This blog post was co-authored with my colleague David Mosen.

A lot has been written about how AI is the next big thing, how it is revolutionizing businesses across industries, even comparing its transformative power to that of electricity 100 years ago. These are catchy headlines that seem in stark contrast with others that predict and observe staggering failure rates for AI projects. So, what is the source of such an apparent contradiction? In this article, we will explore some of the difficulties derived from the particularities of AI projects that can eventually keep them from being successful, from a technology and project management perspective. For more business-related reasons of why AI projects fail, we refer the reader to a pair of insightful articles published in the CIO and Harvard Business review magazines [1][2].

Just as with software projects, AI project management requires proper execution of processes that address business concerns, like requirement, risk and release management. Likewise, causes for failure are mostly shared, including poorly defined goals, miscommunication, insufficient end-user involvement and improper assessment of available resources. On top of that, AI projects face additional hurdles, arguably arising from the introduction of data science into the mix, with its data requirements and the novelty of the methods often involved.

We have compiled a list of the six main challenges that we have been observing during the delivery of 100+ AI projects:

Challenge 1: Trying to run before you can walk

Digitalization precedes data-based projects, like those involving AI. Going from analog to AI is sure to be a treacherous journey, assuming that it is a possible and sensible one. Granted that we are at a time when most large businesses have a clear idea of whether, how and to what extent they can benefit from digital infrastructure. However, it is also true that most are still not data-driven, a last leap that often requires a change in leadership and a great deal of flexibility. In a nutshell, digital data-driven processes are not a prerequisite to a first AI project, but some will need to be implemented if not already in place.

Challenge 2: Stretching the short supply

Data science and AI expertise is not only scarce, but also dispersed in the knowledge space, in the sense that it is still a broadly defined domain that gathers from several better-delimited fields like statistics and software engineering. From a project personnel perspective, this means that single profiles which perfectly match the required expertise might not always be available. Consequently, team harmonization and coordination take a more prominent relevance.

Challenge 3: Communication gap

Together with business and software engineering, data science represents the third leg of an AI project; an addition that introduces further complexity to the interactions across knowledge domains, on top of its intrinsic intricacies. This calls for even further efforts on communication and conflict management, as well as change management. Having a culture that favors keeping an open channel across stakeholders will go a long way.

Challenge 4: You can’t assess your tech cake and eat it too

AI projects require thorough, yet limited-in-time, technology assessment. The AI and data science landscape is still developing at an accelerated pace, with not just new tools and services, but also new paradigms. Thus, projects will face the following compromises: 1) consider the best balance between cutting-edge and maturity; and 2) deciding whether to explore less options in more depth, as opposed to a wider range of options at a higher risk of an a posteriori realization that choices were not a good fit. The hype with AI also increases the chances of using a sledgehammer to crack a nut, instead of selecting a technology stack that matches the scale and complexity of the project. Thus, it becomes paramount to be critical and realistic and, as always, to adapt to the existing environment and actual requirements.

Challenge 5: Locked-in

Another consequence of a yet-maturing AI environment is the high probability of any given technology or vendor to quickly grow out of favor. This situation is combined with the fact that large cloud providers are rapidly becoming the go-to vendors for full-cycle data science development, each offering, in addition to flexible infrastructure, simplified deployment models that involve more proprietary technology. Overall, this makes it important to carefully assess the possibility of lock-in and the level of portability when choosing each of the pieces of the tech stack, even within each vendor’s offerings.

Challenge 6: Data is right, aligned

Software developed within the scope of a project is required to be aligned with business objectives. So does data that is gathered and prepared. Yet, while software requires engineering processes, data requires processes that mix engineering and science, including collection, exploration and transformation. Hence, this process needs different skills and it needs to be tackled from a different angle. Specifically, checking for data availability, completeness, suitability and unbiasedness lean more towards a science approach. These will typically precede any engineering processes and have been demonstrated to have ruined finalized projects for not having been carried out correctly.

Seek professional help

Whether you decide to undertake the journey in-house or to seek the expertise of an AI consulting services partner, the above challenges have to be addressed to arrive to a successful AI project. A success formula will have to bake in a proper AI mindset that includes the environment, processes, people, data and technology into its project execution and management.

At Crayon, our trusted AI-360 framework addresses these and other challenges posed by AI, helping our clients navigate such complicated technologies. We accompany our clients on their journey to achieve AI-enabled excellence, from strategy to implementation and as well as organizational change and IP development. As a first step, we typically engage with our clients to assess their situation and accordingly set a bespoke roadmap for data and AI efforts, essential for the appropriate planning of successive AI projects (Challenge 1). As of early 2020, with 5 AI Centers of Excellence worldwide, Crayon has delivered over 100 AI projects – and counting, thanks to our growing team of 50+ data scientists, data engineers and AI advisors, selected among the best (Challenge 2). They can speak the client’s language, as well as data and software, and are organized in a way that streamlines communication processes to minimize communication gaps across stakeholders (Challenge 3). Further, our success in ensuring the best possible outcomes for our clients unquestionably relies on the fact that the members of our team are experts in their fields and are able to determine the most suitable technologies while carefully balancing trade-offs (Challenges 4 & 5). On the data side, data planning, access and management are key enablers for data-driven organizations, so we aim at designing purpose-driven data infrastructure that effectively support business objectives. If the relevant data infrastructure is already in place though, our data scientists and data engineers work together to acquire deep understanding of the data and assess how to best move forward, always in close collaboration with the client (Challenge 6).

If you do decide to seek the expertise of an AI consulting services partner, reach out to us and we will evaluate the AI-360 journey that best matches your needs. A proper opportunity assessment, like Crayon’s AI-360 Discovery offering, in which the business problem is identified, the data is assessed and a solution is conceptualized, is a good place to start and will mitigate the effects of the above and more challenges down the road during the planning and the development of a full project. Other Crayon offerings, like AI-360 MVP, AI-360 Production and AI-360 Model Management as a Service, will be more suitable if you are further in your AI maturity ladder and are ready to implement a solution. Here as well, Crayon’s AI-360 framework will have your back so that no technical or business challenge keeps your AI project from becoming a success.

References

[1] https://hbr.org/2019/02/how-to-choose-your-first-ai-project (Last visited: February 17, 2020)

[2] https://www.cio.com/article/3429177/6-reasons-why-ai-projects-fail.html (Last visited: February 17, 2020)

(Please note that that this post has originally appeared in the We Are Developers Magazine blog.)

Comments