Project management: Are you backing the right AI projects?
- Steph Organ
- Archive , Blog
- October 22, 2019
- Updated:
Table of Contents
As an executive with an influence over whether your company implements AI and which projects it embarks on, there’s a lot of pressure on you to be successful. The future of AI within your company could rest on you on how your chosen projects perform.
The facts are that not all projects succeed, so even some of these carefully chosen projects will need to be shut down, but this can be hard to recognise by the project leader when they are so heavily invested in the project’s success. It is, however, vital to be able to determine when to pull the plug — in the spirit of Silicon Valley, failing fast is paramount so that success can be harvested elsewhere.
Research by McKinsey & Company found several companies taking a similar approach toward identifying failed projects by using some form of external judgement to guide resource allocation decisions — transparency in this process being key to avoiding political conflicts when dismantling a project.
The core idea among various companies was to change the burden of proof. Some did this by creating a job role to objectively hunt down failing assets. Others used a ranking system to assign projects as;
- Grow
- Maintain
- Dispose
When a project is noted as underperforming or lands in the "dispose" category, it is then down to the project leader to demonstrate potential project revival, or accept the ruling.
This approach eliminates the emotional attachment to projects and the bias that is often felt towards the loss of the resources that have been poured in thus far and instead brings focus around to whether or not the asset can be profitably reformed.
We believe that it is important to create a process like this within your organisation to make it easy to review whether funding is being sensibly allocated and to be able to identify when to walk away from a project.
FAQs
What is the failure rate problem in AI projects?
Industry surveys consistently show that a significant proportion of enterprise AI projects fail to deliver their intended value. The failure modes are well-documented: unclear objectives, poor data quality, lack of internal expertise, insufficient buy-in from the people who need to use the output, and the tendency to define success in technical terms rather than business terms.
For SMEs and mid-market manufacturers, these failure modes are even more acute. With limited IT resource, limited internal data science expertise, and limited tolerance for projects that consume budget without delivering results, the cost of backing the wrong AI project is high — not just financially, but in terms of the organisational appetite for future AI investment.
What are the characteristics of the right AI projects?
The right AI projects for manufacturers share common characteristics: they address a clearly defined operational problem, they have access to the data needed to produce reliable outputs, they have a clearly identifiable user who needs the result, and the ROI case is straightforward to calculate.
Cutting optimisation is a good example. The problem is clear: minimise material waste when cutting bar, plate, or sheet to order specifications. The data exists: order specifications and available stock. The user is identifiable: the production planner or machine operator. The ROI calculation is direct: time saved in planning plus material saved in cutting.
Contrast this with more speculative AI projects — predicting demand six months out, optimising a supply chain with dozens of variables, or detecting anomalies in a process with limited historical data. These can be valuable, but they are harder to scope, harder to execute, and harder to validate.
