Four reasons your AI plan is failing (and how to fix them)

Share this post:

Despite billions of dollars spent on AI and digital-transformation programmes over the last twenty years, failure rates in delivering intended outcomes are now at about 75%. Three-quarters of these initiatives fail.

Why? Because ideation is easy, and implementation is not. A good idea about how to use and automate data is not the same thing as understanding how to operationalise that idea within an organisation.

Clearly, going from “good idea” to business outcome is the stumbling block. So, here are four things — beyond the often-blamed constraints like technology and budget — you can do to lower the risk of failed AI.

First: Start with a laser-focus on business value

To succeed, your program needs to examine what is holding the company back, rather than playing the all-too-frequent ‘dreaming of data science’ game. That’s a fantasy of magical algorithms, machine learning, and deep learning that results in a perfectly orchestrated business underpinned by insights. What you end up with is lots of great ideas, POCs, MVPs and some optimistic looking charts.

Why? Because the idea simply wasn’t a good one in the first place. Is it solving a real need? Is that need understood? Is this the best way to solve it? How do you know? These questions don’t always get asked, and as a result you can start work with half-thought through hypothesis, meaning your results are a question of luck.

The IBM Garage operating model is an orchestrated approach with razor sharp focus on measurable business outcomes, with a safe space to succeed and fail, that’s transparent, and iterative. Data scientists have used this kind of approach for decades, but Garage has the ability to scale and adapt to the needs of many different business challenges.

IBM Garage isn’t just agile or some design thinking. It’s not a place where MVPs go to die. It’s not input- or output-driven, and it’s not a delivery method. Garage isn’t the thing, it’s the thing that gets you to the things (outcomes).

IBM Garage is an operating model, a collaboration framework that delivers at speed and at scale, with an unwavering focus on outcomes. It brings together the way we do things for our clients. It’s not SAP implementation or Garage: it’s and. It’s not data-driven transformation or garage: it’s and.

IBM Garage is a process that quickly gets you to a clear understanding of what’s holding your business back, and a process for testing and building a solution.

Second: Track business value during the project

One you’ve identified it, stay true to your business-value focused plan. Too often, once a project is set up, it’s the execution that’s tracked – the triple constraints – not the attainment of business value that is the purpose of all your effort in the first place.

While everyone talks about data, too often the most meaningful data that’s actually available to us isn’t getting used. Initiatives and programmes focus on KPIs that measure whether the program is on track, rather than whether it’s leading to increased business value. Without the time and skills it takes to understand the value levers and how to track to them, initiatives can veer off course.

Here’s an example: Matalan had a lot of data: 90% of their customers carry loyalty cards. But they were trapped in ‘analysis paralysis’, cutting different views of that data but not taking a step back to think about their most pressing business questions.

Together with them (and board-level sponsorship), we created the Matalan Garage, a space where myths could be busted, and models could be built, measured and scaled. In 12 weeks we identified £3 million in value, changed their approach to data-driven business, and pivoted them to insight driven outcomes. They were highly commended by the Lloyds National Business awards for Data Excellence.

Third: Manage the human change factor front and centre

This is the one that that is at the heart of it all. Transformation programmes introduce new tools and technologies, change job roles and organisational structures, and often introduce new expected ways of working. These programmes are governed at an operational level. Everything can appear to be on track.

But if leadership doesn’t stop to understand what, or how to address their people’s individual fears and desires which will shape their behaviour and redefine future operations, the initial positive results won’t last. Change Fatigue sets in.

Fourth: Make sure to earn people’s trust

For your AI program to succeed, it needs to be trustworthy. It’s a simple idea, but its implications are profound. A recent survey found that 84% of AI professionals agree that consumers are more likely to choose services from a company that offers transparency and an ethical framework on how its data and AI models are built, managed and used.

So, for example, if a bank AI declines a mortgage application, that data-driven decision, whether automated or not, cannot be based on ethnicity or gender. You want it to be based only on data that are relevant for the release of the credit and the ability or not to return it. If machine learning is used, unintended discrimination could be hidden within the instructional examples provided to the AI. If the training examples don’t represent the plurality of all possible situations, it would be difficult for the system to generalize. If the examples contain only cases of mortgages given to men and refused to women, the AI will associate gender with the acceptance of the request: when asked to analyse a new credit application, it could use this feature of the applicant to propose or deny acceptance.

So, just as within your business teams, you want diversity reflected. It’s issues like these that can emerge and erode trust in the initiative and ultimately upend an otherwise well-intentioned deployment of technology.

IBM Garage: A safe space to ask ‘why?’ before you try

A leading discount retailer had a hunch that AI was their ticket to huge expansion, but they didn’t have the in-house expertise to scale. They’d built everything themselves and were struggling with scaling-up.

We implemented the IBM Garage operating model, and with them, quickly upended their business to focus on insights-driven decision-making. It was a big step away from their gut-feel approach of ‘let’s try this’. They are now asking, ‘what are we trying to achieve?’.

Five months in and this client is achieving £2m of additional sales every week. We’re building their in-house data science and AI capability and we’re whilst training their execs to ask the why before they try.

AI can change business fortunes. Start and continue with business value, and focus on how it impacts real people, within your business, and in the world you touch. It’s time to invert that failure rate and build for success.