t understand the context of your business. The organizations seeing measurable ROI aren’t simply adopting better models—they’re giving AI the structure, relationships, and business knowledge it needs to make consistently reliable decisions,” Kasimov said in an interview.

Kasimov came to this view through his work in operations rather than through his study of technology. He bootsstrapped LeadsMarket to nearly $100 million per year before moving into developing AI systems. “We’re entering an era where competitive advantage won’t come from access to the latest foundation model—those capabilities are becoming increasingly available to everyone. The companies that win will be the ones that build an intelligent application layer around their unique business knowledge, allowing AI to deliver measurable productivity gains that competitors can’t easily replicate,” he said. Now that raw model access is a commodity, the company that will win is the one that adds its own context.

The pattern matches what I have reported all year. The challenges in adopting these AIs are mostly not technical, that’s the major gap. When an AI investment fails to generate revenue, the model is rarely the problem, it is simply a tool without a defined function. Independent research points the same direction. McKinsey’s State of AI survey tested 25 organizational attributes and found that redesigning workflows had the biggest effect on whether companies saw profit impact from generative AI. The tools mattered less than the structure around them. There is no business framework for its use. There are no metrics to measure how well or poorly the AI performs. In essence, the overall AI implementation strategy was never complete. The application layer gives that gap a name.

What The Layer Looks Like In A Small Business

Enterprises create or purchase this layer. A small business can also build a bare bones version of it. Most work is done in notes, not new software. The context piece is current written guidance to the business. It covers offers, customers, brand voice, limits and basic facts that a new hire would need to know. The workflow piece links real systems to AI. This connection allows an output from a chat window to be sent to a project tracker or inbox. The checking piece defines what constitutes a correct result before the tool runs. This is what makes AI workflow automation into a process with which you will get predictable results, rather than guesswork. These parts together connect small businesses’ AI solutions to clear business outcomes.

The reason most teams do not go through these steps is because none of this is glamorous. And yet the data on paybacks continues to support this less-than-glamorous process. If you are a small business using a pre-built tool for a defined job, you can see payback in months. Broad projects without context produce the poor enterprise AI ROI numbers shown above. This same pattern occurred when 34,000 small businesses stated their AI was working. However, most were unable to show evidence of an AI ROI.

How AI Increases ROI For Small Business

Source: Institute Of Business AI

Build, Buy Or Document

For most businesses with fewer than 50 employees, the decision to develop versus purchase AI applications is largely self-evident. Enterprise (large) companies purchase custom application layers. Small businesses have a document based process along with properly configuring their existing AI toolsets. Documentation can take a weekend rather than a contract. Create your business context file once and then load it into all of your teams’ current AI toolset. Many now accept persistent instructions or project files. Define one workflow per tool. Eliminate any tool that cannot identify what it does. This is the essence of a small business AI Strategy.

Additionally this will provide an honest assessment of many layer-branded products that could become available in the future. Only some of these will actually represent a true business AI infrastructure while others will simply represent a system prompt with a logo. There are three additional questions which will determine the difference. If a vendors response time to a sales call question is slow, they may respond similarly if you experience an outage.

The Layer Still Needs A Ledger

One thing to note about this argument. A context layer can enhance performance. However, enhanced performance is still not AI ROI. “Enterprise leaders shouldn’t be asking, ‘How do we deploy AI?’ They should be asking, ‘What business problem are we trying to solve, and how will we measure success?’ AI is only valuable when it produces better decisions, faster workflows, lower operating costs, or higher customer satisfaction. If you can’t measure those outcomes, you haven’t created business value—you’ve created another technology experiment,” Kasimov said.

When a business incorporates an application layer for which there is no process for measuring, then they have simply improved the quality of unmeasured work. They have gone from producing unmeasured average results to producing unmeasured good results. Regardless of architecture, the discipline of counting every month remains the same. What were the costs of each of your AI systems? How many hours did you save with them? What deliverables were completed thanks to them? A context layer enhances results. Counting alone provides the metrics needed to create measurable ROI.

AI purchasing over the next 18 months will be layered. The language used to describe each layer, platform or studio will become increasingly complex. This complexity does not impact evaluation. Every layer, platform or studio pitch ultimately boils down to three questions.

What type of AI business context does it support?

What job does it perform?

What number demonstrates that it was worth the expense?