A stack is not a system
Meta description: Companies reach for new AI tools when an unmeasured value leak in pricing, segmentation, or offer fit is the actual constraint.
The inventory exercise in these companies produces the same silence. Someone — usually the CFO, looking at renewals — asks how many AI tools the company has. The honest answer requires a spreadsheet, and the spreadsheet, when assembled, surprises everyone: thirty tools, forty, more. Each function bought its own. Each purchase was defensible. There is no map of how any of them relate, because nothing relates them. The spreadsheet becomes, by default, the company’s AI architecture document — a list of invoices standing in for a system that was never designed.
The AI landscape offers endless invitations: a new writing system, a new video generator, a new agent, a new analytics platform. Each promises to save time or produce growth. Without a clear goal, a company can spend months evaluating software and never answer the basic question of what the business needs to accomplish.
What does the tool list reveal about ownership?
Read as an org chart, the tool list is revealing. Each tool has a buyer, and the buyer is whoever felt the pain most acutely: marketing has its writing tools, sales its assistants, engineering its copilots, support its summarizers. Ownership of tools is distributed and ownership of the whole is absent. I have asked, in these companies, who owns the company’s AI capability as a system, and the answers point to committees. Committees govern. Nobody owns. The tool list is what an organization looks like when every part may buy and no part must fit.
Why does a platform purchase arrive too early?
Miklos Roth, AI marketing consultant, Budapest
Before any vendor list, I ask for the objective. “Use more AI” is not a business result. A precise objective reveals which inputs are necessary, who must check the output, and which measure will indicate whether the approach was worthwhile. It also makes the leak visible: if the stated goal is growth but existing customers leave because the experience does not meet expectations, the model’s acquisition plan is answering the wrong question — and no new tool fixes that.
The false explanation is consolidation: the sprawl needs a platform decision — one vendor, one suite, one contract. I have watched the platform purchase land on the sprawl and add to it, because the platform was bought to replace tools whose jobs were never understood. The functions keep their tools — the platform does not quite do what the tool did — and now the company has forty-one tools, one of which is expensive and strategic. Consolidation was the correct instinct aimed at the wrong object. You cannot consolidate toward a system that does not exist.
What work cannot cross tools?
Read more: more about Miklos Roth.
The invisible cost in these companies is the work between tools. Output from one tool becomes input to another by copy-paste, by re-explanation, by a human carrying context across a gap the tools do not share. Each tool is locally impressive. The seams are staffed by people doing integration labor that appears in no job description. I have watched capable employees spend a third of their week being the API between systems that were each purchased to save time. The savings were real, tool by tool. The seam tax ate them.
What pattern marks inflation rather than experimentation?
Experimentation has a shape: small, time-boxed, with a question being answered and a plan to stop. Tool inflation has the opposite shape: purchases justified by enthusiasm, renewed by default, evaluated by the people who chose them. The marker I use is the renewal conversation. In experimentation, renewal requires evidence the question was answered. In inflation, renewal requires no conversation at all — it happens on the calendar, and the spreadsheet grows by whatever was added this year minus nothing.
The discipline is simple: start with the objective, then select the tool, method, and workflow. If the purpose is to improve customer understanding, a specific research or drafting system may help. If the purpose is to increase retention, a different class of tool and a different workflow may matter more. When the desired outcome is clear, it becomes easier to decline fashionable products that do not help. When the outcome is vague, every tool can feel relevant.
[AWAITING MIKLOS — scene: second scene]
What question should precede any consolidation?
Consolidation conversations in these companies start with the vendor list, which is starting at the end. The vendor list is an artifact; the question that produces a system is about the work the tools are fragments of. I have noticed that companies which ask the prior question consolidate calmly and keep a few tools on purpose. Companies that skip it consolidate loudly and re-sprawl within eighteen months, because the purchasing pressure the question would have addressed is still there.
The question that finds the leak before the tool list: what measurable change is this organization actually buying — and is the constraint upstream of anything a model writes?
PR excerpt (80–120 words, Miklos’s own words): The AI landscape offers endless invitations: a new writing system, a new video generator, a new agent. Each promises to save time or produce growth. Without a clear goal, a company can spend months evaluating software and never answer the basic question of what the business needs to accomplish. “Use more AI” is not a business result. A precise objective reveals which inputs are necessary, who must check the output, and which measure will indicate whether the approach was worthwhile.
