Enterprise AI conversations often start with technology.
Which model should we use? Should we build agents? How should agents communicate? What should we build versus buy? How do we connect AI with APIs, data and enterprise applications?
These are important questions. But before designing an AI platform, there are more fundamental questions to answer:
Business Function — What is the organization responsible for doing?
Business Process — How does the work get done through a set of activities to achieve an outcome?
AI Features — Which AI capabilities should work horizontally across the organization, and which need to operate vertically within specific business processes?
Understanding these layers can significantly influence how we design an enterprise AI platform.
Start With the Business, Not the AI
Consider Digital Sales as a business function.
Its responsibility might include acquiring customers, selling products and growing customer relationships across digital and assisted channels.
Within that function, there are multiple business processes.
For example, a Sell a Product process might look like:
Understand customer need → identify suitable products → determine eligibility → construct an offer → determine pricing and discounts → create the order → confirm the outcome.
These activities may involve different applications, APIs, data sources, business rules and teams.
Technology will change. The business responsibility and the need to execute these processes will remain.
That’s why understanding the business function and its processes should come before deciding where AI fits.
Horizontal AI: Capabilities That Work Across the Business
Some AI features have value across many business functions and processes.
For example:
Search and knowledge retrieval. Summarization. Content generation. Document understanding. Translation. Conversational assistance.
A summarization capability could support Digital Sales, Customer Service, Finance, HR and many other functions.
These are horizontal AI features. Their value comes partly from being reusable across the enterprise.
An AI platform should make it easy for different business functions to consume these capabilities without rebuilding the same foundation repeatedly.
Vertical AI: Intelligence Within the Business Process
Other AI features require much deeper understanding of a particular business context and process.
Return to our Digital Sales example.
AI might determine the next best product based on customer needs and eligibility. It might help construct and optimize an offer based on pricing and discount rules. An agent might orchestrate several activities required to progress a customer from product discovery to a completed order.
These are much more vertical.
Their intelligence comes from understanding the context, rules, data and desired outcome of a particular business process.
And this distinction matters.
Simply creating a Product Agent, Pricing Agent, Order Agent and Notification Agent around existing technical services doesn’t necessarily mean we have created an intelligent business process.
We may have simply reproduced today’s technical architecture using agents.
Let the Business Shape the AI Platform
This leads to a simple architecture principle:
Don’t start by asking, “What agents should we build?” Start by understanding what the business does and how the work gets done.
From there, identify which AI features should be horizontal and reusable, and which need to be vertical and deeply integrated into a business process.
The architecture can then follow a clearer hierarchy:
Business Function
What are we responsible for doing?
↓
Business Process
How do we achieve the business outcome?
↓
AI Features
Where can AI assist, reason, decide, automate or orchestrate?
↓
AI Platform Architecture
What reusable models, agents, orchestration, context, data, APIs, security and observability are required?
This changes the architecture conversation.
Instead of designing an AI platform around the technology available today, we design it around business responsibilities, processes and outcomes – while creating reusable AI capabilities that can serve the enterprise horizontally and specialized AI capabilities that can operate vertically.
A future-ready AI platform therefore doesn’t begin with models or agents.
It begins with understanding the business.