For buyers
When not to deploy AI in your company?
A company sometimes arrives with a decision to deploy AI already made. After an initial review of its processes, we then say plainly that AI is not a good choice for them here and now.
These conversations really do happen. Below, we describe three situations in which we advise against deploying AI. We also give a question that helps identify them earlier.
The process must be put in order first
AI speeds up work with an agreed workflow: inputs, rules and an output. If a process has no person responsible and every case follows a different path, automation will only replicate that mess faster. The organisation must first establish one path and identify the person responsible for the process. We advise on how to put it in order and help carry out that work. Without it, a deployment has no foundation.
The result can be reached more simply
Some problems reported as "we need AI" can be solved by a report, a rule in the existing system or a change to one step in the document workflow. Such a solution is cheaper and faster, and maintaining it requires no new skills. A language model is an AI system that works with text. When a simple filter is enough, using such a model will not repay its cost. AI begins to pay for itself only when a simple tool can no longer cope with a growing number of documents.
Nobody will be able to measure the effect
A deployment without a baseline ends in a disagreement over the numbers. If a company does not know what a process costs today, there is no value against which to compare the result. That is why we begin with a baseline measurement and a single process. Data access is read-only at the outset. The metric is the responsibility of the person who runs the process on the company's side, and we calculate the result from company data using a rule written down before work begins, so it does not depend on our opinion. If a baseline cannot be established, we say so plainly and do not begin the deployment.
From our practice
If nobody currently measures the cost of the process, the first task is to establish that number. Only the result of that measurement makes it possible to assess whether an AI deployment is worth pursuing.
A question for every AI supplier
A supplier that earns money by deploying AI should be able to state in advance the conditions under which it would advise against such a deployment. It is worth asking every supplier, including us, one question: how will you know that AI is not the right tool here?
This question helps both sides. We can focus on solutions with measurable results. The company does not waste a quarter and can put its budget into work that will produce a result.
Have a process that takes up hours and whose result can be measured? Tell us about it at kontakt@uruchom.ai. Within one working day, we will send three questions about the process and propose a 45-minute call. We will prepare the first-stage outline after that call. It will begin with measuring the baseline, and if AI is not the right tool here, we will say so plainly.
Back to all posts