Implementing AI in 3 Steps

Jeff Grisenthwaite

Jeff Grisenthwaite

· 4 min read

ai machine-learning how-to

On a daily basis, we read about the upsurge in artificial intelligence (AI) taking on more complex tasks in the workplace. But for many employees, AI hasn’t emerged to affect their day-to-day lives at work, and they continue to spend hours every week on activities that could be augmented or automated by AI using currently available technology.

One of the barriers to wide-spread adoption of AI throughout the workplace is the common perception that implementing AI requires advanced technical skills or months of development time. For many AI use cases, the implementation is much simpler than that perception. Let’s explore the 3 basic requirements to get started: a decision, some data, and an AI platform.

Step 1 - Select a Decision

Start by finding a business decision that would be useful to augment or automate with AI. Good candidates have these two qualities:

  • High Frequency: The decision is made at least 300 times per year across all employees. Higher frequency is even better, because it will mean more data to feed the algorithms and greater opportunity to save time and improve decision making.
  • Multiple Factors: There are least a few inputs that a person would use to determine the best decision each time. The more inputs, the better, since machines are much better at dealing with dozens or more variables than people are.

Examples of good use cases for AI that are applicable for almost every department include:

  • Manager Approvals: Determine if an expense or a discount on a sales opportunity should be automatically approved.
  • Routing Emails / Tickets: Categorize, prioritize and route emails or support tickets based on the content of the inquiry, the language of the text, the sender and other factors.
  • Forecasting: Predict the likelihood of a sales opportunity to close, the risk of a customer to cancel, or the estimated completion time for a project.
  • Auto-Responses: Automatically completing RFPs and RFIs based on a library of past responses, or responding to a customer question with the likely best answer.
  • Flagging Anomalies: Sift through a number of records to identify the small set of outliers that need to be reviewed or analyzed.

You can decide the role that AI should take in the decision-making process, and you can take a staged approach to introduce AI in a seamless and low-risk manner:

  • AI Recommendation: Enable the AI to analyze the data and provide a recommendation to a person. This allows for the subject matter experts to observe the AI, gain trust that it’s making sound decisions, and also save some time as they are making those decisions.
  • Assisted AI Decision: Set parameters for when it’s appropriate for the AI to make a decision and when that decision should be made by a person. These parameters could include the confidence of the AI in the decision, as well as the stakes or risk for a particular decision based on the factors.
  • Automated AI Decision: For some decisions, an AI will learn to become more accurate that people, or at least accurate enough that the time savings for people is worth having the AI fully automate the decision.

Step 2 - Gather Data

Machine learning (ML) uses data as its teacher. The algorithm will analyze historical decisions and look at the factors for each decision against the decision that was made. With enough decisions and data, patterns emerge that can be leveraged to make predictions for future decisions.

How much data do you need? It depends, but you will at least need hundreds of examples of decisions and the factors that went into them.

Where the data resides does not matter so much, but it will need to be digitized (not sitting in stacks of paper). It also needs to be labeled in a structured way—either fields in a system or database, or a spreadsheet with rows and columns.

If you don’t have enough labeled data yet, the best approach will be to start the discipline of gathering the factors and decisions in a structured way, so that you can use it to feed the AI platform in the near future.

Step 3 - Select an AI Platform

You will need a platform to bring AI to your decision making process. As you select a platform to make AI a reality for your department, consider these factors:

  • AI Capabilities: Talk to the providers about the decision that you’re looking to automate or augment to determine if they have the AI capabilities that you need. These could include machine learning, natural language processing (NLP), optical character recognition (OCR), fuzzy matching, sentiment analysis, image recognition, and more.
  • Data & Integration: You will want this process to be as automated as possible from ingesting your historical and future data to recording decisions and initiating actions in other systems. Talk to the providers about their ability to connect to the systems, spreadsheets, or databases that will be involved in either supplying data or receiving the outcomes of the decisions.
  • Workflow & Tasks: You will want to be able to control whether this is an AI recommendation or assisted AI decision, and you will want to be able to assign tasks to people to be able to take action when the AI decision-making is not fully automated.
  • Flexibility: Change is inevitable. You will want to ensure that you will be able to update factors for the decision, change the source or output system, or modify the workflow. And you will want to be able to make all these changes without requiring an engineer or data scientist every time.

With these 3 steps, you should be well on your way to bringing AI into your department, and beginning to reap the benefits of automated decision-making.