In my work as a data consultant, I have come across numerous cases of business managers with an appetite to implement AI solutions on top of bad data structure. By their versatile nature, Apps like ChatGPT and Claude project an ability to mop up all manner of spills. Unfortunately it's not true. As clever as AI is, it doesn't overcome the adage: Shit in, shit out! Feed an AI with nonsense data and it'll give you nonsense results. Actually, with AI it's highly possible to give it great data and still get nonsense back, but that's another discussion.

With that in mind, I posit another approach. Instead of rushing to get AI in, which risks poor ROI, destabilising your operation and completely bending the business out of shape, do the opposite. Slow down.

That's not to say anyone should rest on their laurels. What I'm saying is that we should all be using AI as a catalyst to improve the core operation by addressing productivity inefficiencies and knowledge & skill silos.

What this means in practice is that any business hoping to jump on the AI bandwagon needs to do a few things first:

1. Conduct a business review: Have an outside set of eyes appraise your entire business. Focus on operational and regulatory risks, up and cross skill opportunities, strategy enhancements over all departments.

2. Conduct a process & procedure review: document existing processes and cross-reference them against documented procedures, identifying inefficient or mis-aligned task steps, or gaps in documentation.

3. As an output of the process review, ensure that system-driven processes dovetail with people skills and authorisations, and if you have to alter and improve those processes, that everything is well documented.

4. Set up a formal process for reviewing procedures with an assigned team.

Ironically, anyone embarking on these tasks can use AI to support in the completion of them. If you complete these steps (or if you're already at this point) then you'll be in a much better position to make the most of an AI solution. But these actions still don't address how to implement AI into your business.

Most AI innovations are using it to adapt traditional workflows. Computers work in a highly logical way with brittle if…then, and/or conditional statements. One of the revolutionary ways that AI changes computing is that it breaks that paradigm, at least in a meaningful way. A Large Language Model (LLM) is still a logical process, but the logic is so complex that it emulates a degree of chaos, i.e. hallucinations, and these models can produce results that no-one (including their creators) could predict.

In this sense they work more like people than computers. Whilst we are logical, we are not logic machines. Our decisions are influenced by any number of conditions including our mood in the moment, recent experiences, even our breakfast can play a part. We operate on gut-feeling much of the time, with imperfect information and memories. What if our computers do the same? Would that make them more adept at managing complex real-world situations? Could we hand over management of a business function to an AI if we gave it a dynamic environment with fluid rules to work with? For me, flexibility & ambiguity might be key to the AI-lead innovation we've all been missing. The smartest businesses won’t use AI to cut people out - they’ll use it to help their people achieve more than ever before.

*https://www.wsj.com/articles/how-the-ai-boom-is-leaving-consultants-behind-c9088fda?mod=cio-journal_lead_story

At ATP, we’re cutting through the noise - helping businesses move past experimentation to execution, building practical AI strategies that deliver measurable results... get in touch: Joe Wright , e-mail: jw@atp.ltd