Your Next Customer Might Send an AI Agent
Most businesses are used to customers arriving through a website, a phone call, an email, or the front door.
There’s another possibility worth preparing for: a customer sends an AI assistant to do some of the work for them.
“Find a supplier who has these parts in stock.”
“Check whether they can deliver by Thursday.”
“Get a quote for 200 units.”
The customer still has the need and the money. They’ve simply delegated part of the buying process.
For a business owner, that raises a practical question: how easily could an assistant get accurate information from your business and take the next step?
You don’t need to rebuild your company around AI to answer that. You do need to think about what information you make available, how reliable it is, and what you’re prepared to let someone do with it.
Start with the customer’s task
Imagine a customer looking for a particular product.
Your website lists it, but availability requires a phone call. Pricing is in a downloadable catalogue. Delivery depends on which warehouse has stock, and someone needs to check.
A person can work through that process. So can your staff. They probably do it every day.
But each handoff adds time and uncertainty.
Now imagine the customer’s assistant can ask your systems three straightforward questions: Do you have it? What does it cost? When could it arrive?
If your business can provide dependable answers, you’ve made buying easier. You’ve also reduced the routine enquiries your staff need to handle.
That’s a useful business outcome on its own.
Where MCP fits
One way to make those connections is through Model Context Protocol, usually shortened to MCP.
The name sounds more complicated than the basic idea. MCP provides a standard way for AI applications to access selected information and functions in another system.
For a business, that might mean checking stock, retrieving an order’s status, or preparing a quote.
Think of it as a service counter for software. You decide which services are available, what information is required, and who is allowed to use them.
It can sit in front of systems you already operate. Those systems still hold the records and enforce the business rules. MCP provides a way for an AI application to work with the capabilities you choose to expose.
There is still integration work involved. Your information still needs to be accurate. And access still needs to be controlled.
A standard connection helps. It doesn’t do the management work for you.
Checking stock is different from promising delivery
This is where the business decisions become more important than the technology.
You might be comfortable letting an assistant check whether you have ten units in stock.
Would you let it reserve all ten?
For how long?
Could it offer a discount? Extend credit? Promise Friday delivery?
Those actions have consequences. They affect other customers, your margins, your inventory, and your reputation.
We already understand this when we hire people. A new employee may be allowed to look up an order without being authorized to issue a refund. A salesperson may prepare a quote but need approval for a larger discount.
AI access needs the same clarity.
You should be able to explain what an assistant can see, what it can do, and when it must ask for approval. Those limits need to be enforced by your systems, with a record of consequential actions.
Start with one useful question
There’s no need to make every department, database, and workflow available at once.
Choose one recurring question that costs your staff time and has a dependable answer.
“Where is my order?” is a reasonable candidate.
The first version might let an authorized customer retrieve the latest status and expected delivery date. No address changes. No cancellations. No refunds.
Then check whether it works in practice.
Are the answers accurate? Can customers access only their own orders? Does the service handle missing information sensibly? Does it reduce work, or simply create a different kind of support request?
That gives you evidence for the next decision.
It may also expose an older problem. Perhaps delivery dates aren’t updated consistently, or stock figures mean different things in different systems.
Those issues deserve attention. Giving an assistant faster access to unreliable information will make them more visible.
Make it easier to do business with you
I’d be cautious of anyone suggesting that adding an MCP server makes an organization “AI-ready”.
Readiness involves the quality of your information, the clarity of your processes, and the limits of the authority you delegate. A connection is one part of that.
But the opportunity is practical.
Customers want answers. Employees want fewer repetitive tasks. Businesses want enquiries to turn into orders without unnecessary effort.
Start there.
Pick one useful interaction. Make it accurate, give it clear boundaries, and measure whether it helps.
You don’t need to predict exactly how customers will use AI. You can begin by making your business easier to work with when they do.
You can explore a small example on my own website: connect your AI assistant to search my published writing and follow the original sources.