The Unglamorous Work AI Depends On

Borrowed

"For tacit knowledge to become explicit knowledge—that is, stored somewhere where it can be viewed, reviewed, and used by others—it must first be converted from the mind to another medium (a database, white paper, report, manual, video, podcast, picture, etc)."

From "Implementing Value Pricing" by Ronald J. Baker [Book]

Learned

Something I’ve been thinking about this week is how much of what a company knows never gets written down, and how much that limits your possibilities with AI today.

We’re working with some brands right now to build AI tools and help them understand what AI means for their businesses. Some of it is search and discovery, some of it is guided selling, and some of it is getting a single catalog to feed every channel properly. Every one of those conversations lands in the same place, and it isn’t the AI. It’s the data.

Take something simple, like whether a product is gluten-free. It’s listed on the packaging, and probably in the middle of the product description, but it isn’t a field. Nowhere in the catalog is there an attribute called “gluten-free.”

A shopper catches it because they can look at the package. A model reading the brand’s product catalog only sees fields, so if the attribute isn’t there, the product likely won’t appear in the answer. Multiply that by everything a shopper actually asks about, like serving count or what age it’s for or whether it holds up on sensitive skin, and you end up with brands that might be the perfect fit for someone who will never encounter them this way.

This is as true for consumer brands as it is for any business, and I’ve been running into our own version of it at Barrel. So much of what we know lives in our people rather than in documentation. If I’m being honest, I’ve had a love-hate relationship with documentation for years. It takes a while to put together, and at the rate our industry moves, it goes out of date almost immediately. What I’ve come around to is that the useful version isn’t a rulebook: capture data in real time, write down the evergreen processes, and leave the rest alone. We’ve done a decent job on the data side, mostly for new business and agency finances. From a process standpoint, though, a lot of those pieces are still missing. We’re actively adding new ways to capture data, such as a project’s expected cost rate.

The challenge shows up the moment you want to evolve something or learn from an outcome. What’s crazy is you don’t know how much you’re missing until you have the data to learn from.

What’s interesting is how much of this work is about looking back. For our quarterly session last week, I asked our growth team to go through the first half of the year and record every proposal, every partner lead, and every marketing dollar spent, along with their own read on why each one went the way it did. All of that already existed across different spreadsheets, systems, and inboxes, just never structured this way. Once it sat together, we could see things like which types of proposals actually won, and how often a promising partner lead died from nothing more than a missed follow-up.

It also showed us what we’d never thought to capture. We’ve changed how we name and break out our SOWs, so that the type of work and the service line are their own fields rather than something you’d have to read a document to figure out. That’s our version of the gluten-free attribute. The information was always there. It just wasn’t anywhere we could easily query it.

None of this work is fun. What we’re really talking about is structured data. You decide which attributes matter, agree on what the values mean, and fill in the gaps, one product or one process at a time. Nobody can do that part for you, and it’s the part that makes everything after it possible.

One thing that will separate people and businesses going forward has less to do with the tools and more to do with the discipline to start capturing and the willingness to go back and shape what’s already there, which is slower and less interesting than most of us would like.

Thought

What do we know that we've never had a reason to write down?