Insights / Company
A founder's guide to picking your first AI automation project
Akash Sharma · Guide · 2026-06-12 · 3 min read
Most founders' first mistake with AI automation is choosing the project that sounds most impressive instead of the one that actually saves the most time. "We're building an AI-powered customer intelligence platform" makes for a better pitch than "we automated one weekly report," but only one of those ships in a month and actually gets used. Here's a short, practical way to pick correctly.
Step 1 — List the tasks nobody wants to do
Not the interesting problems. The dull, repetitive ones: the weekly report someone re-builds by hand, the same three follow-up emails sent every day, the invoice reconciliation nobody looks forward to. If a task makes your team visibly sigh when it comes up, it belongs on this list. Interesting problems attract attention on their own; boring, repetitive ones are exactly the kind that get automated away and never come back.
Step 2 — Measure the real time cost
Time each task honestly for a week. Multiply by frequency. Most founders are surprised how much of a "small" task adds up to over a month — fifteen minutes a day doesn't sound like much until it's five hours a month, every month, for as long as the business runs.
Step 3 — Rule out anything that needs judgment you can't yet define
If explaining the decision rule to a new hire would take you 20 minutes of "well, it depends," it's not ready for automation yet — you need to make the decision consistent first, then automate it. This step saves more failed projects than any other on this list, because it's tempting to automate a process that feels routine but is actually held together by someone's unwritten judgment. Automating that just moves the inconsistency somewhere faster and harder to audit.
Step 4 — Pick the smallest version that's still real
Not "automate customer support." Automate one specific ticket type. Not "automate finance." Automate one specific report. The smallest real version of a project is the one that ships in weeks instead of quarters, and shipping something narrow and working beats scoping something broad and permanently six weeks from done.
Step 5 — Set one number you'll actually check
Hours saved per week, or turnaround time, not a vague sense that "it's helping." Write the number down before you build anything, the same way you'd set a target before starting any other kind of project. A number you check is a project you can defend. A feeling you had is a project that gets quietly deprioritized the first time someone asks whether it was worth it.
What a real first project looks like
If you get through this list and land on something concrete — one task, a real weekly time cost, a decision rule you could actually write down, and a number you'll check in a month — that's a real first project, not a strategy deck.
That's exactly the scoping conversation we have with every new client before recommending anything, through Consulting. The founders who get the most out of their first automation project aren't the ones who picked the most ambitious idea. They're the ones who picked the smallest one that was still real, shipped it, and used the result to justify the second one.
