The Recognition Test

Everybody in software says their product is hard to use. Nobody defines it, so nobody fixes it. I want to define it.

A screen is easy when the person looking at it recognizes everything on it as a fact about their own business.

Hard to use is a lazy diagnosis

Walk any trade show and every founder will tell you the market leader is hard to use and they are easy. Ask what "hard" means and you get adjectives. Clunky. Confusing. Overwhelming. You cannot fix an adjective.

So I started with our own platform. Building a complex automation in Vendasta today, or a custom report, is inaccessible to almost everyone. Only a handful of our most technical people can do it, and they say openly that they struggle. When one person in the company can use a tool, the tool is wrong.

Our mission is to make technology easy, accessible and affordable for small businesses. We owe ourselves a definition of easy we can measure.

The definition

When an owner looks at the screen, every question on it should be one only they can answer. What do you do? What does a good outcome look like? Here is what happened. Is it right? They are the world's leading expert on their business. There is nothing to learn before answering.

The moment the screen shows something they do not recognize (a node, a trigger payload, a webhook, a join) we have handed them homework. We have asked them to become a bit of a programmer before they are allowed to be a business owner.

Count the foreign nouns

Here is the measure. Take any screen. Write down every noun on it. Sort them into two columns.

| Their words | Our words | | :-- | :-- | | Customer, appointment, review, invoice, refund, lead, phone call, business hours | Account ID, trigger, webhook, variable, UTM, boolean, condition, dimension, canvas, step |

Every noun in the right column is a translation the user must do before they can say what they already know. Zero means the screen speaks their language. Forty-two steps on a canvas means it speaks ours.

I call this the Recognition Test. A screen passes when an owner thinks, "I can do this, because everything here is something I already know." It fails the moment they feel they need to call someone technical.

Any team can use it tomorrow. You need a list of nouns and an honest sort. No usability lab required.

Capture, compile, correct

The things an owner recognizes are the things that make up the business logic: the job to be done, what good looks like, examples and corrections. Nobody at Vendasta knows what a good outcome is for a plumber in Tulsa. The plumber does. Our job is to make it effortless to put that knowledge somewhere and have AI do everything else.

| Layer | What the user sees | Recognition Test | | :-- | :-- | :-- | | Capture | Questions about their business in plain English | Passes. Their words only. | | Compile | Nothing by default. AI generates tested, validated software. Experts can open it. | Exempt. | | Correct | The outputs: the email that went out, the ticket that was routed, the report. They mark it right or wrong. | Passes. Judging outcomes is what owners do all day. |

The complexity that used to sit on the owner's screen moves into generated software, where a real toolchain checks it continuously.

Easy at the top, deep underneath

The fair objection: our software solves hard problems. Multi-location franchises. Tiered support routing. Reconciling revenue across billing systems. It cannot be plain English all the way down. Correct. It should not try.

Think of a car. You drive with a wheel, two pedals and a fuel gauge. Every driver recognizes all of it. Pop the hood and there is an engine a mechanic can reach. The engine does not make the car hard to drive, because you are never forced under the hood to get groceries. Most business software does the opposite. It makes every driver pass a mechanic's exam before turning the key.

So the standard has four parts.

1. The top level passes the test. An owner can capture their logic, run their business and correct the outputs without meeting a foreign noun.
2. The depths exist, and they are excellent. When a real edge case shows up, an expert (ours, a partner's or an AI agent) can open the actual code, tests and evals.
3. Going deep is a choice, never a toll. If an everyday job requires opening the hood, the top level failed. That is a bug.
4. What happens below surfaces above in plain words. "Calls after 6pm go to voicemail with a callback promise." A diff tells the owner nothing.

The complexity gets a proper home, with a door on it.

How this scales

Expert-built logic scales with experts. If two people can build complex automations, the company's capacity to encode business logic is two people's hours. Add partners and their customers and the queue grows forever. There are not enough experts in the world, and they take years to make.

Run the same maths through the Recognition Test. Every screen that passes multiplies who can contribute, from two people to every partner and eventually every business. AI compiles the logic and never gets tired. Every correction flows back into the logic, so the system learns at the speed of usage.

AI does the work. Humans orchestrate. The Recognition Test is how we know the human can orchestrate, because everything on their screen is theirs.

If they recognize it, they can do it. If they can do it, we scale.