How to Tell the Future 🔮

*Taps microphone* Is this thing on?

Hey! It’s Madelyn. Long-time subscriber, now part-time sender. Weird!

I feel… passionately… about Attribution. A lot of people are talking about how attribution’s dead. I find that weird.

Here's my whole deal, in case we haven't met. I do marketing, and I do HubSpot, and the part I actually love is where those two overlap. Not campaign strategy on its own, and not portal admin on its own. The place where somebody has to decide what a property is for before anyone can run a report on it.

Attribution is that overlap in its purest form. It's a marketing question you can only answer with a data model.

So when people say attribution is dead, I think what they mean is they built one and it didn't tell them anything. Which, honestly, fair. But I still think you can tell the future with it. Just not in the way anybody wants you to.

Attribution will never tell you what a buyer is about to do. It tells you what they did, and only if you decided ahead of time what the possible answers could be.

Which is not the same as knowing. You're not guessing what someone will do. You're deciding in advance what the possible answers are, and building somewhere for each one to land. If you guess right, great. If you guess wrong, you still learn something, because the wrong answers landed somewhere too.

That's where most attribution projects die, honestly. Someone shows up expecting a machine that predicts, finds out it doesn't do that, and quits before they get to the part that works.

The part that works is dumber than a prediction. You don't know what a buyer is going to do. But if you build some buckets, they have to land in one of them. And if they don't land in any, that's a bucket too.

So here's how I'd do it.

Start from the question, not the property.

Write down what you want to be able to say before you build anything. "Which untrackable things are bringing us high-intent leads" and "which channel closes fastest" are different sentences, and they need completely different buckets. Nearly every stalled attribution build I've seen started in the tool instead of in the sentence.

Four steps, on paper, before you even open HubSpot. Takes about twenty minutes. (Ok, sometimes it takes longer if you get reaaally into it.) It's the only part nobody wants to do.

Then use the fewest buckets that will answer it. Sourced and influenced is two buckets. Two! And for a lot of companies that fully answers "is marketing contributing to revenue," which is the actual question being asked when there's a CFO in the room. You can get more precise later. Attribution is never done, and that's fine. But complexity is where these die, and not because it's hard to build. Because nobody maintains a model with forty categories.

The bucket nothing falls into is the reading.

In one of Hubsessed’s Attribution Unlocked sessions, Em Wingrove explained how she has a workflow where anything her AI can't confidently categorize drops into an "other campaigns" bucket, and landing there automatically creates a task for her to go look at it. The miss is the alert.

The same thing happens at much bigger scale with direct traffic. People will tell you all day that they found you through ChatGPT, and their contact record will say direct traffic, because HubSpot's AI referral source only counts when someone actually clicks a citation. Being the answer is different from being cited with an answer.

A growing pile of direct traffic isn't missing data, it's a reading. It's only readable if you built somewhere to put it, and knew to go look.

One last check before you build a bucket. Say out loud what you'd do differently if it filled up. If you can't answer that, you don't need the bucket. You need a better question.

Speaking of things you can actually forecast…

More to come soon. You’ll hear it here first!

That's the whole forecast. Who knows what you’ll do next. 🔮