The Before You Buy It Checklist

---

Good Research Is Not The Same As A Good Decision

Every wrong purchase in this checklist came after research I would still defend. Sources were checked. Prices were verified against the company's own page. Nothing in the process was lazy.

The mistakes were all one level up. I was answering a question thoroughly without ever checking it was the right question. That failure leaves no trace, which is what makes it expensive. You pay, the tool works, and you never find out what the alternative was.

These six checks take about eighteen minutes between them. Run them in order. Most of the time they confirm what you were going to do anyway, and occasionally one of them deletes the decision entirely.

---

1. Write Down What You Want To Be True When This Is Finished

Do this before you read a single review, article or comparison. One or two sentences, in your own words, describing the situation you want to be in once the money is spent.

I once spent six hours researching a system properly. I checked a research report against its original sources, went through ten practitioner videos, ran two deep research passes and put eleven questions to forty-nine sources. Then I said out loud what I actually needed the thing to do, and almost none of the six hours survived that sentence.

The report had set the agenda the moment I opened it. Every hour after that was me getting better at answering a question I had never asked. Nothing in the process was going to catch it, because everything was measured against the report instead of against my situation.

2. Ask What You Qualify For Before You Ask What It Costs

Eligibility first, pricing second. Almost nobody does this in that order, because pricing pages are easy to find and eligibility rules are not.

I needed a piece of software and had two realistic options, the cheap one or roughly a hundred and fifty dollars more for the version that did everything I needed. Three research agents compared them for weeks and every price they quoted was correct. I accepted the trade-off.

Then someone suggested checking whether I qualified for anything first. Five of the products were free for my situation, including the expensive one I had reluctantly decided against. Same software, same account. The only difference was which web address you arrived from, and every search result sends you to the paid one.

Nothing in the research was wrong. The question was wrong, and no amount of rigour on the wrong question gets you to the right answer.

3. Count How Many Of Your Sources Are Actually Independent

Agreement between people who read the same original is not evidence. It is one opinion wearing several hats, and it feels exactly like consensus from the inside.

Fifteen sources once agreed with me and I stopped checking. The person who invented the approach, an engineering blog from a real company, two critics and ten build videos. All positive. It turned out to be one six month old idea being passed around.

So I changed what I asked for. Instead of information I asked for postmortems, abandonment stories and people writing about why they quit. Same tool, same topic, forty-nine sources came back instead of fifteen. Only two of the systems had any record of running past twelve months.

4. Find Where The Tool Hands The Job Back To A Person

Every tool has a point where it stops and you take over. The marketing describes the part before that point, because that part demos well.

I checked every AI product claiming it could handle a tax return. Not one of them can actually file it. There is no public route into the government's electronic filing system, so a person still logs into approved software and presses the button. Every product does the interesting part and stops there.

That gap is not an oversight and it will not close because a model gets better. It is a permissions problem, and permissions problems do not respond to intelligence.

5. Look Properly At What You Already Have

The setup you are embarrassed by is often the one that works. Simple does not look like a feature, so it is easy to keep trying to give it away.

I run eleven client projects on plain text files and assumed I was years behind. No database, no plugin, nothing running in the background. I asked two different research tools what I was missing and they came back with the same answer independently, which is the only kind of agreement worth much. Every purpose built option had logged problems with losing data. The plain files had none, mostly because there is so little there to break.

I had spent the whole exercise braced to learn I was behind. The finding was that I should stop reading and get back to work.

6. Judge It On How It Behaves, Not On How Well It Reads

Making sense and being right are two different tests, and most of us only run one of them.

I once approved something I had just built, then researched what it was built for, and found six holes in my own hour old work. Not opinions about style. Missing steps that would have quietly produced worse results every time it ran. The thing I approved had been written from what the model assumed, which is a polite way of saying written from nothing.

It made sense on the page, and that is the problem, because everything competent makes sense on the page.

---

The Whole Checklist, In Order

Eighteen minutes in total. Most of the time you will end up buying the thing you were going to buy anyway, and you will know why, which is worth something on its own when it turns out to be wrong.

---

---