Where to start with AI when everything looks possible

Q: How do I choose my first AI use case?

Start from your own friction, not from a tool list: something slow, something expensive, something that breaks every week, or something people already avoid doing. Shortlist only AI use cases that address one of those, check data and process readiness honestly, then pick one pilot and define the metric it must move.

Q: How many AI pilots should a company run at once?

One is the right number for a first pilot. Many things can be done; pick only what is important and start with one. A single pilot with a named metric and owner teaches more than five parallel experiments that share attention.

Q: What is the most expensive AI mistake?

Picking the wrong problem in week one and discovering it in month six. It costs nothing upfront, which is exactly why it is common. The protection is boring: a metric, a baseline, a target, a date, and an owner, agreed before any tool is chosen.

"Everything is about AI nowadays. The question is when to start and in which cases." A business operator said that to me, and it is the whole problem in two sentences.

By Ekaterina Servetnik, 12lane.ai

The hard part of AI in 2026 has moved. Everyone can build now. The advantage is knowing what to build. AI multiplies your options, but it does not decide what will sell, what will save money, or what your customers will actually use.

So the honest starting question is "which problem are we allowing AI to solve". Here is the method I use with B2B companies.

Step 1: Start from friction, not from tools

Walk through your own operation and look for four signals:

Something slow. Quotes that take days. Proposals that take a week.
Something expensive. Smaller customers who cost more to serve than they bring in.
Something that breaks every week. The report everyone rebuilds by hand. The handoff that drops orders.
Something people avoid. The CRM nobody updates. The knowledge base nobody reads.

Any of these four is a better starting point than any tool announcement, because each one already has money attached.

Step 2: Check readiness honestly

Every realistic AI timeline assumes three things: clean data, a written process, and an owner on your side. If one is missing, the timeline is fiction. That is fine, as long as you know it. Getting ready is itself a worthwhile first project, and much cheaper than discovering the gap in month four.

Step 3: Define the metric before the pilot

This pilot moves [metric] from [current] to [target] by [date], with [owner].

One sentence. If your team cannot fill in the brackets, the pilot is not ready to start. The most expensive AI mistake costs nothing upfront: picking the wrong problem in week one and discovering it in month six. The bracket sentence is the protection.

Step 4: Pick one

Many things can be done. Pick only what is important, and start with one. A single pilot with a named metric teaches your organization how AI creates value in your specific context. Five parallel experiments teach your organization that AI is chaotic.

When the first pilot moves its metric, the second use case is easier to choose, easier to fund, and easier to staff. Momentum is a strategy.

Frequently asked questions

How do I choose my first AI use case?

Start from friction: slow, expensive, breaking, or avoided. Shortlist the use cases that address one of those, check readiness, define the metric, pick one.

How many pilots should we run at once?

One. Five parallel experiments share attention and produce five half-answers.

What is the most expensive AI mistake?

The wrong problem, chosen in week one, discovered in month six. It costs nothing upfront, which is why it is so common.

Browse 80 real cases to shortlist from

The AI Use Case Library documents 80 implementations with tools, timelines, and honest numbers, filterable by function, tier, and industry. Leave your email to open it.

Open the full library Book a 30-min call