Is AI a Sprint or a Marathon? Development ‘Sprints’ Get Companies to the Starting Line Fast 

via Shutterstock

AI is a complex process that typically takes years to do right, requiring a huge investment in time, resources and money. At costs that can average upwards of $500K to design, develop and implement a robust, long-term AI solution, companies need to undertake their AI initiatives with precision and care. It can take time to see how “smart” the solution is over time, and how accurately it can predict outcomes, identify patterns or help companies better engage with customers.

But this isn’t the only way to approach AI development. Companies can instead turn to AI design and development “sprints” to quickly kick the AI tires at lower cost than full deployments, and help them test out the AI waters. Since this an easier, less risky and less expensive initial approach, we can expect to see more companies using sprints this year.

AI sprints are short-term, proof-of-concept AI projects that can help companies determine with some level of certainty – within one to three weeks – whether AI can help achieve specific business goals. They’re typically implemented by an AI developer or vendor, and the solution can be the basis for continued AI development and engagement.

AI sprints are a natural evolution of the design sprint concept created by a division of Google, Google Ventures, introduced several years ago. This consists of a five-day process to create a solution through design, prototyping and idea testing with customers. It provides a time-condensed view into the validity and usefulness of a specific solution to a business challenge. Since Google’s launch, many companies have cashed in on the design sprint concept for creating new products and services in a variety of industries.

According to physicist and author, Mark Buchanan, “Artificial intelligence is going to change the world profoundly, although exactly how is still unclear.” Adopting a sprint strategy to AI development takes out the fear, uncertainty and doubt (FUD) factor that is a key challenge to greater AI adoption. Many business users not only find the concept of AI to be extremely complex, but they are also often unconvinced of its value. A sprint allows them to cut to the chase and experience the benefits of AI, while setting a clear roadmap to full execution.

The Innovation Sprint Process

Below are the five key steps that a typical innovation sprint should include:

• Identify the business challenge. Easier said than done, and since different decision makers may have different viewpoints on what the key problem is, it’s important to discuss and reach a consensus of what you are trying to solve. Sometimes, AI may not be able to solve the specific problem identified, so it’s important to clearly articulate the problem.
• Conduct a data audit. Once the business challenge is identified, next determine what data you have that helps address it. Data can come from different sources -- in CRM systems, corporate databases or files in structured or unstructured formats. A data scientist can help identify appropriate data, determine if you have a sufficient amount of it to feed an AI algorithm, and help supplement that data with external data-sets, if needed.
• Build the algorithm. Once sufficient data is gathered, the data scientist feeds it into a prototype to validate its accuracy. If the findings indicate that the data can support the business problem, then the data engineer will clean, label and classify the data; and the data scientist will select the most relevant AI architecture and develop the algorithm.
• Assess the results. Within days, the AI team will determine how well the solution predicted a reliable outcome and make adjustments accordingly. For example, we conducted an AI innovation sprint to help a major health insurance provider identify the reasons for customer churn. By expanding and refining the data-set through several iterations to continually improve results, we were able to develop an algorithm that predicted the likelihood of churn with 93 percent accuracy.
• Operationalize the algorithm. Once the algorithm has proven itself, the solution becomes operationalized – usually within a few weeks. Since the solution, however, constantly learns as it receives new data, its accuracy will continue to improve.

Sprints Gain Popularity

While no one is yet to coin an industry-standard term for these short bursts of AI trials, they’re already in use among leading vendors. IBM announced a Data Science Elite team last year, created solely to help big enterprises push their first AI models into production. According to an industry article, companies like Harley Davidson, Lufthansa, Experian and Sprint (no connection to the name of the process) have used the IBM team to kickstart their AI initiatives through these types of sprints.  As AI is becomes a must-have technology, AI innovation sprints may become the clearest path to the starting line.

Carlos Meléndez is COO and co-founder of AI and software engineering services firm Wovenware.