AI Will Optimize Your Bad Processes

Companies are under pressure to use AI.

AI can summarize documents, classify emails, extract information from messy inputs, draft replies, generate reports, and so on. The promise of faster processes is essentially unbounded.

So, when a business looks for a first AI use case, it often starts where work already hurts:

• customer support that requires a lot of human interaction
• long forms used to move a process to its next step
• customer data spread across notes, files, and people’s memories
• business reports assembled by copying and pasting into an Excel file

The pain is obvious, and the speed gains can be substantial. Painful workflows can be good candidates for AI, but they are also dangerous places to start if nobody first asks why the work is painful.

Bad Flows Are Not Just Slow Flows

Speeding up a slow process is almost tautologically a good thing. A bad flow is something else; something that is poorly defined, or something that should not be done in the first place. It depends on personal memory, informal habits, and undocumented exceptions. Framed positively, these things make up “the company work culture”, what workers absorb during their onboarding. But this is at odds with an AI implementation. It does not acquire that knowledge by sitting next to its colleagues. If the exceptions are not in its context, they do not exist.

Business reports are often part of this: produced in semi-structured Excel files that are prone to copy-and-paste errors, then emailed by the financial team to executives in a chain of CCs.

Imagine asking AI to assemble the weekly report from those files. It removes the hours of copying and pasting. But if nobody knows which file is authoritative, what a column really means, or why two totals disagree, the AI has not fixed the reporting process. It is automated guesswork, compounding the uncertain nature of AI with the uncertainty of the process.

Smooth Guesswork

In a bad flow, friction is often the only visible signal that something is wrong, much like physical pain in the body.

The manual work, repeated questions, slow reporting, and constant checking are symptoms of missing structure. If AI removes those symptoms without changing the foundation, the business may lose sight of the real problem.

Reducing the pain without addressing the cause can have the same effect as a painkiller: life-saving in the short run, but harmful in the long run.

In the previous example, the better solution is to generate the report from an authoritative source in the first place. The business can then serve it through a dashboard or API, audit it in real time, enforce granular access, and offer different views without creating more copies of the truth. All possible improvements because we tackled the right issue.

Solve Root Problems and Improve Without Limits

Structure means accurate data and clear processes. It is the foundation on which compounding improvement can be built. With structure, the system can answer what was done, why, who did it, and which facts were used. Only then can we solve the problem at the most appropriate level. This applies at different levels. Rather than only speeding up support, for instance, can we improve the product or process so that support tickets are never created in the first place¹?

The goal is not to avoid AI, but to use it as a tool for compounding growth. Since AI can speed up almost anything, use it to help replace bad flows with good ones, then optimize them.

Once the flow has structure, the next problem becomes practical: getting the right data and instructions to the model, and returning its answer where the work actually happens.

-- len