10 signs your organization is ready for AI — and the ones that say wait

Sign 1 · Ready · Strongest predictor

Someone already uses AI on their own, without being asked

If one person on your team has already started using AI for their own work, unprompted, you are ready. That habit predicts success better than anything else I look at, including budget, data, and technical staff.

The clearest case in my own record is not something I built. I set up a sales associate at a small supply company with an AI assistant and spent a few sessions teaching her how to give it context. She picked her own problem and wrote her own automation: it reads an incoming order email, enters the order, generates the purchase order, and attaches it back to the thread. Within four months she was running it daily — 78 sales orders in one month, 86 percent of everything the company entered. She also added a feedback step nobody taught her.

The same year, same person, I built and handed her a lead-generation tool. It was good. She never ran it once. Teaching the capability produced adoption where delivering the artifact did not.

Worth being honest about what changed: the company’s output did not go up. What changed is that she starts a task and walks away from it. That is attention freed, not hours saved, and it is still the thing people notice first.

Sign 2 · Ready

A named person will click approve, every cycle

Every system I build keeps a person in control: a screen to review the work, an approval step, a final sign-off. That only works if someone actually shows up for it. Not “the team.” A name.

One small company runs an AI writing pipeline this way: research, draft, a pre-submit check, then a review email to two named reviewers who approve or reject with feedback. Rejections trigger an automated rewrite. It has published eight posts across four straight months, and the owner’s editorial rejections were mined into 10 standing rules folded back into the writer, so the system got permanently better.

That last part is the test. A review gate that never rejects anything is decoration. A review gate whose rejections change the system is the thing that makes an AI project stick after I leave.

Sign 3 · Wait · The failure people do not see coming

Nobody has time to look at what the AI produces

This is the one that surprises people, so it sits high on the list. A pipeline can work perfectly and still produce nothing, because output that nobody opens is not output.

I built a funder-discovery pipeline for a nonprofit. It profiled 58 candidate funders and scored 41 of them highly. The executive director reviewed zero of them in seven weeks. Two grant deadlines passed. Nothing was broken. My own note at the time: it is a lot of steps, but it is just a record — there is no payoff for her at the end of it.

Rebuilding it as a simpler spreadsheet did not fix it either, because the problem was never scanning, it was motivation. The fix that works is flipping the ask from opt-in to opt-out: instead of “here are 58, pick some,” it becomes “I am sending these eight on Friday, reply if you want me to stop.”

Before you automate anything, answer this honestly: who opens the result, on what day, and what do they do next? If there is no answer, build the answer first.

Sign 4 · Ready

You have work that is manual, repetitive, and high volume

The obvious one, and still a good one. When a task is high volume and follows a pattern, it is a fit — and you feel the pain every week, which means the fix gets noticed.

One client had 3,161 product pages that each needed a unique, on-brand title and description. It was weeks of work nobody had time to do. The run updated all of them and flagged 22 it would not guess at, which is exactly the behavior you want.

Another had grant applications running 6 to 40 hours each, depending on complexity. Breaking that down mattered more than the number: about half of that time is research — foundation sites, past recipients, contacts, news. The research half automates well. The drafting half is judgment and stays with a person.

Look for the split before you look for the total. A 40-hour task that is 50 percent lookup is a better project than a 40-hour task that is all judgment.

Sign 5 · Ready

One person is quietly the bottleneck

In most lean teams, one person does the thing only they can do, and everything queues behind them. For one nonprofit it was the executive director’s time, spent on funder research and drafting instead of leadership.

When you can name the person work backs up behind, you have found the best place to start — not because that person is slow, but because the queue in front of them is the cost, and it is easy to measure.

One caution, which is Sign 3 wearing a different hat: the bottleneck person is often also the person with no time to review AI output. Hand the review to someone else if you can.

Sign 6 · Ready

Your tools do not talk to each other

Almost every session turns up a step where someone copies data from one system into another by hand. A catalog in one database, a store on another platform, a spreadsheet in between. That copying costs time and causes errors, and it is often where a small automation pays for itself fastest.

One 4.5-FTE nonprofit tracked its entire fundraising pipeline across four places: deadlines in Google Drive, contacts in a CRM, forecasting in Excel, follow-ups in Gmail. No view showed all of it. A salon ran bookings on one platform and payments on another and reconciled them by hand daily.

The constraint that should govern the fix: a small team cannot maintain a complex system. Connecting what you already use beats buying a new platform, almost every time.

Sign 7 · Ready

You are already paying for a tool that does not fit

A failed or overpriced experiment is a good sign, not a red flag. It means you already know the shape of what you need, which is most of the work.

One client was paying $15,000 a year for a subscription that generated research prospect lists. We replaced it with a free build on public data. The first run produced 1,681 contacts with only three email addresses in common with the old vendor’s export — it was not finding the same people cheaper, it was finding different people. The subscription was allowed to expire.

That is a cost you eliminate on day one, whether or not the new tool ever runs a second time. Line items you are already unhappy with are the easiest place to start.

Sign 8 · Wait

The knowledge lives only in someone’s head

If the expertise is never written down anywhere, there is nothing for a system to learn from yet. Writing it down is the real first project, and it is worth doing whether or not AI ever enters the picture.

Before handing a rebuilt website to non-technical staff, I wrote them a 95-page guide with 37 annotated screenshots, captured inside their own account so the screens matched what they would actually see. The useful part was not the guide. It was that seven times, an instruction could not be written simply — so we fixed the site first and then documented the fixed version.

Documentation is a diagnostic. If you cannot write the steps down plainly, the process itself is the problem, and no tool will paper over it.

Sign 9 · Wait

Your process still changes week to week

If how the work gets done has not settled, automating it now just locks in a moving target. You will spend more time rewriting the automation than you ever spent doing the task.

This is a timing problem, not a verdict. Stabilize the process, run it the same way for a month or two, then automate the version that held still.

Sign 10 · Wait

You are expecting AI to replace judgment

AI is strong at the repetitive 80 percent and weak at the judgment-heavy 20 percent. If the task is mostly judgment, the payoff is small and the risk is high.

The version of this that actually works is narrower and less exciting: AI does the lookup, the first draft, the formatting, the cross-checking. A person decides. That split is why my builds have approval gates in them, and why a rollback path in one client system generates the reset script but will not run it without someone confirming.

All four “wait” signs are fixable conditions, not permanent noes. Every one of them has a first project hiding inside it.