There’s a simple question I think every company evaluating AI should ask:
Could an AI agent run your quote-to-order process today?
Not whether you have Salesforce, ServiceNow, a modern ERP, or an AI strategy. Not whether your vendors have released agents that can configure products, generate quotes, submit approvals, or create orders.
Could you actually give an agent responsibility for moving a real customer transaction through your business?
For many companies, the answer would be no.
And the reason probably has very little to do with AI.
Imagine giving the job to someone new
Think about what happens when a new employee joins your organization.
You can give them process documentation, training, system access, and instructions. But eventually someone tells them how things really work.
Use this price unless it’s this type of customer.
Check ERP before promising the delivery date.
That product technically requires engineering approval, but only under certain configurations.
Don’t trust that field in CRM.
If it’s an international order, send it to Finance first.
That customer has special terms. Ask Susan.
These aren’t unusual edge cases. In many businesses, they are the process.
Experienced employees learn how to navigate them. They develop context. They recognize patterns. They know who to call when the system doesn’t provide an answer.
Now replace that experienced employee with an AI agent.
Suddenly all those little gaps matter.
An agent can’t rely on “you just know”
Consider a fairly normal quote-to-order request:
Create a quote for 50 units of this configuration, apply the customer’s contracted pricing, confirm we can deliver by October 15, get whatever approvals are required, and create the order.
That sounds like exactly the kind of workflow an AI agent should eventually handle.
But look at what the agent actually needs to understand.
Which product combinations are valid? Which system owns the configuration logic? Where does contract pricing live? Is current inventory enough to satisfy the order? If not, can production meet the date? Does this discount require approval? Does the customer’s credit status affect the order? At what point can the transaction move from CRM into ERP?
None of those are AI questions.
They’re business architecture questions.
If the answers live across six applications, three spreadsheets, two inboxes, and the heads of a few longtime employees, adding an agent on top doesn’t eliminate the complexity.
It exposes it.
This changes how I think about AI readiness
Most AI-readiness conversations start with technology.
What model should we use? Which agent platform? What should we automate first? How do we connect our data?
Those questions matter, but I think there’s a more useful place to start:
Can your business explain itself?
Take one important process—quote-to-order is a great example—and try to define exactly what an agent would need to know to execute it.
Where does authoritative customer data live?
Where does product configuration happen?
Who owns pricing?
What makes a quote valid?
What triggers an approval?
What makes an order ready?
Which system has authority at each step?
What happens when something doesn’t follow the happy path?
If those answers are difficult for your organization to provide, you’ve discovered something valuable before deploying a single agent.
The goal isn’t one giant system
This also challenges an assumption that has shaped enterprise technology for years: that simplification means putting everything into one platform.
I don’t think that’s where this is heading.
CRM may remain the best place for customer and opportunity context. CPQ may own configuration and pricing. ERP may continue to own inventory, production, and financial transactions. Order management may orchestrate fulfillment.
That’s fine.
The important thing is that each system has a clear responsibility and that the rules governing movement between them are understandable.
In that environment, an AI agent can become the engagement layer across the business.
The salesperson doesn’t necessarily need to understand which API retrieved inventory or which system calculated the price. They need to communicate the desired outcome.
The agent figures out how to get there.
But only if we’ve done the hard work underneath.
Try the test
Pick a real transaction your team completed recently.
Then ask:
Could an AI agent have completed this from beginning to end without someone quietly filling in the gaps?
Every time the answer is “no, because someone would need to know…” you’ve probably identified an opportunity.
A missing business rule.
An unclear system boundary.
An undocumented exception.
A manual approval.
A piece of tribal knowledge.
A data problem.
An integration gap.
Those are the things I’d be working on now.
Because the companies that get the most value from agents won’t necessarily be the ones that buy the most AI.
They’ll be the ones whose businesses are easiest for AI to understand and operate.
Before asking what an AI agent can do for your quote-to-order process, ask whether your quote-to-order process is ready for an AI agent.
The point about AI exposing process gaps rather than eliminating them is important. In O2C, clear rules, reliable data, and defined handoffs are essential if automation is expected to handle real transactions effectively.