|
Well, hello, and welcome to this week’s edition of the eCommercer.
There’s a particular kind of joy in going back to something you thought you’d left behind ages ago. For me that’s coding. I wrote the first versions of all our sites twenty-odd years ago, then hired people far cleverer than me and I just let my skills go rusty. I assumed that door had closed. It turns out it hadn’t. AI has propped it wide open again, and I’ve spent a good chunk of this year happily building things I’ve no business building.
So this week’s guest was always going to be a good one, because he’s been living in exactly the same place, only deeper.
An old friend, a lot of emails, and $50,000 of learning
Drew Morgans runs Marknology, a full-service Amazon and marketplace agency out of Kansas City. He’s a returning guest, and he’s also a mate. I went to see him and his family a couple of years back, on the way home from dragging my daughter round the Goonies house in Oregon, and there’s a barbecue place near his office I still think about.
Drew has spent around $50,000 and the best part of a year going deep on AI. Not the hype version. The actual build. He started by feeding fifteen years of his own work into it, roughly 500,000 of his own emails, 300 podcast episodes, blogs, SOPs, and won and lost proposals. The idea was simple enough, which is to turn all that past work into something useful in the present. As he put it, the real meat’s in the emails.
The problem everyone hits is context
Then he ran into the wall the rest of us keep hitting, which is context. AI tends to work sequentially. It looks over here, then over there, then somewhere else, and it never quite holds the whole picture at once the way a good friend would. So Drew’s clever assistant kept handing him answers with holes in them.
His fix is the interesting part. Rather than a cleverer prompt, he built a rules layer he calls Drew IQ, around 350 “if this, then this” operator rules that run before the AI ever gets involved. On top of that sits a context engine that surfaces everything about a given brand, the margin, the inventory, the ranking, the stock position, in about half a second, locally, before the expensive model is ever called. He sums the whole philosophy up in a line I rather liked.
| |
“Cheap model in, expensive model out.”
Drew Morgans
|
Cheap local models do the grunt work of ingesting and tagging overnight. The good, pricey model only gets involved at the sharp end, when something is client-facing or, as he puts it, “trying to be me”. He also made a decision to keep the whole system model-agnostic. He’s named it Winston, and if his AI provider vanished tomorrow he could switch to another and keep running.
Scaling the work you could never scale
Drew had an interesting idea called, ‘scaling the unscalable’. Any good Amazon operator knows the deep work that actually moves the needle, cross-referencing rankings against margin against inventory against what real people are saying in the reviews and on Reddit, all to work out why a listing gets the clicks but not the sales. It’s brilliant work. It’s also so slow and fiddly that agencies simply can’t afford to do it at scale. Drew’s using AI to do that non-scalable work far more often, and handing clients insights they’ve genuinely never had before.
I know exactly why Drew’s enjoying this, because I’m doing the same thing on my side of the Atlantic. We’re cancelling software subscriptions left, right and centre, because we’ve built our own. Our own project management. Our own SAAS, bespoke to what we actually need. And I’ve been having far too much fun with it. I grew up in the eighties, so naturally I built KITT from Knight Rider as a voice interface for Claude on my phone. Big microphone button, a mock KITT voice answering back, little dials going up and down. Utterly daft, and genuinely useful when I’m driving.
The bigger point Drew and I agree on is that for years the job was to master the next skill, and then the next. Now the job is to look up and see where the whole thing is heading. It’s made both of us feel like the early days again, leading from the front rather than managing from behind. I didn’t see that coming, and I’m rather enjoying it.
If you’re a founder wondering where AI actually pays off, rather than just where it’s loud, this is a very practical episode. We get into one super-agent versus a crowd of specialist bots, why he stopped chasing every shiny new tool, and what he reckons happens to the rest of us over the next year.
|
