I migrated a bunch of data to HubSpot this week.

You know what HubSpot did with it?

INVALID EMAIL.
INVALID EMAIL.
INVALID EMAIL.

Well, it didn't actually say that. It just rejected the whole damn thing.

All the data.

None of it went over.

Because of a handful of bad emails I didn't even collect.

And yeah, I know what you're thinking.

"Just clean them up."

Which would work if there weren’t 36,000+ records in this data set like the one a client handed me last week?

You know how you go to do something with a tool because it’s going to be a whole lot easier to do it like that than by hand?

And then it just falls flat on it’s face and now you are staring down this monumental task that you REALLY don’t want to do because it’s going to be a massive pain in the backside.

And getting it done is the job that has to get done to move things forward.

You know that place of being caught between a rock and a hard place?????

Yeah this is one of those moments where doing it by hand would be fine if it was a small data set that you could wrap your brain around.

But trying to do it by hand when there's 36,000 rows of emails… is like trying to push that boulder up the hill and never being able to get there, and watching it roll back down hoping it doesn’t crush you on the way.

Because the reality is, after the 65th row, you're not going to do it by hand

So then, now what? Ask Claude to figure it out?

Well, what about this?

Rough estimate: 36,000 rows at that column density is probably 15-20 million tokens.

Claude's context window is 200,000 tokens.

You can't even get 1% of that file into a single session.

So now, where are you?

The chances of getting this done quickly and easily are Slim and None, and Slim's leaving town.

I know some of my friends, software engineers, data engineers, and GTM engineers alike, would say, "Let's pull out Claude Code."

And that's fine if that's where you want to go and you know what you want to do.

So this file I'm talking about has 36,000+ rows…

Is all that data clean, or is it just the email that's bad?

How are you going to know?

What I'm getting at here is we never built a process that helped us understand what was going wrong…

Which is to say, we never built a process to understand our data quality in the first place.

Because we never built a process, we never turned it into a system.

Because we never turned it into a system, we never automated it.

So every time bad data shows up — and it always shows up — you're back at the beginning.

Solving the same problem again

Burning time on something you've done before, but didn't formalize in any concrete way.

And now everybody's excited about AI.

AI can build the system fast.

AI cannot build a system you never defined.

You still have to know what the process is before anyone — human or machine — can turn it into anything.

That's the part that didn't get easier.