BEFORE BUILDING THAT AGENT, DO THIS FIRST
Written by Jessica Murray
Last week was a busy week. I spent it in NYC, participating in Tech Week. For those unfamiliar, each year, Andreessen Horowitz orchestrates a decentralized week-long tech conference in key tech ecosystems. This is my second year attending in New York. There were >1,500 events that covered pretty much any topic in tech you could imagine.
This piece won’t be all about the trip. I bring it up because a topic I often discuss with entrepreneurs surfaced repeatedly in 1:1 conversations and panel discussions: Before you automate, understand the business problem and map the process.
So often, the temptation is to jump to a solution: “[Insert] is so manual. Tell me how to use AI to automate it.”
One of my first questions is typically: “Have you mapped out the process?”
More times than you might think, the answer is no.
When that happens, it’s back to the basics.
The tool is only as good as the thinking and people behind it
Remember, AI is a tool. A powerful one, but definitely not omniscient.
Think about your favorite SaaS platforms built for the prior wave of automation: Asana, ClickUp, Monday.com, Zapier, HubSpot, Zendesk, Slack. Each has power. Each is also capable of creating more noise than structure when configured around a poorly understood process.
AI is no different. The quality of the output is a direct function of the clarity of the input (i.e., context), the logic of the workflow and the humans who shape it. Put a broken process in, get a faster broken process out.
In a 2024 study, RAND outlined five root causes of AI project failure. Three of them point directly at this problem:
#1 centered on the business stakeholders’ misunderstanding of the problem.
#3 focused on being enamored by the bleeding-edge technology vs. solving real problems for the intended end users.
#5 was AI being applied to problems it wasn’t capable of solving.
The teams that will see the most impact from AI-related projects will be those that do the unglamorous work of understanding the business problem and underlying process first.
Five steps before you get to build
Before jumping straight to the finish line, slow down (just a smidge). These five steps will help you get to the right place for your business.
Step 1 - Define the problem you’re trying to solve: If you lack clarity here, it will be challenging to develop an appropriate solution. Take the time to understand the fundamental business problem before taking further action.
Step 2 - Identify the stakeholders and get the right people in the room: The people closest to the work are valuable at this stage. They know where the friction lives. Bring them into the process early.
Step 3 - Map every step in the business process: When dealing with automation or AI, knowing the steps to completing the work is key. Not vaguely either. Get specific. Map each step. Note which tools are already being used to support the work and where breakdowns or inefficiencies exist.
Step 4 - Define the ideal state: Once you have a clear picture of how work flows today, you can more concretely imagine what you want the future to look like. Create that artifact for additional context.
Step 5 - Consider solutions: Only after you understand the business problem and workflow should you start getting “fancy.” You’re now in a better position to determine what set of capabilities will solve the problem at hand, match your internal capabilities, and, if the business problem is external-facing, effectively support your customers.
AI may not actually be the answer. At least now, you’ll know.
AI may or may not be the optimal way to solve the business problem at hand.
But, if it is, your life will be much easier at implementation if you follow the steps above.
AI is all about context. That context includes a clear understanding of how work happens and what you need the AI to do.
What helps with that?
Process mapping (a la Steps #3 and #4). Not a novel concept. But consistently, one of the first steps people want to skip.
To circle back to NY Tech Week, the last event I attended was hosted by IBM and titled: “How AI is transforming the workforce model – and what comes next for talent strategy.”
The panelists included the Chief People Officer of Bank of America and the Chief HR Officer at IBM. At one point, one of them said (and I may not have the quote 100%, but it’s close enough):
“You have to understand your processes and you have to understand the underlying work that goes on and what problem you're trying to solve. Then say, yes, this makes sense to use a really high-powered AI tool, a basic AI tool, or no tool at all.”
This isn’t just my soapbox. It isn’t only these two senior executives, either. It came up many times with different groups of people last week. That’s a theme worth paying attention to.
If you’re trying to figure out where AI or automation fits in your business, don’t start with the tools. Start with the problem. Follow these steps and see where you land.