Most failed AI implementations didn't fail because of the AI. They failed because the business wasn't ready for it. Bad data. Undocumented processes. A team nobody briefed. This checklist exists so you can find those gaps before you spend anything. (If you're earlier than that — still figuring out what custom AI software is and whether it fits — start with the guide to custom AI software for small B2B service businesses.)
There's a post that resurfaces on Reddit every few months. Someone spent $50,000 on an AI build, eight months passed, and they're back on spreadsheets. The top comment is always some version of: "Nobody told us we needed clean data first."
That's not a technology failure. It's a readiness failure.
The businesses that get real results from AI automation aren't the ones with the biggest budgets or the most sophisticated problems. They're the ones that did the boring prep work before anything got built. They knew where their data lived. They had a process they could actually explain. They had a person who would use the tool once it existed. There are 33.3 million small businesses in the US (US Small Business Administration, 2024) — and almost all of them have at least one workflow that would benefit from automation. The readiness gap is what separates the ones that capture that value from the ones that buy software and shelve it.
Twelve questions. Four categories. Honest answers only.
Your data
This is where most implementations fail first.
1. Is the data this process uses in one place — or scattered across inboxes, spreadsheets, and five different tools?
2. Is the format consistent? Same fields, same naming, same structure — or does it depend on who created the record and when?
3. If someone asked you to pull 90 days of records for this process right now, could you do it without cleaning anything first?
A "no" to any of these isn't a technology problem yet. It's a data problem. Fixable — but it needs to be fixed before the build, not during. (Even with clean data, whether the problem itself is worth building for is its own question.)
The AI doesn't clean up bad data. It automates what's already there, mess included.
Your process
AI automates patterns. If your process doesn't have a pattern, there's nothing to automate.
4. Could you write a step-by-step description of this process right now, no prep, in 15 minutes — without hedging?
5. Does this process happen at least 10–15 times a month?
6. Is the output mostly the same shape every time, or does it change significantly depending on the situation?
A lot of times, question 4 is where businesses realize their process lives in someone's head and not on paper. That's not a dealbreaker, but it means documentation comes before automation. Either way, that work is valuable.
If the answer to question 5 is no, the volume math probably doesn't support a custom build. Worth knowing before you go further.
Your team
AI projects don't die because the tool didn't work. They die because nobody used it.
7. Does the person currently doing this task know you're exploring automation — and are they on board?
8. Is there someone internal who can manage or troubleshoot the system once it's built, even at a basic level?
9. Will leadership visibly back the project, or is this a one-person initiative hoping to get noticed?
Question 7 is the one most people skip. If the person running the process thinks they're being replaced instead of freed up, adoption rates collapse. In reality, the best implementations happen when that person is part of designing the system — they know the edge cases, they know where it breaks, they have skin in the game.
Question 8 is about what happens after we leave. If nobody internal can handle basic troubleshooting, that's a gap to plan for explicitly — not a reason to skip the build, just something to solve up front.
Your expectations
This is where most timeline and budget failures come from.
10. Can you define what "this is working" looks like — with a measurable outcome — before anything gets built?
11. Is there a human who reviews AI output before it touches anything customer-facing or financial?
12. Are you prepared for a first version that's 80% right, with room to iterate to 95%?
Question 10 sounds obvious until you try to answer it. "We want this automated" is not a success metric. "We want proposal creation cut from four hours to 45 minutes, measured after the first full month of use" is. Define the outcome before the build and you'll know six months later whether it worked.
Question 11 is non-negotiable. Specifically in the case of anything that goes out under your name — emails, reports, client deliverables — there should be a human in the loop before it lands. This isn't a hedge against AI capability. It's how good systems stay good. The review step is where the system learns what it doesn't know yet.
Question 12 is about calibration. The first version is never the final version. The businesses that get real results treat early output as a strong draft and invest in the iteration cycle. That's where the gains compound.
How to read your answers
All or mostly yes? You're ready. The work ahead is picking the right problem and the right partner.
Mixed? You're close. The "no" answers tell you what to address first — most are fixable in weeks, not months.
Mostly no? Not a reason to stop. A reason to sequence properly. Clean the data, document the process, brief the team, then build. The businesses that skip that phase are the ones writing Reddit posts about $50,000 builds.
The boring prep work isn't a phase you skip. It's where the return on the investment actually starts.
If you're not sure where your specific situation lands, the quiz on our homepage takes three minutes and gives you a straight read on whether you're in the right place to start.