Could you have built this in 2019?

Here is a test that takes about a minute and tends to sting a wee bit. Look at whatever you have automated with AI. Now ask: could a competent developer have built this in 2019?

If the answer is yes, you have not changed how your business works. You have bolted a faster engine onto the same process. I picked this framing up from Allie K Miller’s newsletter and I have not been able to unsee it since — including when looking at our own work.

What fails the test

• A chatbot answering your ten most common questions. Decision trees did that a decade ago, just more stiffly.
• An auto-responder acknowledging an enquiry. That is a mail rule with better manners.
• Generating a first draft of something you then rewrite entirely. Faster typing is not a new process.

None of that is wasted. Let me be plain, because I do not want this read as a telling-off: automating a rules-based job is often exactly the right first move, and it is where most businesses should start. It is a ceiling, not a mistake. But it explains the flat feeling a lot of owners describe — the sense that they have adopted AI and nothing much changed. Nothing much changed because nothing much changed.

What passes it

The work that genuinely could not have been done before tends to share one feature: it involves reading unstructured, messy human language at volume, and forming a judgement about it. Rules cannot do that. Now it can be done every night while you sleep.

• Reading every enquiry from the last quarter and telling you which phrasing on your site is producing the wrong sort of lead.
• Reading two years of reviews across four platforms and naming the complaint that keeps recurring in different words.
• Turning a rambling five-minute voice note from a site visit into a structured quote, with the assumptions listed.
• Watching your own quotes and telling you which ones convert, and what the winners have in common.

Notice these hand you a decision, not a task. That is the tell. A 2019 process gives you output to process. A post-2019 process gives you something to decide.

Run it on your own setup

Firstly, list every place AI touches your business today — honestly, including the ad-hoc chatbot use nobody wrote down. Secondly, put a yes or a no beside each one. Thirdly, if it is all yeses, that is not a failure. It is a map of where the next gain is, and it is usually a bigger gain than the ones you have already taken.

One honest admission, since I would rather credit the test than hide behind it: plenty of what we build for clients passes only partly. A first-response system is mostly a 2019 process with better language. The bit that is genuinely new is what it notices while it works. Our free audit scores exactly this — how much of your operation is still manual, and which of it is now worth handing over.