Key points

• A 2025 study published in the Quarterly Journal of Economics analyzed 5,172 customer support agents at a software company. After the company introduced an AI assistant tool, the number of customer issues resolved per hour rose by about 15% on average.
• A 2026 study published in Management Science ran a randomized experiment with 640 small business owners in Kenya over about five months. Some were given a GPT-4-based business advice tool, but the study could not statistically confirm an average improvement in sales or profits.
• Even as AI gets more capable, the business owner still has to decide what to hand off and how that work connects to customers and sales.
• In the end, a business delivers value to customers, gets paid for it, and keeps that work going, so value has to come before money.

Working with AI, I can try more things on my own. I create content, draft replies to customers, and build features for the service. With the output right in front of me, it feels as if the business has grown just as much.

But the fact that things are piling up in front of me is separate from whether the business is doing well. For something to really be a business, customers need a reason to use it, and someone has to pay for it. You have to be able to keep doing the work with the money you receive. That doesn’t change in the AI era either. In the end, a business delivers value to customers, gets paid for it, and keeps that work going.

A business doing well vs. doing a lot of work, fast

AI really does help a lot with work. A 2025 study published in the Quarterly Journal of Economics analyzed 5,172 customer support agents at a software company. After the company introduced an AI assistant tool, the number of customer issues resolved per hour rose by about 15% on average. Solving more customers’ problems in the same amount of time can reduce wait times and the burden on staff.

Research that looks at the business as a whole shows a different side, too. A 2026 study published in Management Science ran a randomized experiment with 640 small business owners in Kenya over about five months. Some were given a GPT-4-based business advice tool, but the study could not statistically confirm an average improvement in sales or profits.

Doing an already-defined customer support job better is different from applying AI’s advice to your own business. Even as AI gets more capable, the business owner still has to decide what to hand off and how that work connects to customers and sales. It’s genuinely welcome that AI can create marketing content, but that content only has real value if it solves a problem for the customer who sees it or leads them to take some action. If you make 100 pieces of content a day and they don’t contribute to sales, what’s the point?

It turns out not every cost AI cuts becomes my profit

When AI lowers the cost of translation and content production, there’s room to improve profits. You can also use AI to handle customer consultations and save on labor costs. Adding more channels for receiving inquiries doesn’t add much of a burden. But competitors use similar tools, too.

So if competitors get similar efficiency from AI, the core of competition now becomes what service we offer that carries value only we can provide. Getting paid isn’t enough to call it a good business. In the end, a business delivers value to customers, gets paid for it, and keeps that work going, so value has to come before money.