AI to assist projects has been around for decades – why aren’t we using it?

I recently read a fascinating article entitled ‘Using Artificial Intelligence Techniques to Support Project Management’. It discusses AI-assisted decision-making on projects, knowledge-based interactive graphics and hybrid computer-human systems. It sits well with the large number of articles and reports that both APM and IPMA have been sponsoring recently, and with the launch of the Project Data Analytics Taskforce. There is only one problem – this article was written in 1987.

The author was engineering project management guru Ray Leavitt. And while Professor Leavitt has always been a far-sighted individual, his article points to the uncomfortable fact that we have been able to use AI to assist us in managing projects for decades. The real puzzle is why it has not been more widely adopted.

So, what’s the excuse?

We can’t argue that technology has only recently developed to at last make it feasible. Data to conduct this type of activity has been hanging around in Primavera and related ERP systems for decades. And we don’t need gargantuan amounts of computing power for these calculations. Data at the level of a project is not ‘big data’. There may be millions of data points within a project, but I would be really surprised if you pointed out a programme made up of more than 50,000 projects. 50,000 of anything is not big data.

So, responsibility for the failure to exploit AI in project management must lie elsewhere. A recent investigation that I undertook with colleagues at Sheffield, Manchester and Southampton universities, sponsored by APM, may help explain this conundrum. The study is called Project Data Analytics: The state of the art and science. We began by asking a wide range of project practitioners two simple questions:

• What are you doing with project data analytics now?
• What will you be doing next?

We used the answers to these questions to create a summary of where we are with project data analytics and, importantly, the barriers that are preventing its wider uptake. In analysing the results of our interviews, we identified that we needed to establish some succinct definitions in order to discuss results between ourselves and our interviewees.

Three key definitions

Project data analytics is the use of past and current project data to enable effective decisions on project delivery. This includes descriptive analytics (presenting data in the most effective format) and predictive analytics (using past data to predict future performance).

Machine learning is the name given to computer algorithms that ‘learn from doing’. In project terms, machine learning has, at its centre, algorithms that are used to spot patterns between some characteristic of projects or programmes and some aspect of project performance. This process gets more accurate the more it is used. Machine learning is a fundamental part of predictive project data analytics.

Artificial intelligence refers to the study of ‘intelligent agents’, autonomous non-human entities that can take in information from their environment and act upon their environment in a way that enables them to succeed in their goals. Intelligent agents need to have mastered machine learning and aspects of predictive data analytics. In a project context, some people have speculated that an intelligent agent could enhance or change the roles and status of many project professionals

We summarised our findings in a 2x2 matrix.

This highlighted several issues that explain the slow uptake of AI and project data analytics by the project management community.

The ‘geeks meet business’ divide

For most project practitioners, ‘python’ is a big snake and not a computer language for machine learning applications. Many project practitioners’ eyes glaze over when ‘geeks’ start talking in terms of ‘random tree forest algorithms’ or ‘data cleansing’. Until we can find a clear, unambiguous way to talk about these technologies, we are not going to be able to take advantage of what they have to offer.

Attitudes to data

The age-old GIGO (garbage in, garbage out) rule applies. If the data disciplines associated with our use of project control systems lead to poor data, then we can learn absolutely nothing by analysing that data. A first step in any wider use of project data analytics must be better data discipline, and this needs to start immediately.

Basic statistical challenges

APM defines a project as “a unique, transient endeavour undertaken to achieve planned objectives”.

If projects were truly unique, there would be no point in trying to ‘pattern spot’ across projects, which is what predictive project data analytics is trying to do. Given that we accept there are degrees of ‘uniqueness’, a very important question is: what sample of projects is it appropriate to compare other projects against? This problem of ‘reference class’ underlies any predictive capability and is one that has not yet been solved satisfactorily. Practitioners are still trying to establish a mechanism for comparing ‘apples with pears’.

What data do we really need to make better decisions?

Finally, and arguably most importantly, practitioners still do not know what data is required to make better decisions in managing projects. This fundamental question remains a mystery. Project data analytics may yield some incredible insights, but it can only do this by using the data that is available. When it comes to identifying what data is needed to make better project decisions, we are definitely in the territory of ‘unknown unknowns.’

Perhaps, given all of the above, it is no surprise that Ray Leavitt’s expectation of AI’s use in projects still hasn’t been realised over 30 years later. If we are serious in our intent to use AI and predictive analytics in projects, then we must find a solution to the problems outlined in this article.

Download the paper Project Data Analytics: The state of the art and science here.

_{Image: Phonlamai Photo/Shutterstock.com and Naomi Brookes}