Good project management depends on good decisions but those decisions often have to be made with imperfect information. Project managers routinely have to balance competing priorities, assess uncertain risks, interpret large amounts of information and decide when the available evidence is sufficient to act. They also have to compromise (and negotiate such compromises) because no project has unlimited time and money to do everything perfectly.
Artificial intelligence has considerable potential to help with this complex decision-making. Its ability to process large volumes of data, identify patterns and analyse multiple variables means it can provide project managers with insights that would be difficult, or extremely time-consuming, to uncover manually.
However, there is an important distinction between supporting a decision and actually making one.
AI may be able to identify that a project is becoming increasingly likely to miss a deadline, for example. What it cannot necessarily determine is whether delaying delivery is preferable to reducing scope, increasing costs or asking an already overstretched team to work harder. Choices such as these require an understanding of the context in which the project is being implemented, the relationships between stakeholders and team members and the likely human consequences of making one decision in preference to another.
Where AI tools can really help in a project management environment is gathering fuller evidence from similar projects to help the project manager make better decisions rather than actually making decisions.
Why AI can improve project decision-making
Projects generate considerable amounts of information: schedules, budgets, resource data, risk registers, progress reports, meeting notes, change requests and performance data, to name but a few. The difficulty is not necessarily obtaining information but in turning that information into something useful.
AI can help by processing data at a scale and speed that would be difficult for an individual project manager. This has several potential advantages.
Faster analysis is perhaps the most obvious. AI systems can examine substantial datasets quickly, allowing emerging patterns or anomalies to be identified earlier. On a complex project involving hundreds of activities, for example, AI might identify delays that imply a future scheduling problem before that problem becomes immediately apparent.
There is also the potential for better use of historical data. Organisations often possess information from previous projects but make surprisingly little use of it. AI can potentially analyse previous schedules, costs, risks and outcomes to identify patterns that may be relevant to a current project. It can essentially provide the ideal “Lessons Learned” process that so few projects actually deliver.
This could also improve risk management instead of relying on the risks that the project team think of during the initial risk assessment. AI can very quickly compare the current project with previous projects and suggest risks or dependencies that deserve further investigation. That does, of course, depend on risks being fully documented for all projects.
Another advantage of AI is gathering information, comparing figures, summarising reports and identifying exceptions, which consume valuable management time. If AI performs more of this type of work, project managers can devote more attention to interpreting the findings and deciding what should be done.
But is AI really objective?
Plenty of times I’ve heard the argument that AI is not biased in the way humans often are, and so can make less biased decisions, but I, personally, would be cautious about trusting this assumption. It’s true that AI tools do not have long-standing personal loyalties or are involved in office politics – which can enable it to provide an alternative perspective – but that doesn’t mean its conclusions are entirely objective. AI systems have been trained on existing data, including opinions, so these are inherent in the content it creates. Historical project data might, for instance, contain inconsistencies and biased views originally created by a human. Consider an organisation that has repeatedly underestimated a certain risk and not adequately recorded it’s repeated occurrence. In such a situation the project data on which an AI system was trained does not accurately reflect reality.
There is another difficulty: correlation is not the same as explanation.
AI systems are able to identify certain factors on a project that might suggest the project could overrun on its schedule. That is very valuable information, but the project manager needs to assess whether those factors are actually causing delays on the current project or whether other factors are the real problems. This is why we should not treat answers from AI systems as definitive but instead as possible evidence that could lead to a possible solution i.e. with caution!
Predictive analytics: identifying issues early in the lifecycle
Predictive analytics is one of the most promising applications of AI in project management. By examining historical and current project data, AI can identify patterns associated with particular outcomes. It might detect that a combination of resource shortages, delayed approvals and increasing numbers of change requests is associated with a greater probability of schedule slippage. That doesn’t mean AI systems can predict the future.
Projects involve people, organisations and external factors, all of which can behave unpredictably. How predictive analysis can help a project manager is by providing information about the possibility of something occurring. So, as an example, a project manager will not identify a missed milestone only when the status of a task is updated by a team member, but could receive advance warning from AI that there is a chance the milestone will be missed. There is then the opportunity to intervene early on to keep the task on track.
AI and project risk management
Risk management is another part of project management where AI tools provide supporting evidence to identify problems early on. Traditional risk identification is reliant on the experience of the project team members who might recognise situations that have previously resulted in risks occurring. But if no one on the team makes the connection between a current issue and a potential risk then risks could be overlooked by the team. When using AI it can proactively examine data from previous projects and other relevant data sources and might identify recurring causes of, for example missed deadlines. The caveat here, as always, is whether such information was recorded from previous projects.
Risk management isn’t only about identifying risks but also deciding how to solve or mitigate the problem because there are always other factors to take into consideration. Take as just one example a technically correct solution that is very expensive to implement. This cost might be prohibitive to the stakeholders so the project manager still has to negotiate possible compromises.
Natural language processing can unlock unstructured project information
Not all useful project information sits neatly in a spreadsheet. A significant amount is contained in emails, meeting minutes, reports, lessons-learned documents, stakeholder feedback and other written material. Natural language processing (NLP) allows AI systems to analyse this type of information at scale. That creates some really interesting possibilities.
Consider a complex construction project with multiple workstreams. An AI system might be able to highlight concerns about a particular supplier or resource because issues have occurred in different teams, even though the issue may not have been added to the risk register.
Collectively, they could represent an emerging project risk so this ability to connect fragmented pieces of information could become one of AI’s most valuable contributions to project decision-making.
AI is a decision-support systems not a decision-maker
A good way to think about AI within project management is as a decision-support system – helping project managers make better-informed decisions.
Let’s take a project that is not running according to schedule. An AI system could model several possible scenarios, such as:
- maintain the current resources and accept a later completion date;
- add resources and estimate the additional cost;
- reduce scope and assess the effect on expected benefits;
- change the sequence of activities and identify the new risks created.
The project manager would then have a clearer understanding of the available choices, but AI cannot necessarily decide which choice is right.
A delayed completion might have serious contractual consequences. Reducing scope might undermine the business case. Adding resources might be theoretically possible but practically impossible because the required specialists are unavailable. A client relationship might make one option politically unacceptable.
These are not data problems – they require context, judgement and an understanding of people.
The danger of automating bad decisions
There is also a less comfortable side to AI-assisted decision-making. The speed of analysing data is only useful when the underlying decision is sound. If an organisation has a history of weak governance so that lessons learned are never recorded fully from past projects, then using AI to analyse poor data runs the risk of exacerbating poor decision-making by the project managers. To counteract this the human scrutiny of AI outputs needs to be even more thorough.
Project managers need to understand where information has come from, question surprising recommendations and recognise when a situation contains factors that the AI system may not adequately understand. The ability to challenge an AI-generated recommendation could therefore become just as important as the ability to use AI tools in the first place.
Better decisions through human–AI collaboration
What this all means is that the best use of AI in project management is to use it for the tasks that humans are slow at doing (e.g. analysing large datasets) and focus human effort where people are better than AI – leadership, negotiation, conflict resolution etc. For example, a project manager faced with an unpopular compromise to be negotiated with a senior stakeholder needs far more than data to achieve this successfully.
This distinction suggests a useful division of labour. AI systems can look for patterns, predict what might happen next and assess what options are available. Then the project manager can determine what this means in a particular situation, and what to actually do.
A good way to think about AI in project management is not as an alternative to project professionals, but as a tool that gives those professionals more information to work with.
