A not-insignificant percentage of our clients first approach us after trying and failing to implement AI. The most common reasons why are:
- The tools they bought don’t work.
- They didn’t automate the whole task (the team has to babysit the computer).
- The reliability of the AI wasn’t there.
- The AI was too expensive per-run.
Generally, these companies ask us “what went wrong?” and then “how can we do it better?”
Kudos to them for stepping back up to the plate. Here’s our process for automating stuff. Whether you work with us or with anyone, I recommend following it.
Step 1: Understand the Process First
Start with the problem. With AI & automation, the problem is that a task takes too long or is not worth the human’s time.
Get the SOPs.
Map the workflow from trigger to output.
Talk to the people who actually do the job.
Understand the systems involved, the decisions being made, and what happens when something goes wrong.
Most organizations have processes that live only in people’s heads. The “official” workflow and the real workflow are not the same thing.
You cannot automate what you haven’t defined. If the process is ambiguous before automation, it will be more ambiguous after.
The first question is not “what can AI do?” It is “what exactly happens here, and what does good output look like?”
This work is not sexy and should not be done by the CEO. Someone in operations should be doing this work.
Step 2: Automate It
Once the process is documented, build.
For straightforward workflows, tools like Claude or off-the-shelf automation platforms handle the job. For complex, multi-system workflows, hire developers.
Claude Code has transformed what was a $100,000 buildout in 2018 into a $20,000 buildout in 2026. Developers are seriously not as expensive as they used to be. We recently had a prospect tell us what they wanted (it was a lot), then they asked “is this like $1,000,000? $2,000,000?” And we said “No, maybe like $100,000 or $200,000.”
Build end to end.
A complete automation has a defined trigger, processes the work without ambiguity, handles exceptions, and produces an output the next step can use. If it doesn’t do all of that, it isn’t done.
Sante Realty Investments is a good example of what this looks like in practice. Their deal-sourcing process was manual, slow, and kept their best people buried in low-value research.
We started by mapping the workflow end to end: where did the data come from, what criteria mattered, and what did a qualified deal actually look like?
Once that was defined, we built an automation that pulled data from multiple vendors, compared it against their investment criteria, and color-coded deals green, yellow, or red for committee review.
The result: 150 deals prequalified in under 20 seconds. That’s a 1,000X improvement over the manual process. The people who had been doing that research by hand were freed up to focus on actual deals.
Step 3: Start With Your Best People
When companies roll out AI, the instinct is to focus on struggling employees. Lift the bottom. Close the skills gap.
The problem is that your lowest performers are also the least likely to use AI well. They’re slower to experiment, less likely to integrate new tools into their workflow, and more likely to use AI as a crutch.
Your highest performers are the opposite. They’re already looking for edges. They’ve got a laundry list of things to do once they finally have the mundane work off their plates. They understand the work deeply enough to know exactly where AI helps. When you give them better tools, they use them.
Productivity gains compound at the top. The gap between a good employee and a great one is already large. AI widens it. That’s where ROI shows up first.
The pattern is consistent across every industry we’ve worked in. Your best people get more out of AI because they have the judgment to direct it.
The Sequence Matters
Most organizations get this backwards. They chase tools before defining work. They automate partial workflows. They roll out AI broadly and hope adoption sticks.
The organizations that win do three things in order: understand the process, automate it end to end, and start with the people already equipped to use it well.
That sequence isn’t complicated. It’s just harder than buying a subscription and calling it a strategy.