One reason that artificial intelligence (AI) initiatives fall short is that organizations’ data isn’t ready and consumable.

Determining whether data has the intelligence to deliver the desired outcomes is an important upfront task for any business looking to incorporate AI. The old adage “You can’t manage what you don’t measure” increasingly applies to an organization’s wealth of information assets. Does it have the right data on hand and in the right condition to predict, say, which customers might defect?

Figure 1

Here’s how to determine your organization’s data readiness for AI. We also offer recommendations on how to prepare your data for an AI deep-dive.

Develop a new set of metrics for data.

Determining whether your data is up to snuff for AI often requires a new set of metrics. Many enterprises measure their data organization’s success in ways that don’t target business value. For example, key performance indicators (KPIs) for IT regularly include whether the system is ready when reports need to run, or whether a particular report or data feed was executed within the agreed upon time window.

For AI to be successful, however, data metrics need to measure business value and the ability to deliver the desired outcomes. That measure doesn’t exist today. Measuring your data IQ will help you hone in on business value.

What’s more, decision science assumes that your organization has the data it needs to be effective, and that it’s in the proper shape. Many organizations find themselves coming up short. They’re unprepared for the breadth of data needed for AI. Take retail sales data, for example. Understanding store performance requires invoice-level data tagged to specific departments such as women’s coats or men’s action wear. To validate their hypotheses, data scientists might also request details on online promotions associated with the sale. They might ask for intraday sales information to measure the efficacy of the promotions.

Hunting for data is time consuming. For data scientists, time spent searching for data and then getting it into the shape they need takes away from the model creation that’s at the heart of AI. In our work, we observe that data scientists often spend more than 70% of their time on data preparation, leaving them just 30% of their workday for model engineering.

The takeaway: To maximize the time spent on model creation, make data preparation an integral part of the data platform’s function. Assign responsibility for it to data engineers. Make it the goal of the data organization to deliver analytically ready data sets for the use of data scientists.

Evaluate whether data is available at the point of action.

AI requires fast, easy access to data. How automated are your operations? How available is data at the time of action? Inventorying decision points in your company’s value chain will help you fill in the blanks and begin to craft answers to these questions. With a master list of business processes, your team can then identify the decisions that need to be made for each process. Inventorying business processes was common in the early days of enterprise resource planning (ERP) systems. Today, both application and data transformation tend to be treated as separate initiatives. Consequently, they fall short of leveraging data and insights that can deliver superior process outcomes.

Embedded analytics and insights at the point of action can make a big difference. For example, wholesale buyers typically have little information available when they create purchase orders. Adding intelligence to the process can result in different decisions: Imagine a requisition screen with pop-up alerts that feature not only vendor pricing and on-time shipping performance but also alternative vendors whose price may be higher but who always ships on time. It’s information at the point of action.

The takeaway: By looking at the whole of your organization and the decisions that make your business run, your team can begin to close the gap between application and data transformation. An inventory of your business processes can begin with the following questions:

• Which decisions are critical and impact profit and loss (P&L)?

• Which decision points are powered by analytics? Which aren’t?

• Which processes are prone to decision errors that can be eliminated by providing more information to process participants?

Score your data IQ

Now that you know the points in your operating model most likely to benefit from data, it’s important to determine whether the data you have is in the right form and shape to provide those benefits. Quantity doesn’t always equal quality.

Data IQ frameworks provide a quantitative measure of whether the information your organization has on hand will assist team members in the decisions your organization seeks to make, such as trying to determine a customer’s long-term potential. Frameworks turn up information that otherwise isn’t apparent. For example, it’s not uncommon for organizations to discover that their data is of poor quality or too limited in scope to provide, say, a window into how the customer engages with your company. The few months of data available on each customer may not be enough to base predictions on. Even several years’ worth of data can turn out to be far less robust than anticipated if it lacks the necessary details.

The takeaway: Use a data IQ framework to measure the intelligence of your organization’s data landscape and its readiness to deliver analytics for your business objectives.

• Do the data elements correlate? Are there redundancies?

• Is there diversity of data, or is there too much concentration of similar types of data?

• What potential anomalies exist among data patterns and characteristics due to inconsistent business processes and data-capturing techniques?

Identify your organization’s strengths and weaknesses in data intelligence so your AI effort delivers the outcomes you want.

Next: Read Part 3 “ How Will Your Organization Use the New Data?”

To learn more, visit the AI & Analytics section of the website or contact us.