Hi there,
Over the past two months, nine clients told us the same thing, unprompted: fix the data first, then start on agents.
That's the default advice in the industry right now, and there are numbers behind it.
|
THE NUMBER EVERYONE QUOTES
Gartner predicts that by 2026, 60% of AI projects will be abandoned because the data never reached “AI ready.” 63% of organizations either don't have a proper data management approach, or aren't sure if they do. Gartner, February 2025. Survey of 248 data management leaders. |
So the conclusion looks obvious: data first, agents second. It sounds right, and it is exactly where agent projects get stuck.
Ask the same firm the obvious follow-up question, how do we make all of our data AI ready, and their published answer is this: “There is no way to make data broadly, universally AI ready in advance.”
Which means “is our data ready” has no answer. Ready for what?
| Data is never ready in general. It's only ever ready for one specific job. |
So start from the other end. Pick one job you want an agent to do, list the information that job needs, then find out where each piece of it actually lives.
|
THE PLAYBOOK
Pick one decision. Write down six things.
The rule: if you can't write that fifth line, the problem was never the data. And line six, who gets the final call when the rule doesn't settle it cleanly, is the hardest part of this and the easiest one to skip.
|
Every piece will land in one of three places, and each one points to a different next step.
|
01
|
In a system
It's already there.
More common than people expect. The information already exists, it's just never been connected to this job.
Start today
|
|
02
|
In a person · most likely
In someone's head, their email, or a spreadsheet only they understand.
Gartner's own numbers say 80 to 90% of the data companies generate is unstructured. This is normal, not a failure on anyone's part.
A couple of hours, then ready to start
|
|
03
|
Nowhere · rarest
It doesn't exist anywhere.
A real gap, and in our experience the rarest of the three. But now you know exactly what to collect, instead of everything.
Collect the right thing
|
All three assume the number you find is trustworthy. If you can't tell whether it's accurate, that's a different problem.
Because you were always going to need that conversation with them. The agent just moves it up the calendar.
| ||
|
Panopto Workplace Knowledge and Productivity Report
|
You're paying that today, with or without an agent.
Our own lead-routing agent stalled on exactly this. There was no rule for which rep gets which lead, because one person had always made that call from memory. Writing it down took an hour. Now the team can see it, the next hire inherits it, and the agent runs on it.
This hour is where the whole exercise works or falls apart. Three things that help.
|
✓ Ask for real cases, not the official process. The tidy process written down on paper is usually not what anyone actually follows. ✓ Ask what they check when a number looks wrong. This part never gets written down anywhere. To them, it isn't a rule, it's just experience. ✓ Read your notes back to them. Every correction they make is a piece of knowledge you just captured. |
You don't need the whole data foundation before you start. Building one agent is what forces the rule out of somebody's head and onto paper. Do that three or four times, and the foundation you end up with is made entirely of things you actually needed.
|
The worksheet
One Decision, One Page
The six lines above on a single printable page. Fill it in for one real decision, take it into the hour with the person who makes it, and you'll leave knowing which of the three answers you have, and whether you can start now.
Download the worksheet →
|