The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

5 steps to building and implementing AI solutions

These five recommendations are a few points to keep in mind as you add AI solutions to your current business operations.

5 steps to building and implementing AI solutions
[Adobe Stock / metamorworks]

Most businesses are able to recognize when they have a challenge that artificial intelligence (AI) can solve. However, when they move on to planning and implementation, they often fail to get their AI solution off the ground quickly. When trying to solve any problem with AI, it’s important to consider guidelines, resources, tools, and partners that can help bring such a solution to life.


Considering 87% of data science projects never make it to production, your team should aim to master these five steps when developing and implementing an AI solution.


Though it may seem like a no-brainer, failing to clearly establish what a project is supposed to accomplish can spell the early downfall of an AI project—at the very least, it’s where shortcuts become costly. It is vital that your team clearly define the problem statement you’d like AI to solve.


Once you’ve clearly defined the problem, you can begin to work through a series of steps to help determine how the model should function through specific use cases. Developing use cases should involve executives, managers, and customer feedback. This stage is crucial to getting a successful AI project underway.

Developing Structured Use Cases

Use cases can ensure that your AI development remains achievable and focused, and are typically derived from your core problem statement.


Strong use cases contain the following:

  • Clear objectives 
  • Established KPIs that determine how to measure success
  • A defined case owner to own the building, testing, and validating of the use case
  • A determination of what data is required for the AI to achieve the defined objectives
  • The identification of edge cases or unique use-cases that sit within the data required, as sometimes the data is not as consistent as expected and the details in the use cases have an impact on the success of a project
  • An evaluation of the ethical and legal opportunities and threats to determine that your AI solution brings value to your business, team members, and customers
  • An outline of existing technology and current capabilities that will support the development of the solution
  • Potential roadblocks and a plan to address them if and when they arise

This simple process can be repeated for as many use cases as needed to understand how your business will approach its AI strategy.



Once you’ve identified the problem you’re solving, the use cases for your AI solutions, and the outcomes you’d like to see, you’re on to the next step of gathering a dataset for the model to use. Preparing data requires a few steps:

  • Identifying necessary data for your solution
  • Determining availability of data and where it is sourced
  • Profiling the data (the process of examining, analyzing, and creating summaries of data)
  • Sourcing the data (pulling in data from multiple sources)
  • Integrating the data (combining data from various sources into a single data set)
  • Cleansing the data (filtering out insufficient data and, in turn, creating quality data)
  • Preparing the data for learning (establishing the data in a format your AI model can use)

Depending on the guidelines you’ve developed, this seven-step process could evolve to include additional procedures.



At the core of AI is machine learning. The way your model learns determines how the data provided will be used and if human intervention is required at the beginning or throughout the entire learning process of the model.

There are two types of training models: supervised and unsupervised learning. Supervised learning involves training the machine using a sample of labeled class data to teach the machine right versus wrong. Once the machine reviews thousands to millions of data samples, the goal is for the model to understand patterns independently.

Unsupervised learning involves the machine learning independently by trying to identify a pattern based on provided data. In this case, the machine is not told what data is helpful versus what is not helpful, or which data is correct versus incorrect.


According to IBM, “deep” machine learning can leverage labeled datasets (labeled by human intervention), also known as supervised learning, to inform its algorithm. Some deep machine learning does not require labeled datasets and can ingest unstructured data. As you narrow in on what you’d like your algorithm to learn on its own and the nature of your data, you can determine which learning model is best.


AI partners and vendors can come in many forms depending on the solution you’re trying to build. As you iron out the details of your algorithm, data sources, and ideal outcomes, you can determine which vendors will help you take your AI from a prototype to a fully implemented solution. Leveraging AI consultants in implementing and modeling can help your business determine your solution requirements and greatly speed up your time to production.


As you work with your partners and determine the right tools to help you build your prototype, it’s important to note that there isn’t a single tool or dataset that will solve your problem. This is where an experienced AI partner can help your teams determine the right set of technology, talent, and techniques to create the solution for your AI model.


Determining at what points humans play a role in your AI development will determine how your AI model functions. At some point in your development process, you will require the support of human intelligence experts. This can look different for every solution, so it is crucial to understand when human intelligence plays a role in order to get ahead of staffing requirements and keep your project deadlines.


AI can be an incredibly valuable tool to your business, including increasing your valuation or allowing your business to develop a proprietary data machine that can keep your business ahead for years to come. These five recommendations are a few points to keep in mind as you add AI solutions to your current business operations.

Jeff Mills, Chief Revenue Officer at iMerit