AI for commercial leaders: 13 concepts worth understanding.
I’ve been spending a lot of time talking to commercial leaders about AI, and something has become increasingly apparent. We’re having sophisticated conversations about transforming entire businesses with technology that many decision-makers don’t fundamentally understand.
I’m not suggesting executives need to become engineers. But understanding a few of the mechanics behind AI, what drives its economics and how it connects to business systems, is becoming pretty important when making investment decisions.
So I’ve pulled together 13 concepts worth understanding. Some are admittedly geeky, but they help explain what it actually takes to move AI from an impressive demonstration to a useful business capability.
The intelligence and economics
1. Tokens
Tokens are the units of information AI models process. They can represent words, parts of words or other information, and are a major driver of usage costs.
At enterprise scale, the volume of information processed across millions of interactions can have a significant impact on operating economics.
2. Inference
Inference is what happens when a trained AI model is put to work. Every time it answers a question or performs a task, it’s running inference.
Training builds the model. Inference is the ongoing activity businesses pay for when using it.
3. Reasoning Models
Reasoning models devote additional computation to working through complex problems, evaluating alternatives and completing tasks that require multiple steps.
More reasoning can improve results, but it can also increase cost and processing time. You probably don’t need advanced reasoning to classify routine customer inquiries, but it may be valuable when evaluating complex commercial scenarios.
Connecting AI to business knowledge and systems
4. Embeddings
Embeddings help AI systems identify relationships between information based on meaning rather than simply matching keywords.
For example, they can help connect a question about customers leaving with documents discussing churn or retention. They’re one of the building blocks that make large collections of company knowledge searchable and useful.
5. RAG (Retrieval-Augmented Generation)
RAG allows AI to retrieve relevant information from external sources and use it when generating a response.
Think of a sales assistant accessing current pricing, product information or company policies rather than relying solely on what the model learned during training.
6. Tool Calling
Tool calling allows AI to request actions from external software.
Rather than simply explaining how to update a customer record, an AI application could retrieve the record and submit an authorized change to a CRM. It’s one of the capabilities that moves AI from answering questions to performing work.
7. MCP (Model Context Protocol)
MCP is an open standard that provides a consistent way for AI applications to connect with compatible tools and information sources.
Think of it as a standardized interface between AI and business systems, reducing the need to build every connection from scratch. Security and permissions still need to be managed.
Moving from AI assistance to execution
8. AI Agents and Agent Orchestration
An AI agent can pursue an objective by selecting actions, using tools and completing multiple steps with some degree of autonomy.
Rather than simply summarizing a sales pipeline, an agent might identify opportunities requiring attention, prepare recommendations and initiate approved follow-up activities.
Agent orchestration coordinates multiple agents, tools and workflows to complete more complex activities.
The commercial opportunity extends beyond automating individual tasks. It’s about rethinking how work moves across the organization.
9. Agent Harness
An agent harness is the infrastructure surrounding an AI model that helps it execute work reliably. Depending on the application, it manages instructions, tools, context, memory, errors and approvals.
Think of the model as the intelligence and the harness as the operating environment.
Two applications can use exactly the same AI model and deliver very different business results because of how the surrounding system is designed.
10. Human in the Loop (HITL)
Human in the Loop refers to designing AI workflows where people review, approve or intervene in certain decisions or actions.
An AI sales agent might prepare standard quotes independently but require management approval before offering discounts beyond a defined threshold.
As AI takes on more operational responsibility, deciding where humans retain authority becomes an increasingly important part of organizational design.
How AI is changing discovery and advertising
11. AEO / GEO (Answer Engine and Generative Engine Optimization)
Traditional SEO helps businesses appear in search results. AEO and GEO focus on how brands, products and information appear in AI-generated answers.
AI search systems may retrieve information from websites and other sources, evaluate its relevance and use it to construct a response. Clear product information, accessible content, credible sources and consistent brand information can improve the opportunity to be discovered and referenced.
For commercial leaders, the shift is significant. Customers may increasingly form preferences and narrow their choices within an AI conversation, before ever visiting a company’s website.
12. AI Advertising
AI platforms are also introducing advertising into conversational search and discovery experiences.
Unlike traditional keyword advertising, these placements can be informed by the context and intent of a conversation. A customer researching a new product category, for example, might encounter a relevant sponsored offering while exploring their options.
The important distinction is between organic AI-generated recommendations and clearly identified paid advertising. Commercial leaders will need to understand both, including how visibility is earned, how media is purchased and how effectiveness is measured.
The commercial implications
13. Agentic Commerce
Agentic commerce describes AI agents participating in product discovery, evaluation and purchasing, potentially completing transactions with appropriate authorization.
A customer might ask an AI assistant to compare products, evaluate options against their preferences and complete a purchase.
For commercial leaders, this raises an important question: What happens to brand preference, customer relationships and distribution when an AI agent increasingly influences which products customers consider and buy?
How the pieces fit together
Understanding these concepts individually is useful, but the real value comes from understanding how they work together.
Think of an AI-powered sales assistant that researches accounts, prepares recommendations and supports customer interactions. Here’s a simplified view of the components that could make it work, and the layers where AI is changing how customers discover and buy.
| Component | Simple analogy | Role in the business |
|---|---|---|
| AI model | The brain | Provides the underlying intelligence and reasoning capabilities |
| Tokens and inference | The cost of thinking | Drive much of the ongoing cost of using AI |
| Embeddings and RAG | The knowledge library | Help AI find and use relevant company information |
| Tool calling and MCP | The connections | Allow AI applications to interact with business systems |
| Agent harness | The operating environment | Supports reliable execution, tools and error handling |
| Agent orchestration | The coordination | Connects multiple agents and workflows |
| Human in the Loop | The oversight | Establishes where people retain approval and accountability |
| AEO/GEO | The discoverability layer | Helps brands and products be found and represented in AI-generated answers |
| AI advertising | The paid visibility layer | Creates opportunities to reach customers during AI-assisted discovery and evaluation |
| Agentic commerce | The transaction layer | Enables AI agents to evaluate, select and potentially purchase products on behalf of customers |
This is a simplified analogy, rather than a literal technical architecture. Reasoning models are a type of model capability.
Understanding the technology is only the beginning
I think one of the biggest misconceptions about enterprise AI is that choosing the right model is the most important decision.
The model matters, but it’s only one component of the system required to create business value.
The harder challenge is connecting AI to the right information, integrating it into business processes, establishing appropriate decision rights and ensuring the economics make sense.
Commercial leaders don’t need to engineer these systems. But they need enough understanding to challenge assumptions, evaluate opportunities and recognize when an impressive AI demonstration is still a long way from becoming a viable business solution.
As AI becomes increasingly capable, the competitive advantage won’t simply come from access to better models. It will come from how effectively organizations integrate those capabilities into the way their businesses operate.
In our opinion, starting with process, people and workflows to deliver against customer and business needs is the first step. Technology is an enabler and should follow. Most, however, are starting the other way around, which often leads to increased cost, complexity, and quite frankly, issues.
