How to Speed Up Artificial Intelligence Initiatives with Agile Approach

Enterprises increased their AI spend by 62% in 2019. This means that these enterprises have realized the potential value of AI and are thus thinking about harnessing the power of AI to set them apart in today’s very competitive space. Even with the budget and number of AI initiatives increasing year by year, the greatest challenge that enterprises are facing is to implement these initiatives and realize their full value. The problem is not implementers. Usually, most are qualified not just with relevant certifications like the Scrum Master certification, but with expert-level skills and experience.

Strategic data science initiatives are characteristically high-risk and exploratory. To mean that they may or may not be the perfect solution for businesses. In the absence of appropriate risk mitigation strategies in implementing AI initiatives, chances are that there will be nothing to write home about their success rate. It takes the incorporation of an agile approach and a culture shift to realize the full value of AI initiatives in enterprises. 

Problem with scaling AI the traditional way

The traditional approach to building AI solutions employed the linear waterfall delivery method. In the waterfall approach, the development cycle flows in one direction such that the entire solution must first be built before it can be deployed and its value realized. The challenge comes with anticipating all issues and accounting for them before the solution is built. Even with perfect planning, this is impractical if not downright impossible. 

While the waterfall approach worked in developing solutions that prioritized business needs, it may not be appropriate for building solutions that prioritize customer needs which is the ultimate aim of AI. AI employs data and predictive analytics to establish the needs of the customer before designing a solution. 

The outcome of adopting the waterfall approach to scaling AI initiatives is that there is a high likelihood of building the wrong solutions. Secondly, much time and resources are taken in fixing problems that are discovered after a solution has been deployed. 

Benefits of Agile in AI

The agile approach to development came up to overcome the rigidity in the waterfall approach that prevented enterprises from realizing the full value of their AI deliveries. Although it was designed for software development, the agile culture is applicable to any aspect of the business operations. The Agile approach has greatly benefitted projects: 

  • That are large 
  • That are complex 
  • That have unknown solutions 
  • Whose requirements are likely to change in the course of developing a solution 

Agile works by dividing the project into smaller manageable cycles that can be built, deployed, and delivered iteratively. This approach prioritizes customer needs. It places emphasis on collaboration between teams and between the teams and customers to encourage feedback on every iteration. Just as it is in development, in AI this means 

  • End-users are part and parcel of AI solution development from the start
  • A shorter time developing effective solutions 
  • Faster and optimum realization of the value of AI deliverables 
  • Better collaboration which helps teams focus on customer problems 
  • Higher team productivity
  • Ability to discover and fix problems and act on feedback before launching the product 
  • Competitive advantage 
  • Elimination of reworks hence better resource use
  • Ultimately, a shift from data and technical skills to focus investments on AI solutions where the real value is

Leveraging the Agile approach to speed up AI initiatives

Fundamentally, leveraging agile for AI comes down to incorporating agile into the roadmap of an AI initiative on a case by case basis. This means that the Agile approach will be adopted throughout the lifecycle of the initiative from inception through to execution. It will be important to note that agile as a culture should not be overlooked merely because it is being integrated into individual AI use cases. 

Overall, agile plays an important role as a culture across the organization as well as on individual initiatives. Therefore, having the buy-in of the entire organization right from the management level should be a priority. Remember, the value of an AI initiative in an agile production environment does not only impact the customer but also the final product. 

Here are five ways in which the agile approach can be harnessed to accelerate AI initiatives. 

  • Identify value 

Value equals objectives. 

Every project starts by identifying the objectives of the initiative and then determining whether the solution will deliver the expected value before it is built. This is done by doing a cost-benefit analysis to be sure that the initiative will yield value worth the investment made on it. The idea will then go through an exploration process to determine the expected outcome followed by developing a proof-of-concept to demonstrate that the initiative is feasible. 

  • Factor in time-to-value 

Within the overall value also lies the time-to-value. While a proposed solution may be appropriate enough (in an agile environment) to meet a business need, the time factor could be the reason why it doesn’t fit the bill. Consider an initiative that is worth the investment and resources accorded to it and one that will solve the business need. However, if it will take a longer time for stakeholders to buy in the initiative or a longer time to deploy it to the production environment, its value may not be fully realized. 

  • Shift to an Agile culture 

It takes a culture shift for the benefits of the agile methodology to be achieved. The agile methodology is a culture that encompasses three core aspects including 

  • Cross-functional team collaboration 
  • Data-driven decision making 
  • Adaptable experimental development 

Shifting to an agile culture, in essence,  means a shift from siloed to collaborative operations, experience-based to data-driven decision making, and from the rigid waterfall to the experimental and adaptable agile development process. Ultimately, the entire organization will have been educated about and bought in the agile culture. Secondly, the development process ought to be iterative and adaptable enough to accommodate changes in requirements based on end-user feedback to bring out the most value in a product before the final delivery.  Finally, decision-making is more effective when based on recommendations from AI algorithms rather than personal judgment or intuition. 

  • Iterative and experiential development 

Iteration is core to agile development. 

Development is done in cycles to allow room for testing the model frequently and refining the features of a model in AI to increase its accuracy and adapt its suitability for its use case. The successful development of a model also depends on the capabilities of the infrastructure developed for it. Of importance to note is that much as aspects of the infrastructure can be reused, infrastructure should be specific to the model being built. 

  • Cross-functional collaboration 

Aside from the appropriate infrastructure, it takes the right mix of skills to deliver value. Development teams are cross-functional. In an agile setting, each person comes in with specific required skills and his/her roles should be clearly defined from the outset. This is to ensure that there are no grey areas when it comes to task allocation, no overlaps when carrying out duties, and that accountability and project ownership is achieved in each member of the team. 

Conclusion

The AI industry has matured. Incorporating agile methodology only works to strengthen it for better high-value project outcomes. Through iterative development, teams learn and get better as they carry out their tasks. Models, on the other hand, get better and more functional with iterations. Finally, the overall advantage of incorporating data in AI is that models are built to scale. 

Leave a Reply