How to Build an AI Program in 2024

At the heart of most organization's to-do list is to build a well thought out AI program for their company agenda. How do you take ground in something so critical, while not over-investing? In this article I'll walk through the six steps that every organization needs to consider and the relationship between them for building a comprehensive strategy at the end.

Step 1: Executive Alignment

The successful organizations start with understanding the Mission of their Business and carefully considering how AI will impact the way that mission is executed in the world. This is not simply a list-of-ideas, but a comprehensive consideration of the threats (to current business) and opportunities (to future business) that exist or will be created in the market as a result of AI. The company needs to consider the mission in the world (such as providing Financial Results to Clients) and then how AI will impact the delivery of that mission (hint: it's a LOT). The list of considerations should include:

1. Does the Executive Team understand the art-of-the-possible?
2. Does the rest of the business understand the art-of-the-possible?
3. What is the mission of our business?
4. What efficiencies will AI provide in our existing business?
5. What would our business from the ground-up look like?
6. If we needed to re-create the business, what would it look like?
7. How do we enable AI as a skill across all employees?
8. What responsible guard rails do we want on our execution of AI?
9. Who are the critical business leaders to include in discussion?

Step 2: Group Envisioning Sessions

For the company that is just getting started with AI, you may start this at the Executive Level. For a company that is more mature, you might be moving through envisioning sessions cyclically. The goal here is to identify opportunities for AI in two lanes, commodity and mission-driven. The commodity opportunities will create value as a normative way of doing business. The mission-driven opportunities will be specific to your business, GTM, industry, or disruption. Both are necessary. The intent is to widen the tent and include members of the business to have a clear understanding of where your business can achieve wins from AI.

You might be looking at priorities such as the following:

Step 3: Scenario Selection

From this step you are very intentionally selecting opportunities for improvement. These opportunities are organized to achieve success and ranked based on priority to the organization. Note that this is where the lanes will diverge and commodity will move down its own path and achieving success will have its own set of directed goals. The opportunities you success need the following key requirements:

1. They must have a business sponsor with authority
2. They must have the necessary data
3. They must have ROI

As these scenarios are evaluated, either in commodity (think Copilot) or Mission Driven (think automating Customer Service), all of these need to be satisfied. You don't want to fall into the trap of just turning it on and hoping for the best. As you evaluate ideas, put them into a list like the following, enabling you and the executive team to select from the best ideas to consider. Remember you'll have both Incremental ideas (ones that are more efficient processes for doing what you do now) and Disruptive ideas (ones that change the market). In many cases the incremental ideas are your first activities, which might be what gains the momentum to go after disruptive ideas.

Here is an example of a simple value analysis:

Step 4: POC and Pilot

The step of POC and Pilot is to prove something to yourself, then to the business. You need to prove (1) that the idea works and (2) that it creates value before moving to production. The goal of POC is to prove the idea works. This might take only a few days, or a few weeks, depending on how difficult the idea is. The second is to move to pilot, which teaches the organization about whether the solution can provide real value to a set of business users. This is an important step because it quickly evaluates not only the technology, but how easily it can be applied to a business process. It also evaluates your team. You will have some that quickly adopt and achieve 10x success from adopting AI... and others that will resist and drag their feet. Quickly identify the landscape of all elements and use this to build your plan to move into production.

Things you might evaluate are:

1. What average amount of time will automating customer service requests reduce on a per employee basis?
2. What might I save in production using a supply chain optimization model in contrast to my current planning techniques?
3. How might my financial customers react to an interface that increases engagement and teaches them to make intentional choices?
4. How do the first 300 users react to having a tool like Microsoft Copilot on a daily basis as they perform routine tasks?
5. What is the impact on developer speed and quality as a result of having a tool like Github Copilot as the produce code or modernize a system?

All pilots need metrics of success and they should be financial if possible. The goal is to create more revenue or drive operational efficiency across the organization. If either of those (or both) can happen... you earn the right to scale.

Step 5: Production

The production step is not only scaling a single AI solution (whether commodity or mission-driven), but creating the necessary automation for that AI solution. If you remember scenario selection, we focused on picking a use case that would have a sponsor. It is at the production step that the sponsor is critical. If the sponsor does not exist, the solution will find itself struggling to deploy to production and challenged to receive the benefits. The sponsor must advocate and push a solution with the business users, gaining adoption because it might not come naturally. Remember that AI skills are NEW SKILLS, just like using a computer was to the previous generation. The 600 sales people you have will not immediately adopt the tooling all together. Some will adopt right away, others will follow, and some will wait. The sponsor is critical in achieving adoption of the solution.

The second element is that as a solution moves into production it leverages a solution like MLOps to achieve an operationalized state. Far too many AI solutions sit in "pseudo-production" where they never really get operationalized and as a result fail at the worst possible times. MLOps ensures the quality of operations are at a similar level to other high value operational systems.

Finally, the production system needs to be continually monitored for value and used to justify future projects, or to serve as a lesson to future pursuits. The best companies have a clear understanding of the outcomes produced from their AI projects.

So... how do we production? BRICK BY BRICK!

Step 6: Scaled Pattern

You thought you were done? You are just getting started! The goal is not to produce one AI outcome in a box... it's to create an entire company that is transformed through AI. This is made possible by scaling the patterns in commodity (float all the boats) and mission (improve the direct company) across every aspect of the business. You'll notice in the graphic below that this is buttoned up by the idea "Human Skills", which is the idea I talked about last week... that every person has the opportunity to use AI across their job role. That every business area can create value and we're just getting the machine going. You are creating not just one Value Factory, but scaled Value Factories across the product groups of your organization and enabling capabilities that have never been seen before in your industry.

THE TIME IS NOW. Let's make this the best 2024 ever!

Nathan Lasnoski