How to Avoid your AI Project Failing?

How to avoid your AI project failing?

How to Avoid your AI Project Failing?

LESSONS LEARNT AND BEST PRACTICES FOR THE DESIGN OF AI SERVICES

Written by Alessandro Giulianelli

The world is changing faster than ever before, and in many areas COVID-19 accelerated the rate of this process. To keep up with the fast pace, especially in R&D departments, the challenge is to bring forward innovation and entrepreneurial spirit adopting a lean approach.  This means to test one hypothesis as soon as possible and if not confirmed, trash it. Fail fast, fail often. However, when dealing with AI-oriented services, some figures show that there may be the need of a different approach.

AI FOR EVERYONE

According to the Gartner’s Hype Cycle for AI 2020, 47% of artificial intelligence investments were unchanged since the start of the pandemic and 30% of organizations actually plan to increase their investments in artificial intelligence. One out of three CEOs launched AI initiatives in their organizations and they regularly redefine resources, reporting structures and systems to ensure a successful project development. For what concerns the application contexts, AI projects continue to accelerate in healthcare, bioscience, manufacturing, financial services and supply chain sectors despite greater economic & social uncertainty. “AI is starting to deliver on its potential and its benefits for businesses are becoming a reality”, defining one megatrends according which we’re seeing a democratization of Artificial Intelligence, that means that AI is no longer the exclusive subject matter of experts.
This trend is confirmed also by the Futurescape 2021 of IDC that foresees that by 2022, 65% of CIOs will digitally empower and enable front line workers with data, AI, and security to extend their productivity, adaptability, and decision making in the face of rapid changes.

LESSONS LEARNT AND BEST PRACTICES

However, we should not forget another report from Gartner that says only 53% of AI proof of concepts are successfully transitioned to product. Meaning that almost half of AI projects fail, due to the many factors playing a crucial role in this context, as such as the reliability of data, the lack of skills, missing domain understanding and uncertainties in customer relationship.
Based on the experiences developed within the Konica Minolta R&D group, these are some of the most commonly known difficulties that may rise when designing an AI project:

  1. Business misalignment. The biggest risk with AI-driven technologies appears when we consider them as business steroids, capable of boosting technology to solve any issue. Following the hype around big data, everyone talks about AI, NLP techniques or ML algorithms without understanding the differences and complexities. However, with accurate data-driven insight it is possible to achieve an effective business decision making. To align the expectations of the organisation management with the real potential of AI, we need to define the right assumptions. The developers’ team (the Dev Team) need an exhaustive set of information together with a strong motivation for the action. In few words, we need to focus on the why rather than on the how, and be less task oriented.
  2. Technology driven approach. When the motivation behind a project is not clarified in the first place, the risk is that the Dev Team is filling the gap with their technological skills, substituting the why with the what. That’s when IT companies create false expectations in teams that commit for delivering a new product or an innovative service starting from the technology. Quite the contrary, clarifying the motivation means focusing on the business value. Following a customer centric approach is at the core of Konica Minolta Digital Services R&D manifesto: with a methodology of Applied R&D, we identify projects that can bring value to a customer rather than developing a project on the basis of its technological content.
  3. Data strategy. Even though everyone knows data is the oil of the 21st century, many are not aware of the importance of data content, data quality and data governance. The type of data and its quality should be taken into consideration in the very early stage of the project. Then, when dealing with data, it is of utmost importance to avoid silos within the same company: in most cases, siloed departments pose serious obstacles to the understanding of business values and to the modelling the proper domain knowledge. A good coordination of the Dev Team together with business units is essential for the good start of an AI project.
THE ROAD AHEAD

Overall, in the last years, many organisations have learnt a lot about developing projects with artificial intelligence, and nowadays we’re making less mistakes than in the past. Therefore, to keep improving the way we design AI projects and decreasing their failure rate, there is the need to make a switch in the organisational culture engaging the whole company into a real digital transformation. At the basis of this evolution, at Konica Minolta we place the Customer-Centric AI Project Cycle:  a process that can be split in five steps to support a robust definition, management and development of a solution based on artificial intelligence.

customer-centric-AI-project-cycle

The Customer-Centric AI project cycle – Konica Minolta Digital Services R&D

  1. Framing the business problemis key to determine the business values and related objectives. That should be the first step in every project, however it is fundamental within the data science area. And knowing who your target customer is, should be the primary goal of initial research. For instance, a good start could be to cluster customers on the basis of their industrial context, as such as agricultural, manufacturing or services for instance. Then, regardless of their sector, focus on the departments that a company usually comprises: HR, finance, marketing, sales, supply chain, inventory and procurement and IT.
  2. Then, following the Design Thinking approach of constantly putting people first, execute a set of interviews for every target customer and for every given department. This is a way to identify the right use-casesmaking each AI solution customer-centric. We believe Design Thinking is crucial to secure the potential of AI, because it helps to focus on the change we may introduce in the customer behaviour with the new solution. Being aware of the technological feasibility step in this phase, we need to focus on the business problem first, and then think of data; tools and algorithms will follow naturally.
  3. Third step is about mapping and collecting data: based on the chosen industry and department, we need to start gathering data, map and put in place data governance procedures. Setting up a mixed-role AI teams is essential in this phase, as highlighted in this list of good habits of organisations with successful AI. Developers, Data Scientists, Architects, Salesmen, Marketing experts, Managers all together bring their expertise and keeping up with Design Thinking, everyone contributes to figure out what the userneeds are and how to fulfil them. This approach is quite the contrary than looking at a problem and immediately working for a solution.
  4. Once you know the business goals, understand the customer domain and have the data correctly managed, then you can face the task of identifying the technology stack for the specific use case. In this phase, we all should remember that the solution is not technology-driven but customer-oriented and it hardly can be replaced with generalized solutions.
  5. The last step is about measuring. Choosing the right Key Performance Indicators (KPIs) can make the difference between a successful product and an expensive exercise. The following are two KPIs that focus on the end user:
    • Customer Satisfaction Score (CSAT) is one of the most crucial customer-centric KPIs because it tells us if our project results match our customer’s expectations. It’s important to notice these are our customers’ expectation and not ours.
    • Another important KPI is the System Usability Score (SUS)Many times, some AI driven features are explained so badly that it is impossible to understand what they do and why someone spent time deploying them. Even if when dealing with AI, this misalignment may happen, embracing the Design Thinking approach makes you thinking at the system with the end-user’s eyes, and it enables you to focus on what really matters for your customer.
AI ALONE IS NOT THE ANSWER

Overall, artificial intelligence has a huge potential of impacting our lives in every aspect. However, AI is not the source of innovation as it is only a set of technologies, and we should restrain from thinking about technology as the sole-domain of innovation. We must keep the customer at the core of our approach, resisting the temptation of the powerful lure of AI solutionism. In the end of the day, our customers care for improved processes and better quality of services, and they are rarely interested in the technology that is behind the product that is solving their needs.  And it’s within this context that Konica Minolta Digital Services R&D is developing cognitive services that explore the orchestration of AI to effectively manage the growing complexity of every workplace. Get in contact with our researchers to develop together a use case project that focuses on your need.

Photo by LinkedIn Sales Navigator on Unsplash

X