Pick a Workflow for Your First AI Project

Your first AI project should probably not be a chatbot. I know that sounds strange, because a chatbot is the first thing most people picture when they think about AI at work. It's visible, it demos well, and it makes you feel modern. The problem is a chatbot is hard to measure and easy to deploy badly, and for a lot of small and medium businesses it solves a problem they don't have yet.

In this solo episode, I walked through why a workflow usually beats a chatbot as your first move, and how to pick a first project that you can prove worked. We build chatbots and voice agents all the time, so I'm not talking down the tech. The point is the order matters.

Here's the core difference. A chatbot is non-deterministic. The inputs come from people outside your business, and you don't know what they'll type. The outputs change too. That's a lot to plan for, and most of the work lives in the knowledge layer underneath, the part nobody sees in the demo. A workflow is the opposite. It has a beginning, an owner, inputs, decisions, and outputs. You plan what each of those are. So you can measure it, find where it breaks, and improve it. For a first project, that's exactly what you want.

Gartner predicted that at least 30 percent of generative AI projects will be abandoned, and one of the main reasons is unclear business value. Picture a company that adds a chatbot to its site. It answers some questions and collects a few leads. A month later leadership asks what it improved. Did it raise qualified leads? Cut support tickets? Save the team time? Sometimes the answer is a shrug. That's where the project dies.

Now look at what happens when a lead comes in. Someone has to read the form, decide if it's qualified, update the CRM, write a response, assign the next step, and remember to follow up if the prospect goes quiet. That's where the money leaks. Leads sit too long, good prospects get slow replies, and CRM records stay half-empty. A flashy chatbot on the front door does nothing if that flow behind it is broken. You'll feed more into a pipeline that already doesn't work. We had a stream recently about exactly this, people sitting on perfectly good leads and following up with nobody.

So I'd rather your first project be the workflow itself. AI summarizes the inquiry, checks it against your qualification rules, drafts a reply using approved examples, updates the CRM, assigns the next step, and flags anything that needs a human before it goes out. It won't look as flashy as a talking bot, and it sits closer to the revenue, where you can measure it. Did we respond faster? Did more leads get followed up? How much copy-paste disappeared? Some of that doesn't even need AI. A plain deterministic automation can move the data around, and you sprinkle AI on the parts that need judgment, like prioritizing or writing in your brand's voice.

When you do pick a first project, run it through a simple filter. Start with something repetitive that happens every day or every week, not once a year. Make sure it has clear inputs, like a form, a ticket, a contract, or a meeting transcript. Make sure it has clear outputs, like a draft email or a routed ticket. Keep the risk manageable, so a mistake doesn't break the company. Pick something the team already complains about, because pain drives adoption. And pick something you can measure. Two examples I love that aren't glamorous at all: drafting a first proposal from your discovery notes so nobody stares at a blank page, and the weekly leadership report, where AI pulls the numbers, highlights what changed, and drafts a plain-English summary for a human to approve.

One more thing that quietly kills projects is missing context. We call it business legibility. If the rules, the examples, and the data aren't written down somewhere the system can read, the workflow won't fit. Skip the project where you have to go gather context first. Pick the one where you already have good inputs, good outputs, and examples of what good and bad look like. Get the context for the next project in parallel if you want, but start where you're ready.

At the end of the day, your first AI project should create momentum and make people say "oh wow, this helps." Once you fix one painful loop, the next one gets easier, and your AI ability compounds one workflow at a time. So before you build a chatbot, write down five workflows in your business that happen every week and pick the one that's repetitive, painful, measurable, and low risk. If you want help spotting it, our AI Opportunity Map at perezcarreno.com will surface three to five workflows, flag the risks, and hand you a two-week action plan. The chatbot can wait its turn. When it's ready, it'll have a job, boundaries, and a number you can point to.