Start with the task, not the tool

The useful question is not what AI can do. It is a much smaller question: what in my week is repetitive, mostly text, and does not require my judgment to get to a first draft?

Write that list before you look at a single product. It usually takes ten minutes and it is almost always longer than people expect. Proposals that follow a pattern. Follow-up emails. Job postings. Meeting notes turned into action items. The monthly report that says the same thing with different numbers. Marketing copy nobody has time to write.

That list is your buying criteria. Anything that does not touch it is somebody else’s priority.

Four categories that consistently earn their cost

These hold up across industries and are where almost every organization sees the first real return.

• First drafts of anything written. The largest and least glamorous win. Proposals, emails, job postings, policies, newsletters, and social copy. The value is not that the output is finished. It is that starting from something is dramatically faster than starting from nothing, and most of the time cost in writing is the blank page.
• Summarizing and extracting. Long documents into the three things that matter. A recorded meeting into decisions and owners. Fifty survey responses into themes. This is reliable, low-risk work because you can verify the output against the source in seconds.
• Reformatting and cleanup. Messy notes into a structured document. A list in the wrong format into the right one. Inconsistent data made consistent. Unglamorous, genuinely tedious, and something a computer should have been doing for twenty years.
• Thinking out loud with a competent second opinion. Not a decision-maker, but a useful place to pressure-test a plan, generate options you had not considered, or ask what you might be missing. This is the use that surprises skeptics most.

What is not worth it yet, for most small businesses

Being clear about this matters more than the list above, because this is where money gets wasted.

• Anything requiring a real integration project. If a tool needs to connect to your systems and your data before it does anything useful, the cost is not the subscription. It is the six months and the person managing it. Large organizations can absorb that. Most cannot.
• Anything sold as replacing judgment. Tools that promise to decide who to hire, what to bid, or which customer to prioritize are asking you to outsource the part of the job you are actually paid for. Use them to prepare a decision, never to make one.
• Anything where being wrong is expensive and hard to catch. These tools produce confident, plausible output that is sometimes incorrect. That is fine for a draft email and genuinely dangerous for a regulatory filing, a legal document, or a specification, unless somebody qualified checks every line.
• The eleventh subscription. Tool sprawl is the most common failure. Most organizations get nearly all their value from one or two general tools used well, rather than nine specialized ones used twice.

How to judge a tool in a week

Ignore the demo. Demos are built to work.

The one skill that matters

Most disappointing results come from asking vaguely and expecting the tool to fill in what you did not say. Three habits fix the majority of it.

Say who it is for and what it is for

Not “write a job posting.” Instead: a job posting for a diesel technician in a rural market where the biggest competitor is the dealership, aimed at somebody with five years of experience who is tired of being treated like a number. Context is the whole difference.

Give it an example of your own

Paste in one you wrote that worked and ask for something in that voice. This single move does more for output quality than any clever phrasing, because generic writing is what you get when you supply no reference for your own.

Treat the first output as a draft, then push back

Say what is wrong with it. Too long, too formal, does not sound like us, missing the part about the schedule. The second and third versions are where the useful work happens, and people who quit after one attempt conclude the tool does not work.

Three rules for a small organization with no IT department

• Do not paste in anything you would not email to a stranger. Customer records, employee information, anything confidential or covered by an agreement. Decide this explicitly and tell your team, because in the absence of a rule people will assume it is fine.
• You own the output. If it goes out under your name, on your letterhead, to your customer, it is yours. That means somebody reads it before it goes, every time.
• Check anything that looks like a fact. Numbers, dates, names, citations, regulations. These tools are excellent at language and unreliable about specifics, and the confident tone is what makes that dangerous.

What to actually do with the hours

The point of any of this is capacity. An organization that gets four hours a week back and puts all four into more email has not gained anything.

Decide in advance where the time goes. More conversations with customers. The strategic work that keeps getting postponed. Walking your operation. Developing somebody on your team. Leaving at a reasonable hour. Whichever it is, name it before you start, or the week will absorb it without anybody noticing.

Related reading: the comfort ceiling, on what tends to happen with capacity that nobody assigned a purpose.

Where this lands by industry

The task list changes. The approach does not.