Artificial Intelligence is becoming a part of almost every aspect of the business. Although calculating ROI in AI is very important, it’s not so straightforward. Netflix’s recommendation engine is a good example of it.
In 2006, Netflix announced the Netflix Prize. It was for improving the accuracy of their recommendation system called “Cinematch”. The target they set was to achieve a 10% improvement.
The Korbell team achieved an 8.43% improvement in the first year of the competition which was worth 2000 hours of their efforts resulting in a combination of 107 algorithms. These algorithms are still part of their recommendation engine.
When the competition ended two years later, they had an impressive compilation of hundreds of predictive models. However, as they moved from DVDs to streaming, the focus shifted and they had to make significant changes to the algorithm.
Netflix expanded its business into more countries and added new features such as users accessing their services through multiple devices. Now, with continuous optimization of their personalization algorithms, the outcome is 75% of what its customers watch is through some sort of recommendation.
Netflix, in recent years, reported that its new content recommendation engine saved the company $1 billion annually by reducing the customer churn rate. You can see how important and strategic the recommendation engine is for their business. Also, this is not a one-time effort but a continuous process of optimization of their recommendation engine to achieve this phenomenal ROI.
How to calculate ROI in AI
Calculating the ROI of investments in AI is sometimes challenging because the result is the combination of efforts of multiple aspects of the business. Here are a few considerations that might help in calculating ROI in AI.
Identify a specific problem to solve
The first and foremost is to identify what specific problem we are solving by the use of AI. When Harley-Davidson decided to use an AI-based marketing automation tool at one of their dealerships, the objective was to improve the lead generation and conversion rate.
Work out costs of implementing AI
The next step is to work out all the expenses related to the implementation of Artificial Intelligence. This could include:
- Prototyping and designing,
- Effort cost in developing, testing, and deploying AI solution,
- Licenses and tools if required,
- Data costs for capturing, maintaining, storing, and protecting it, and
- AI algorithm(s) maintenance and improvisation
Assess the benefits of implementation of AI
Identifying possible benefits from an AI investment might be trickier than calculating costs. The reason is it can include tangible as well as intangible benefits, as
- Increase in revenue,
- Efficiency gain,
- Improved productivity
- Compliance
- Competitive edge
- Brand value
Compare cost and benefits
Finally comparing costs against benefits can help understand the impact of AI investments on the business. A better way of looking at it through a few possible scenarios of best case and worst case.
While looking at both the costs and the benefits, creating a ‘proof of concept’ (PoC) can tremendously help to identify possible risks in investment. It can demonstrate the feasibility of the practical implementation from the ROI perspective.
Conclusion –
Investments in AI need to start with an experimental outlook than a typical software development project. A proof of concept is an important step in assessing the costs and benefits of AI to get a more accurate picture of the ROI.
A comprehensive exercise of working out the costs and benefits can get the ROI calculations closer to reality.