Most AI spending goes wrong before anything is built. Not because the tool was bad, but because nobody stopped to ask what the money was supposed to buy.
Business owners ask me some version of the same question all the time: "Should we be doing something with AI?" The honest answer is almost always yes, but the useful answer is a set of questions. When a business can answer them clearly, the project that follows is usually small, cheap to try, and easy to judge. When it cannot, no amount of technology fixes that.
Here are the five I would answer on a single page before spending anything.
1. What does this job cost you today?
Start with a job, not with AI. One specific piece of work: answering the same questions on WhatsApp all day, writing quotes, chasing unpaid invoices, preparing the weekly report nobody enjoys.
Then describe what it costs right now, in plain terms:
- Who does it, and what else they would be doing instead.
- How many hours a week it takes, even as a rough estimate.
- What slips when that person is busy or away. Slow replies, forgotten follow-ups, mistakes on a Friday afternoon.
This is your "before". Without it you will never know whether the "after" was worth anything. It also has a useful side effect: sometimes writing it down shows the job is smaller than it felt, and the right move is to fix the process, not buy a tool.
2. What does a mistake cost?
Every AI system will be wrong sometimes. The question is not whether, it is how expensive a wrong answer is and how fast you can catch it.
A wrong opening hour sent to a customer is awkward. A wrong price, a wrong dosage, or a wrong answer about a contract is serious. Those are different projects, and they deserve different amounts of caution.
The best first projects live where mistakes are cheap and visible: a person can see the output before it matters, or the damage from an error is a quick apology. Leave the high-stakes work for later, when you understand how the system behaves and your team has learned to trust it with the easy things first.
A simple way to frame it: for this job, would you rather it said "I am not sure, let me get someone" ten times too often, or guessed once when it should not have? Your answer tells you how tightly it needs to be controlled.
3. Is the knowledge written down?
AI is very good at using knowledge. It is useless at inventing yours.
If the answers to your customers' questions live in one experienced person's head, the first project is not an assistant. It is getting that knowledge onto paper: the real questions, the real answers, the exceptions, the things you never promise. This is unglamorous work, and it is where most of the value is. A business that has written down how it works can use almost any tool well. A business that has not will be disappointed by all of them.
Look for three things:
- Answers to what customers and staff actually ask, in the words your best person uses.
- Rules for decisions: when you offer a discount, when you escalate, what counts as urgent.
- Examples of good work: the reply you are proud of, the proposal that won, the report the boss liked.
If these exist, you are further along than you think. If they do not, you have found your first step, and it costs nothing but time.
4. Buy, set up, or build?
Once the job is clear, there are roughly three ways to get there.
Buy a product that already does this job, used as it comes. Fastest to start, least effort, and often good enough for common needs like meeting notes or drafting.
Set up an existing platform properly: connect it to your inbox or your customer records, give it your written knowledge, define what it may and may not do, and design how it hands over to a person. This is where most small and mid-sized businesses get the best return. The tools exist; the work is in fitting them to how you operate.
Build something custom. Worth it when the job is central to how you make money, when nothing on the market fits, or when your data and process are unusual enough to be an advantage. It is also the slowest and needs someone to maintain it.
Many first projects jump straight to "build" because it sounds serious. Most of them should have been "set up". Ask honestly whether the job is special, or whether it only feels special because it is yours.
5. What would make you stop?
This is the question almost nobody answers, and it is the one that protects your budget.
Before you start, decide what failure looks like. For example: if after six weeks the team still does most of the work by hand, or customers complain more than before, or the owner of the project stops checking it, you pause and rethink.
Writing this down does two things. It stops a weak project from living forever because nobody wants to admit it did not work. And, oddly, it makes success more likely, because a team that knows how it will be judged tends to pay attention to the right things from the first week.
Pair it with the opposite: what would make you do more? If it handles most of the routine questions well and people use it without being reminded, that is the signal to widen its job.
Put it on one page
Here is the whole thing as a template you can fill in with your team in under an hour:
- The job: one sentence. What it costs today, in hours and in what slips.
- The risk: what a mistake costs, and who catches it before it matters.
- The knowledge: where the answers, rules and examples live, or who will write them down.
- The route: buy, set up or build, and why.
- The stop and go signals: what makes you pause, and what makes you expand.
If you can fill every line, you are ready to spend, and you will spend less than you feared. If you cannot, you have not lost anything. You have found out cheaply what needs to happen first.
Why this is worth an hour
The businesses I see getting real value from AI are not the ones with the biggest budgets or the newest tools. They are the ones that were clear about the job before they started. Clarity is cheap. Rework is not.
If you would like to go through your page with someone who has done this across different kinds of businesses, you can book a call at cal.michelbaakliny.com. Bring the page, even if half of it is blank. That half is usually the most useful part of the conversation.
