For years, I built Scale Insights with one major flaw.
Some people tried it. Hated it. Left.
Then a few months later some of them came back, because the thing they actually wanted to do couldn't be done anywhere else.
And they painfully grinded through the learning curve while cursing us.
For four months.
Then it clicked. And our raging users became our raving fans, and told everyone about us.
The irony?
The thing that made them rage in the first place, we couldn't solve. Not for years.
We poured everything into the side we were already good at instead. More signals. More parameters. More control. More ways to say exactly how you want your PPC to behave.
Also, more things to rage about.
11 algorithms, and one of them alone has over 100 parameters.
The madness.
And I'm still adding more. SQP. Inventory. Ranking data. Hourly data.
That's always been my philosophy though. I never wanted to build software that told you there's one correct way to run PPC. You know your brand better than we ever will. So here are the building blocks. Now go build your own thing. My job was to make sure whatever you built could go bigger and hit harder.
And for people who know exactly what they're doing, it works.
Ivan, my co-founder, has scaled past $100m a year running his own strategies through it. Millions of automated actions a month, across hundreds of thousands of campaigns, in multiple regions.
Agencies run completely different playbooks for every client, on the same system.
That's absurd control. You decide something should happen and it happens. Precisely, every time, on schedule, no drama.
That's why people came back after cursing at us.
But what about everyone else? Everyone who didn't already know what they wanted to automate?
For years our answer was: go watch our strategy series, where big sellers and agencies break down exactly what worked in their accounts. Reach out to them if one sounds like a fit. And lately, let our customer success team go through your account with you.
Every one of those answers was a person.
The intelligence was always human. Ours, theirs, somebody else's. Never the software's.
We got really good at automating almost anything. And stayed abysmal at telling you what's actually worth doing.
The obvious fix is what some are already racing to build right now. An AI that looks at your account, decides, and makes the changes for you.
I think that's a trap. The autopilot trap. An expensive one to walk into.
I'll come back to why.
But first you need to see the problem it's actually supposed to solve.
Our algorithms can watch your account and act on it millions of times a month, watching things you'd never keep track of yourself. But you have to tell them what to look for.
What they can't do is come to you first.
They won't tap you on the shoulder and say: you're leaving five figures on this product. Demand's there, you're converting above the market, your rank's been stuck for months, and you can still afford the traffic.
They won't tell you the change you made three weeks ago quietly made things worse.
They won't tell you which product to cut, which one to fix, and which one deserves a lot more money.
And the hardest one? You don't know what you don't know. It's sitting in your account right now, and you never thought to ask.
Try this one. Did your ad spend build organic rank you keep, or did you just rent traffic and hand it back?
And before you even spend: is the prize big enough to be worth the climb? How far below the real demand are you actually sitting? Can you afford the fight long enough to hold the position once you win it?
A sharp seller can work all that out. For one keyword. Once.
Now do it across hundreds of search terms, on every product, every week, without getting sloppy.
At some point that stops being about discipline. There are just too many connections for one person to hold. That was as true of me as anyone.
And here's the part that took me a while to see.
Every dashboard we've ever built started with a guess. We guessed what you'd want to know, then built a screen for it. Decent guesses. But still ours. If your question wasn't on the list, too bad.
Then AI came along, and suddenly you can ask anything. Here's the part people miss though. A great question on top of raw data still gets you a confident wrong answer.
Why? Dump a hundred thousand rows into an LLM and it chokes. It runs out of room to think, starts adding numbers by vibes, and hands you a beautifully written conclusion that's off by 10x.
So we flipped it. All the heavy computation happens on our side first, exact and deterministic. What the AI gets is signals: what's off, what's trending, what your last change actually did, how your ads move your rank and your rank moves your sales. So it can do the one thing it's genuinely brilliant at. Interpreting.
You bring the question. We bring everything it needs to answer it really well. And sometimes, to tell you things you never thought to ask.
No more menu. Ask whatever you want.
So why not go all the way? Let it decide and make the changes too?
That's the autopilot trap. Two reasons.
First one's simple. AI works in probabilities. Ask it the same thing twice and you might get two different answers. That's fine when it's drafting an email. Not so cool when "push this product harder" means bumping every bid by 3 times at 2am, pushing top-of-search to 900%, then tapping itself on the shoulder and popping champagne for a job well done.
When real money moves, I want hard boundaries: how far a bid can go, under what conditions, and what the thing is never allowed to touch without me.
The second one's the real one.
If we built something that decided what you should do, we'd be breaking the one rule this company was built on. We've spent years giving you suggestions, strategies, starting points, and letting you make the call. Then AI shows up, and suddenly we're the ones making it for you?
No. So we opened it up instead.
That's what MCP is. AI can reach into everything we've built, whenever it needs to. The signals, the analysis, the automation. Everything it needs to work a decision out properly.
Then people started building things we hadn't planned for.
One guy in our beta told Claude what kind of automation he wanted. Plain English. Claude turned it into actual rules, pushed them into Scale Insights, and they ran. He made the calls that mattered: which products, what numbers he needed to hit. The rest just happened.
Nobody here built that for him.
A long-time Scale Insights user found tens of thousands of dollars sitting in gaps in an account they'd worked for years. Clean account. The kind you'd look at and figure there's nothing left to find.
Someone built her own dashboard as a Claude artifact. Her metrics, her layout, her questions. Not the screen we picked for her.
Someone else built a skill to see how his external traffic was moving his organic sales and rank.
None of that was on our roadmap.
Which is the whole point. We never wanted to dictate your advertising strategy. So why would we dictate your workflow, your dashboard, or which AI you work with?
You want a different answer? Ask. You want the whole thing working a completely different way? Build it.
Intelligence on top of absurd control.
For years we handed you the building blocks to automate your PPC your own way.
Now you can build the thinking on top of it too.
And that five figures hiding somewhere in your account? The one nothing ever tapped you on the shoulder about?
I've waited a long time to ship this. I can't wait to see what you do with it.
~Luke