Top 5 Reasons AI Projects Fail
Beyond investing in data science functions, learn what businesses really need to set themselves up for success.
Research tells us that 84% of all businesses intend to use AI in 2023. We also know 85% of all AI projects fail. With the potential outcome of that many businesses failing to effectively launch AI projects, the time is ripe to address the common stumbling blocks. Below are my top five reasons that AI projects fail (and a couple of bonus reasons).
1. There’s no easy access to reliable data
The purpose of data science is to find patterns in data, which in turn enables predictive or prescriptive insights. It follows that without access to data, there is no data science. Sounds simple, but — as always — the devil is in the details.
Access to data ought to be easy. When it isn’t, practitioners are forced to take it on themselves to create workarounds so they can do their job. As a result, the projects often go off-track, timelines are missed, and risky ungoverned processes are created. This renders the data unreliable, and even worse, erodes trust in the data team.
The good news is, there is a range of modern architecture solutions which, with the right level of investment, can eliminate this problem.
2. No value prioritisation
A successful data science project is defined by the value it adds to business objectives. This suggests the need for an articulated list of use cases, solving specific business problems with clearly defined outcomes and estimation of value.
There are a number of product management frameworks — such as RICE and MoSCoW — that can be adapted to data science and will come in handy when the use cases are well understood. However, estimating the value of a brand-new data science project is extremely difficult. This is where product thinking is pivotal. The single most effective approach for value estimation is experimentation — quickly build a simple data science product, and test how well it influences the intended outcomes. Many experiments will suggest no value but will do so quickly and with little investment.
3. No capability strategy
The data science profession is new and rapidly evolving. We also need to acknowledge that data roles have evolved in the past few years. It’s no longer necessary to rely on the same person to do all data work. Maybe 10 years ago, “Unicorn Data Scientist” was a valid job role, but the discipline has grown to enable a wider variety of roles. As part of this evolution, practitioners go into the field with an acute awareness that this is a profession, not just a passion or a hobby, so factors such as market-relevant pay, great working environments, and infrastructure that enables them to thrive are all important.
4. Working silos
To be successful, data science requires a multidisciplinary environment. It combines domain expertise, mathematical modelling and computation to produce results that enable experts to do their work better. This is true for human-centric data science, and the artificial intelligence process is no different.
The value of a data science project is realised through the business actions that it enables, which is why, when it comes to the team, the sum of the parts is much greater than the individual components. An effective target operating model removes any guesswork and defines clear accountability. To use analogies: a data science team without engineering involvement is a passion, but without business accountability it’s just a hobby.
5. Lack of platform for scalable operations
Data science is often viewed as a luxury good — a very expensive cost centre that doesn’t always justify the capability investment. To clarify, while data science projects come at a cost to set up and develop, if applied at scale and used at high frequency, they keep generating returns with little further investment.
So let’s talk about the factor that significantly increases efficiency and allows the consumption of data science at scale — the MLOps platform. These tools manage the lifecycle of data projects at scale and allow for quick deployment of new products (hello, experimentation!). The good news is that the market has caught up and there are technology solutions, either by the large providers such as AWS or Microsoft, or by small independent start-ups, which can be tailored to suit any organisation’s needs.
I also promised two bonus factors for successful data work, and you may already be guessing what they are, as they appear throughout: culture and technology. These go hand in hand — data science thrives in organisations with data-enabled culture and technology solutions, which allow for effective decision-making.
Do these stumbling blocks resonate with your experience? What would make your top five?
Slalom is a global consulting firm that helps people and organizations dream bigger, move faster, and build better tomorrows for all. Learn more and reach out today.
