The Simple Reason AI Strategies Keep Falling Short
Your portcos are wasting tokens and proliferating amateur decision-making
Sponsored by Fine Tune Expense Management
By Rich Ham
Companies are spending big on AI now, and often for good reasons. The range of potential applications keeps expanding, and every portfolio company seems to have another process somebody believes can be automated, accelerated or improved. As Claude, ChatGPT, Gemini and others square off, the natural response is to compare models, evaluate platforms, and start testing use cases.
But smart operators are starting to ask the right question: what does the AI actually know about the problem I am asking it to solve?
That question gets more critical the narrower the problem becomes. A general-purpose model can know a good deal more than the user about manufacturing or finance, broadly speaking, while lacking the information needed to solve a specific and nuanced operating challenge. The decisive facts may never have been published. They may be scattered across internal systems or closely held by people who have spent years doing the work. In these cases, AI models produce lackluster work informed by public material that may be stale, incomplete or written by somebody with something to sell.
While powerful models are increasingly available to everyone, the information that makes them useful for a specialized task is much harder to obtain. An experienced practitioner curates their intelligence over the course of years, knows which information matters, what is missing, and whether the available material deserves to be trusted. Those who have turned this information into institutional knowledge now hold the missing piece to most companies’ AI strategies: good, clean, niche data.
More tokens won't fill the data gap
Teams buy LLM licenses and consume tokens, the units used to measure a model's processing, as they prompt and re-prompt a general-purpose system. They keep refining instructions, hoping it will find the decisive information. If nobody supplies the missing data or methodology, that effort can become an expensive loop, with employee time adding to the bill.
The consequences become more expensive when somebody acts on the answer. An operating forecast built on false assumptions can distort a buying or investment decision. Even an accurate summary of incomplete records can mislead users. Across a portfolio, the cost of those decisions can exceed the token spend itself.
I think about AI much the way I think about power tools. Someone who learned the work with manual hand tools understands the material, every standard, benchmark and measurement and what a finished job should look like. An expert with a power saw knows how to measure twice to cut just once. Without experience, amateurs just make more bad cuts, faster.
Why portco spend is especially exposed
Procurement is particularly vulnerable to this combination of incomplete data and insufficient category knowledge. Armed with contracts, invoices, and sourcing platforms, the function typically believes itself to be well-supplied with information. But in complex service categories, those records are nearly always lacking critical facts needed to determine the health of a commercial arrangement. Suppliers largely control what the buyer receives.
Invoices arrive in different formats and with different levels of detail. Equivalent services may have different names, while identical labels may describe different services. Charges can be bundled; invoiced quantities are often wrong, and information needed to validate a fee tends to be absent. A tidy invoice can still be missing the most important fields, and recognizing this requires category expertise that even the sharpest generalists nearly always lack.
A capable procurement professional might negotiate an indirect service once every three to five years. The supplier, meanwhile, works in that category every day. This imbalance shapes what information gets shared, and which terms receive attention. A buyer can understand sourcing perfectly well while lacking the knowledge needed to challenge a category-specific definition or recognize a provision that will become expensive later.
In uniform rental, a low unit rate tells you little until you understand the inventory it applies to, and all the related peripheral charges which will ultimately determine the bill. In security guard services, an hourly rate also needs to be evaluated against staffing and whether billed coverage was delivered. A model asked to find a good price typically never receives enough information to judge either arrangement properly.
Easy-to-use tools can conceal those gaps. AI can compare proposals and recommend the lowest apparent cost, but an incomplete scope or inconsistent assumptions can turn a polished analysis into a rubber stamp for a weak deal. An invented benchmark or misunderstood contract term can influence a multiyear commitment. More tokens spent analyzing the same documents will not necessarily reveal the missing commercial facts. Somebody has to know what else to ask for.
The operating intelligence PE should value
As AI makes general intelligence cheaper and more widely available, proprietary operating information and the workflows built around it can become more valuable. That is a PE thesis with applications well beyond procurement, extending to everything from insurance brokers to specialty maintenance businesses, healthcare administration, logistics, compliance providers and payments businesses, among myriad others. Years of operating records, interpreted by people who understand the work, may become a more productive asset when AI makes them easier to use.
Fine Tune has provided my firsthand experience with this truth. For more than two decades, we have concentrated on complex indirect services, including uniforms, janitorial, waste and recycling, security guard services, pest control, energy and utilities, and many other facility service categories. Our work has produced proprietary pricing, contract, invoice and service information a general-purpose model cannot simply find online. Our eMOAT platform brings together category expertise, supplier intelligence, contract data and performance history across client programs.
Our people use AI every day. But they also know what a complete dataset should contain, what suppliers tend not to volunteer and how to validate what arrives. They understand which differences need to be normalized before comparisons can mean anything. We have codified much of that experience into processes and playbooks, informing our AI with both proprietary information and methods developed by people who understand the specific expenses we manage.
This is the hand-tool experience in practice. Our category specialists learned to obtain the information, challenge supplier explanations and follow results through to actual spend. AI can extend that work by processing more information and identifying potential exceptions. The people who know the category can validate whether a finding reflects an opportunity or an incorrect assumption. Good inputs improve the system without eliminating the need to check its work.
What PE firms should ask before spending
Everyone can buy access to the same AI technology. The key is knowing what to feed it, having information your competitors cannot obtain, and possessing enough expertise to know when AI’s answers are wrong. For PE firms, that is a reason to look harder at the operating intelligence inside a business before paying a premium. Before funding another initiative, ask what information the system needs, whether it will have that information and who knows enough about the job to recognize what is missing. Find out how the team will test its answers against real results, and what a competitor using the same model would struggle to reproduce.
As access to powerful AI becomes commonplace, the investment question shifts to what a business knows that its competitors do not. The best AI investments build on the data and judgment developed by experts through years of hands-on work.
Rich Ham is co-founder and CEO of Fine Tune Expense Management. Fine Tune helps PE-backed and other multi-site organizations reduce and control complex indirect-services spending through category-specific expertise, supplier management and ongoing oversight
