Chapter 7: Bottle the Expert
At the Düsseldorf boat show, Hallberg-Rassy had a drone pilot on staff. Full time. The footage was gorgeous: white hulls, blue water, sails perfectly trimmed.
But that's not what I remember. What I remember is Jonas, their head of sales, walking visitors through the HR 370. Oak or mahogany interior, and why owners in different climates choose differently. The settee versus the easy chairs, depending on how you'll actually live aboard. Why the code zero sail suits one sailing style and the self-tacking jib another. Why the deck drains exit under the waterline: a small thing you'd never think to ask about, offered at exactly the right moment.
It was a perfect configuration session: needs translated into choices, constraints explained before they were hit, trade-offs narrated in the customer's language. And it existed in exactly one place: inside Jonas.
That's the problem with expertise in complex product companies, and no amount of drone footage fixes it. A video can make a buyer want the boat. Only bottled expertise lets every salesperson sell it correctly, every time. The problem isn't pretty assets. It's trapped expertise.
The Johan problem
Every manufacturer has a version of this sentence, and I've heard it verbatim: "We can't change that rule today. Only Johan knows how it works."
Johan is brilliant. Johan is also a single point of failure with a pension date. Every retirement is a data-loss event, and unlike a server crash, there's no backup running. Sales tenure makes it worse: when your reps average a few years in the role, competence doesn't compound, because it keeps leaving. The bottleneck in scaling complex sales isn't talent. It's memory.
The companies that solve this treat product knowledge the way they treat software: explicit, versioned, testable, owned. And in practice the bottling has two distinct layers, which is the part most teams miss.
The first layer is rules: the hard constraints that make a machine valid. "The sleeper cab only works with the high-horsepower motor." These become the constraint model from Chapter 3, and they must be deterministic, because they guard physics, safety, and certification.
The second layer is narrative, and it's the newly valuable one. For every module and variant, a short structured story: what it's for, when to choose it, when to avoid it, what it implies downstream. A hundred words per variant. "The sleeper cab suits routes above 600 kilometers per day with overnight stops; skip it for regional distribution; it requires the high-HP motor and adds curb weight." Rules keep you safe. Narrative makes you fast. And narrative is precisely what a language model needs to speak usefully about your products, because the LLM knows what a truck is. It just doesn't know your trucks.
How do you start? Not with a form. Put your best expert in front of the product for ninety minutes with a camera or a good microphone, a colleague asking buyer questions, and no slides. Jonas walking the boat, captured. That transcript is your first logic spec: the rules hiding in it go to the constraint model, the stories go to the narrative layer, and the thresholds ("above 600 kilometers a day") get made explicit instead of instinctive.
Done once, properly, the asset outlives everything around it. I know one ruleset that has survived three UI redesigns and two CRM migrations, untouched, still correct, still selling. The interfaces were fashion. The knowledge was structure.
Anti-pattern: The Pretty Asset Trap. Investing in ever-better product content (photography, video, 3D, brochures) while the decision logic stays in experts' heads. The buyer gets excited and then meets a salesperson who can't configure what the video promised. Bottle the expert before you buy another camera.
With products modular, prices governed, and knowledge bottled, the structure is in place. The rest of this book is about the payoff, and it starts with a question that sounds reckless: what if you let people outside the company sell your machines?
Every retirement is a data-loss event. Put your best expert inside the system, not just in the video.