Your Next Customer Might Be an AI: Here's What to Do About It

What small businesses need to fix before the next wave of AI shopping.

I've spent the better part of the last two years helping small and mid-size businesses in and around The Woodlands figure out what to do about AI. A recurring problem in those conversations is that “we need AI” can mean several very different things.

Sometimes an owner wants help answering customer messages. Sometimes they want their systems to share information. Sometimes they just want a task to happen without having to remind someone every day.

Those aren't all AI problems. Understanding the difference matters more than choosing a tool.

At the same time, AI is changing how people find products and, increasingly, how they can buy them. I think small businesses should pay attention to that change. But the preparation is less about buying something new and more about getting their existing information and processes in order.

Small businesses are person-based, not position-based

This is one of the biggest differences I see between the way software is sold and the way a small business actually runs.

Software tends to assume defined roles and repeatable processes. A small business often depends on a person who handles several jobs and knows which exceptions to make.

Think of a shop where the same person handles returns, purchasing, and the phone. The written return policy might be simple. The actual decisions depend on things that person knows: what was promised to a regular customer, which supplier has been sending damaged goods, or when the owner is willing to make an exception.

That knowledge is part of how the business works. It just isn't necessarily recorded anywhere.

Adding AI doesn't automatically make that knowledge available. A tool can read the policy you give it, but it may not know the exceptions, the history, or who has the authority to decide. Even a capable system needs access to the right context and a clear point at which to hand the decision back to someone.

That's why I would start by documenting one process. Not a huge operations manual. One page describing what normally happens, what the common exceptions are, and who decides when things aren't clear. AI can help write that page, too, but the people doing the work have to supply and check the details.

A lot of requests for AI are really requests for automation

“When an order ships, text the customer.”

“When inventory drops below a set level, send an alert.”

These are rules. Ordinary automation can handle them, often with less cost and less uncertainty than an AI system.

Interpreting a messy customer email is a different problem. So is turning product specifications into a readable description or identifying patterns in sales history. Those are reasonable places to test AI.

The distinction isn't absolute. A useful workflow can combine both: AI interprets the message, a rule routes it, and a person approves the response. But separating the parts helps you understand what you're buying and where it might fail.

I also see owners group the website, payment terminal, spreadsheet, and chatbot together as “the IT/AI stuff.” That's understandable when you're busy running a business. But these tools solve different problems. Sometimes the missing piece is an integration between systems you already have, not a new AI subscription.

Where I would start using it

For a small retailer, I would look for a narrow task where the output is easy to review and the benefit is easy to measure.

Product listings are a good candidate. Give the AI accurate specifications and ask it to draft titles, descriptions, and frequently asked questions. Then check every claim. It should help explain the product, not invent materials, dimensions, or benefits to make the copy sound better.

Customer support is another. Start with sorting messages and drafting replies from your actual policies. Keep a person responsible for approvals, particularly for refunds, complaints, and unusual requests. Measure the time saved after review, not just how quickly the draft appears.

Inventory forecasting is worth considering when the data is ready. But if sales records, stock counts, and purchasing information disagree, fix that first. A more sophisticated forecast won't resolve uncertainty about what you currently have on the shelf.

I wouldn't start with the most impressive demo. I'd start with a task that takes time every week and ask whether the tool makes the whole job easier, including checking and correcting its work.

Customers have another way to find you

The change outside your business is that shoppers can ask an assistant to do some of the research they used to do themselves.

Instead of opening several tabs, someone might ask, “Find me a durable rain jacket under $120 that can arrive by Friday.” The assistant can compare options and suggest a shortlist. Your store might enter that conversation before the shopper has ever seen your homepage.

There is evidence that this channel is growing. Adobe reported that traffic from generative-AI tools to U.S. retail sites grew about 693% year over year during the November–December 2025 holiday season. That is a growth rate, not AI's share of all retail traffic. It doesn't mean traditional search has been replaced, or that your own store will see the same pattern.

For me, it's a reason to test how your business appears in AI-assisted shopping, not a reason to abandon everything else that brings customers in.

Ask the kinds of questions your customers would ask. See which products get recommended and what information accompanies them. Treat the results as a spot check rather than a definitive ranking; answers can vary. Then compare what the assistant says with what's actually on your site.

Your product data is becoming another part of your storefront. Clear titles, dimensions, materials, prices, availability, shipping terms, and return policies help both people and software evaluate what you sell. Good photography and a credible brand still matter. They just don't replace the basic facts.

“Premium construction” tells a shopper less than a specific, accurate description of how the product is made. And a recommendation isn't much use if the price changes or the item turns out to be unavailable at checkout.

Helping someone buy isn't the same as buying for them

It's useful to separate three things that often get grouped together as “agentic commerce”:

• An assistant helps someone research and compare products.

• An assistant helps complete checkout, with the person confirming the purchase.

• An assistant chooses and buys within limits the person has authorized, without approval for each individual order.

Those require different levels of trust and control. Evidence that AI influenced a sale is not evidence that an autonomous agent placed the order.

The checkout infrastructure is real. In September 2025, OpenAI announced Instant Checkout and the Agentic Commerce Protocol, developed with Stripe. The launch supported purchases from U.S. Etsy sellers inside ChatGPT. Importantly, the announced flow required users to confirm their order and payment details. It wasn't an assistant independently deciding to spend their money.

My expectation is that more delegated purchasing will develop unevenly. Reordering a familiar household item seems a more likely starting point than choosing an expensive or personal purchase. That's my outlook, not a guarantee about how quickly customers will adopt it.

For a business owner, the useful questions are practical: Can your system provide accurate availability? Can it accept and track the order? Who handles a mistake? What requires human approval? Ask your platform which capabilities are available to your business now, rather than assuming an announcement means you already have them.

What I would do this week

1. Write down one process that lives in someone's head. Include the exceptions and who approves them. This is useful whether you adopt AI or not.

2. Separate rules from tasks that need interpretation. Automate the predictable steps first. Identify where AI might assist and where a person should stay involved.

3. Check five product listings. Correct missing specifications, unclear shipping terms, and inconsistent prices or stock information across your sales channels.

4. Test how customers can find you. Try ordinary search and an AI assistant using realistic shopping questions. Note missing or incorrect information rather than expecting one query to tell the whole story.

5. Run one small, measurable experiment. For example, test whether AI reduces the time needed to prepare listings without introducing errors. Count review time and subscription costs. Decide in advance what would make it worth keeping.

I think AI will become a normal part of commerce. But “using AI” isn't a business result. Fewer mistakes, less repetitive work, better information, and customers finding the right product are results.

That's where I would start. Make one part of the business work better, measure what changed, and let that tell you what to do next.

Sources

• Adobe: AI-driven traffic surges across industries, January 12, 2026. Retail referral growth during the 2025 holiday season. This is Adobe's own analytics reporting, not a prediction for an individual store.

• OpenAI: Instant Checkout and the Agentic Commerce Protocol, September 29, 2025. Primary announcement documenting the launch and its user-confirmed checkout flow; not a statement of current merchant coverage.