What to expect in your first 90 days of using AI

Nobody tells you that adopting AI has a rhythm. There is a burst of enthusiasm, a dip when it does not behave as hoped, and then, if you keep going, a steady climb to something genuinely useful.

Knowing that pattern in advance stops you giving up at exactly the wrong moment.

Days 1 to 30: try, learn, pick a winner

What to do

• Choose one or two tasks that take real time each week, such as inbox sorting, writing quotes or typing up notes. Our guide to choosing your first AI project helps here.
• Write down how long those tasks take today. You will need this later.
• Give two or three keen people access and let them experiment.

What to expect

Early results will impress you, then a few poor answers will disappoint you. That is normal. Most disappointing answers come from vague instructions or missing information, and both are easy to fix.

Days 31 to 60: make it reliable

What to do

• Turn the instructions that worked into saved templates everyone can use.
• Give the AI your own material: price lists, policies, past examples of good work.
• Agree simple rules about what must always be checked by a person.

What to expect

This is where the time savings start to become consistent. It is also when people who were sceptical begin to ask for access, usually because they have seen a colleague leave on time.

Days 61 to 90: measure and extend

What to do

• Compare the time those tasks take now with your notes from the first week.
• Ask the team what still annoys them. That is your next project.
• Decide whether anything deserves a proper tool, such as a connection to your CRM or a customer facing assistant.

What to expect

Most small businesses we work with find several hours a week saved per person by this point, though results vary a great deal by business. More importantly, the team now understands what AI is good for, which makes every future decision easier.

The common stumbles

• Trying to do everything at once. One task done well beats ten done badly.
• Not measuring the starting point. Without it you cannot prove the value.
• Leaving it to one enthusiast. If only one person uses it, it disappears when they are busy.
• Skipping the ground rules. A short note on what data can and cannot be shared avoids worry later.

How to know it is working

Ask three questions at the ninety day mark. Are we spending less time on the tasks we chose? Is the quality the same or better? Would the team complain if we took it away? Three yeses means you have something worth building on.

If you would like a hand planning your first ninety days, our free AI adviser will give you a 30 day starting plan, or you can talk to us.

Frequently asked questions

Do we need special software to start? No. A business plan for ChatGPT or Claude is enough for most first projects.

What if it does not work for us? Then you have learned that cheaply. Usually the issue is the task chosen, not AI itself.