Seeing Results with AI: Start Here
Welcome to part 1 in this (maybe) mini-series, inspired by last week's post on lacklustre experiences with AI. Over the next few posts, we'll dig into what it takes to make AI work for you, or to verify, with confidence, that it doesn't.
Problem Selection
This is the very first step where things go off the rails. The most common ways:
Picking no problem. A vague blanket mandate to "do AI" from the board, the CEO, or plain fear of missing out.
Picking a vague problem. You need to be able to explain what success looks like. Not necessarily with hard numbers; qualitative goals are fine. As long as you can articulate a difference between the status quo and the desired end state, we've got something to work with.
Picking the wrong problem.
That last one deserves unpacking. What makes something "the wrong problem"? If we assume you need to find something to do with AI, then a wrong problem is one where AI can't help. But even among things AI can do, some are pointless:
In a workflow with a bottleneck (and they all have one!), speeding up anything other than the bottlenecked part is pointless, AI or not.
Letting AI generate things where reviewing them takes as long as it would take an expert to create them saves no time. It just leads to exasperated experts.
And don't forget: maybe the process you're looking to automate shouldn't exist at all.
(If the main bottleneck can't be fixed with AI, that's fine. It just means you fix that one first before looking for AI solutions elsewhere.)
Where to Look for Good Problems
A few keywords to get your creative juices flowing. Chances are, you already have a good intuition for which parts of your organization fit these. If not, a value stream mapping exercise can surface them more rigorously.
Repetitive, frequent, high volume.
Requires skilled workers but not all their mental faculties. ("Senior engineer must review every document...")
Narrow context. The input itself, some internal documentation, maybe a few specified sources; that's enough to perform the task.
Clear downstream impact. Speeding up the task, or freeing the people who do it for higher-value work, has a demonstrable positive effect on the business.
Beyond existing workflows, dig deeper. In any scenario involving the intake and processing of information, imagine what you could do if AI handled much higher volumes. Any task that involves "scouting," scanning sources and surfacing relevant finds for human follow-up, can benefit massively from AI doing the legwork at scale.
Armed with these pointers, you can come up with AI use-cases that hit harder than "I dunno, maybe write emails faster and summarize them?"
