AI Strategy
AI strategy from the foundations up
Strategy in the context of this article is based on former Dean of the Rotman School of Management Roger Martin’s definition of strategy as “where to play and how to win”. If you are expecting a plan or a roadmap from this article I recommend you invest ten minutes to watch his excellent you tube video, A Plan is not a Strategy.
The model
This is a bottom-up model, meaning the more foundational aspects are at the base. So rather than simply diving into AI Agents because they are powerful, or because everyone else is, take the time to set the foundations. From the foundation up we have:
AI Mandate and Guardrails
This is the most foundational layer, the base of your strategy. A few examples of what this layer can encompass:
- Your company position on AI, for example are you taking a wait and see approach or looking to actively exploit it?
- Guardrails for the use of AI. What boundaries do your people work within?
- Formal structure, e.g. is there a Chief AI Officer, who do they report to and so on
AI Research and Discovery
AI is a fast moving field. Simply putting a mandate and guardrails in place and calling this ‘job done’ just won’t cut it in today’s environment. You need to understand how AI is changing your industry, regulatory trends and the like. If you are looking to actively exploit AI you are up for some research. Companies using AI successfully today are experimenting their way to success. In tern this research will guide your mandate and guardrails
Knowledge Base
Knowledge without action is useless, and action without knowledge is foolish.
If AI Agents (next section) are all about action, then simply implementing AI Agents without giving them any knowledge about your company, its people, processes and expectations is a recipe for confusion. Storing knowledge in the heads of key people was never ideal, in the AI era this practice may simply become too expensive.
A knowledge base, in the full sense of the word can encompass:
- Company information, the kind of thing you find on company intranets currently, people, contact details and so on
- Processes. If you expect your AI Agents to ‘behave themselves’ and not go rogue, you need to have documented processes for them to follow
- Protocols and Specifications. This is an under-rated area. Large Language Models (LLMs) are imprecise by their very nature. Specifications help to bridge the gap between flexibility in communication and precision in execution. See this medium article for a in-depth explanation of this concept from a technical perspective
- Company databases. Your accounting system, your CRM system and so on are all rich sources of information that can be mined to give your people an advantage
- Company communications. The constant flow of emails, chat messages and even verbal communications are in some ways your richest source of information as they are real and current. Of course privacy is a massive factor in deciding if and how you should plug conversations into your company’s AI searchable knowledge repository
AI Agents
With Agentic AI being so new, some of the terminology being used is becoming quite confusing. My definition of an AI Agent:
An AI agent is an LLM-powered system that combines natural language understanding with access to a knowledge base and/or external integrations (e.g., APIs, tools, services) to perform tasks, retrieve information, and make decisions on behalf of a user.
Put simply an AI Agent knows stuff and it can do stuff.
AI today is predominantly used to find information. So you may ask an AI chatbot something like “find me the cheapest flight to London in a months time”. And the agent will dutifully sift through a bunch of online airline information, will ask you some questions around how many people are travelling, the exact date and so on and some up with flight information. An AI Agent will let you go a step further and allow you to say “Book it!”. And your AI Agent will communicate with the airline’s systems and book your ticket. The attractiveness of having this kind of convenience at your fingertips is not hard to see.
One interface to rule them all
In my long experience in software, the most enduring complaint I have heard from users is:
“why do we have so many software systems I have to log in to?”
— every user ever
AI Agents, once established, promise to give users the thing they have been asking for for so long — a single screen to log into and do all their work from. Your agent will “log in” to these different software systems and work with them on your behalf. Just like … well… an agent.
AI Enabled products and services
Putting AI in your product and/or service offerings may seem a million miles away for your company right now. Certainly building an AI enabled SaaS offering is not for everyone and their are some very big players in this space.
Consider this though, how many of your people right now are using ChatGPT to word and format their emails? AI is already part of what you offer in this sense.
Also, your business has customers. If you offer a digital channel of any kind, it is going to be impacted by the rise of AI. If your competitors are offering AI friendly on-routes to their digital offerings the timing of AI adoption may not be fully in your control.
This tier is the riskiest — an AI blunder in one of your service offerings can be embarrassing and really damage your credibility, hence why this area is at the top of our model, not something to be jumping into before you have laid the foundations.
Where to start
Where to start with all this is of course a decision only you and your organization can make. I’m predicting that because they will be just so darn useful, AI Agents will have a massive uptake when the technology does start to come together in a year or so. If you take only one thing from this article, take the time to build foundations of governance, knowledge and research before you buy that one way ticket on the AI Agent train.
