Published September 2, 2026 · Altvina Insights · 5 min read

The One Question to Ask Before Your Next AI Purchase

Before the next tool goes on the card: which recurring task does it replace, when did a person last do that task by hand, and what process does it plug into? Three checks, under a minute, runnable on your own business today.

A pattern we keep seeing in smaller firms: the sign-up for a new tool happens late in the day, on a peer recommendation, with the billing details already saved from last time.

Annual costs less than monthly. The demo was genuinely good. Signing up takes less time than reading the terms, and whether it changed anything is a question for later.

Judgment rather than measurement, and offered as such: the waste gets decided in the minute before the sign up, not during it.

One question, and the three checks inside it

Which recurring task, by name, does this tool replace?

Sounds obvious. In practice the honest answer tends to come back as a category. Reporting. Client comms. Admin. Categories cannot be replaced. Tasks can.

That question splits into three checks, all of which can be run standing up, before anything goes on the card.

1. A task has a person, a trigger and an output

Say the answer out loud and listen for those three things. Something with the shape of "every Friday, whoever is on duty pulls the week's job sheets into a summary the owners read on Monday" counts as a task. Wording invented here for illustration, but that is the shape.

Stop if the best available answer is a department name. Whatever gets bought will end up serving an idea, and ideas have no handovers, no deadlines, and no person who notices when they break.

2. The last time a person actually did it by hand

Scheduled does not count. Neither does "we really should be doing that". Only the last time somebody sat down and did the thing.

Recent, with real friction, is the good answer: live work gives software something to grab onto. Blank looks are the more interesting answer, and worth more than the purchase decision itself, because either the task is not needed or it is quietly being skipped and somebody downstream is absorbing the consequences.

Automating a task nobody performs mostly makes an unperformed task look handled.

3. The process the tool plugs into

Does a written sequence exist, even a scrappy one, saying who does the task, in what order, and who receives the output? Or does the whole thing live inside one person's head and habits?

Head-only means the software will not inherit a process. It sits next to one. The person carries on working their way, plus the new job of keeping the tool fed.

What a thin answer costs

Tools bought against a vague task rarely fail loudly, going by what we have seen. They fail by becoming a place where work waits.

Cancelling one is hard, because nobody can prove it is not helping. It renews. Whoever set it up defends it in a meeting. The onboarding email stays unread, the seat stays assigned to whoever signed up, and eventually the line item survives because everybody assumes somebody else depends on it.

A weak answer does not stop the purchase from happening. It predicts where the gap turns up next. Painful becomes invisible.

Failed the gate? Do the task by hand first

Deliberately, one named person, steps written down as they happen. Rough notes on a phone are fine, and a documentation project would be overkill.

What comes out the other end is the task, the person, the trigger, the output, and the two or three points where the thing actually snags. Which is the shortest useful brief anybody could hand a vendor.

Going by what we have seen, it is also how some purchases quietly stop being necessary. The snag turns out to be a missing handover or an unclear owner, and neither of those is sold as a product.

Real task, done recently, written down somewhere: buy the thing with a clear conscience. Onboarding goes faster when "working" already has a definition.

Why one question beats a scorecard

Purchases do not usually go wrong from a shortage of checklists. They go wrong at the pricing page, with the card out and the trial about to lapse.

A longer framework would look more rigorous and get opened less. One question that fits in someone's head while a salesperson is being persuasive is the version that actually gets used.

Run it on the next purchase. Then run it backwards, which is the uncomfortable version: take the most expensive tools already being paid for and ask the same things. Which task, by name. Last time a person did it by hand. Which process it plugs into. Our hunch, offered as a hunch and nothing firmer, is that one of them comes back thin on every count.

Thin answers right across the stack

One weak answer is a purchasing decision. Weak answers across most of the stack point somewhere else: the work was never redesigned around the tools, so the tools got layered on top of the work, and the subscription list is now a record of that.

Before spending anything on that, there is a free way to find out which kind of problem you have. Our Routing Hub diagnostic asks where you are still structurally embedded in the work, and sorts what it finds into three piles: a workflow problem, a hiring problem, or a knowledge-transfer problem. It runs in the browser, takes a few minutes, and gives you the answer on the page. A fair number of people read that answer, see which pile they are in, and go and fix it themselves. That is a good outcome and it costs you nothing.

The Altvina Blueprint is built for that situation. A fixed-scope, fixed-fee engagement that hands over a Bottleneck Diagnosis, an Operating Roadmap, a Decision Framework, an Expert Deployment Brief, and a Recommended Path Forward laying out every option, including doing nothing.

Unsure whether that fits? Book a fit call: short, qualification, and not free consulting.

Content and Accuracy Disclaimer

This article was drafted with AI assistance and reviewed by the Altvina team. We rigorously fact-check all content to ensure reliability.

Should you notice any inaccuracies or outdated information, please contact us so we can correct it. Your feedback helps us maintain high standards of accuracy and transparency.