We’re not talking AI on the scale of Hollywood films such as robot ‘humans’ and personalised billboard advertising here. Rather the focus is on projects that use intelligent technology to optimise what a business already has; forecasting demand in order to match supply for example so that customers are not disappointed but excess stock does not go to waste, or determining where a retailer should place a specific product in-store to maximise sales.
The high risk of failure is a major problem as organisations strive to transform their business operations to be fit for a fiercely competitive world. It was the case before 2020, but the pandemic has shone a spotlight on the need for fast, efficient and effective ways to operate that will aid recovery.
So why is the failure rate so high? Is AI just too difficult? Not in our experience.
The key lies in the initial innovation, or proof-of-concept (POC). However brilliant and ground-breaking it is, if it can’t be rolled out across the organisation then, ultimately, the work has been in vain. Not only that but, because costs for projects of this nature tend to be front-loaded, it is likely to have swallowed around 70% of the budget allowed for the total enterprise-wide implementation.
Our practical, hands-on work across a range of industry sectors and with a wide variety of organisations gives us valuable insight. Analysing our findings sees some core themes emerge when identifying the issues that cause roadblocks in AI and digital transformation projects. We believe resolving them requires every project to incorporate the following three practices:
1. Innovations that scale
All too often POCs are undertaken in isolation, for the sake of proving something can be done. The end of the initial project defines its success – without accounting for whether the idea can be effectively used to meet the needs of the whole organisation.
In contrast, truly successful innovations include metrics that factor in the scalability of the blueprint; this adopts a business-critical mindset requiring projects to demonstrate that, as well as achieving the core objective, scale is a prerequisite.
Including two countries in a POC for example, and judging its effectiveness on whether it delivered on the goals for both regions automatically builds in depth and reach. Similarities and differences between the two streams are identified, with this insight informing how best to expand into the next territories. Process and structure have been developed and proven as part of the POC; when the project starts to be more widely integrated, the foundations have already been laid.
2. The right skillset for the long-term
A scaleable implementation is dependent on breaking down the silos that traditionally exist within organisations, whether between business and IT, business and software engineering, etc.
Achieving this requires a multi-disciplinary team that includes a data scientist (the mathematician), product manager / owner (for the business perspective), solution architect (to bring IT skills), and machine learning (ML) engineer with good ML operations (MLOps) knowledge (to provide coding at scale). But digital transformation runs deeper than the right skillset.
Each of the skills needs to be incorporated into every phase of the project, from day one to its close – and beyond (the solutions architect and ML engineer roles are often omitted, or introduced after key decisions have been made). Business continuity requires the people that will run the implementation on a day-to-day basis to be trained before the POC team leaves. The viability of a project is its capacity to be embedded throughout the organisation for the long-term.
3. Transferable tools
Project survival once the POC team has finished also calls for the right technology. Innovations require an accompanying ‘toolbox’ that has been planned-in from the beginning with the objective of enabling enterprise-wide rollout and long-term maintenance (along with training the people who will use these tools).
Gone are the days of companies requiring extensive budgets and long lead times to build or buy-in large organisational IT solutions. The rapid evolution of cloud technologies and services (such as Google AppSheet and Apigee) makes it quick, easy and cost-effective to develop tools and software customised for the project in hand that can be deployed as needed as the initial project scales up.
Digital transformation is complex
These guidelines may seem overly simple for what is potentially a complex problem, but time and time again we see the results of them being omitted. The positive news is that effective AI and digital transformation is not out of reach; but it requires good business governance, process and disciplines, along with an ability to see the whole picture and understand why this is critical to the overall success of the organisation.