What the customer is actually paying for

The invoice says one thing. The money is usually for something else. Before you can score a business, you have to know what its customers are buying when they hand over cash, and there are only three honest answers: bytes, outcomes, and accountability.

Three kinds of deliverable

Bytes. The customer pays for a piece of information or art: an article, a logo, a transcript, a resume, a set of floor plans, a tax summary, a jingle. Bytes are the machine's home turf. Producing acceptable bytes is precisely what these systems are built to do, and every improvement in the models lands directly on the deliverable. A bytes business is not automatically doomed; elite bytes, bytes with a name attached, bytes tied to a relationship, can deflect. But the default exposure of a bytes business is high, and the burden of proof sits on the "why would they still need me" side of the ledger.

Outcomes. The customer pays for a changed state of the physical world: a clean kitchen, a mowed lawn, a working refrigerator, a full dumpster where a garage full of junk used to be. No model mops a floor. Outcomes also carry a second layer of protection, because producing them usually requires presence, and presence builds the local trust that machines cannot fake. The junk removal course in this catalog, rated 2, exists in a world where customers could photograph their garage and get an instant AI quote. That automates the estimate, pile two. The truck, the stairs, and the dead washing machine that has to come down them remain stubbornly human.

Accountability. The customer pays for a human being who is answerable. This is the strongest deflection of the three, and it hides inside thousands of businesses. The notary loan signing course, rated 2, is the purest example in the catalog: a notary's signature is not information, it is legal personhood witnessing an act. A state commission, a bond, a journal, a body in the room. The machine can explain what a loan document means, and explaining it is useful. It cannot be liable for the signing. Medicare billing, rated 4, shows the same spine in a data-heavy trade: claims scrubbing is automatable, but a practice needs someone accountable to HIPAA, to payer audits, and to the money actually arriving.

The strip-the-bundle exercise

Most services sell a bundle of all three, which is why unexamined "will AI kill X" arguments go in circles. Strip the bundle. Take a residential cleaning visit and lay out what is actually being purchased: labor in a home (outcome), someone trusted with a key (accountability), and a checklist report with photos (bytes). Two of those three deflect. Now take "social media management": posts and graphics (bytes), a content calendar (bytes), and an owner-facing human who explains what the numbers mean and resets expectations when a post flops (accountability). The bundle is roughly half bytes, and the social media marketing agency course scores 6 in our catalog, the highest band we still teach.

The strip tells you where the price lives, too. If customers mainly pay for the bytes layer of your bundle, expect the price of that layer to fall toward zero, because the supply of acceptable bytes just became infinite. If they mainly pay for the outcome or the accountability layer, the bytes layer automating is margin, not mortal danger.

The question that cuts through

Here is the test I use when I cannot decide. Ask: if a machine produced the bytes layer of this service perfectly, instantly, and free, would the customer still need me?

The cleaning customer would. The machine does not bring a vacuum. The loan signer would. The signature is the product. The Mediocre-but-acceptable blog post customer would not, and neither would the basic logo customer or the product description customer. And the interesting middle cases answer in the middle: the medical practice would still need a biller but would need fewer hours of one; the small business would still want a marketing person but only the one who can actually be blamed and hugged.

That "still need me, but fewer hours" answer is the signature of the middle bands, the fours and fives, and it defines the adaptation problem you would be signing up for: run the same business with the machine doing half the work, and win the clients who want the human on the hook.

A warning about fake accountability

One abuse of this framework deserves its own paragraph. A lot of businesses claim an accountability layer they do not really have. "Consultants are accountable for results" is often false; check the contract. If your customer can cancel anytime, bears the consequences of bad output themselves, and cannot recover damages from you, the accountability you are counting on as your moat may be decorative. The deflection is only real if the customer experiences it: a license the state can strip, insurance that actually pays, a relationship the customer cannot get elsewhere, or a physical result only your hands produce.

Bytes, outcomes, accountability. Classify the deliverable, strip the bundle, and you have the first two of the five scoring questions. Blame comes later in the course: when it goes wrong, who catches it.