Everywhere you look you see AI. And increasingly AI is being touted as the next great thing in project management. Vendors are progressively adding AI into their existing PPM tools. Often the vendor takes the simplest path which involves the flavour of AI that they are most familiar with. This is generally some form of Generative AI (think ChatGPT). But is this the best use of AI for the job? And is this the correct type of AI for the job? In this article we will explore the different types of AI as they relate to project management and provide a heatmap of what is hot and useful, and what is not the best.
Before we get into the different types of AI, we need to ask: “Why use AI in project management?”. Can’t we just use dashboard and reports?
AI versus dashboards
The reason for this article is because many people are applying AI where it isn’t adding value. Which brings us to a fundamental question: Do you need AI, or do you just need a dashboard or report? And why can’t you use a dashboard or report instead of AI?
The answer to the question is largely that a dashboard is static and embeds the rules you added to it at the time you created it. This is a process driven approach. The problem with it is that you need to update the dashboard as the process changes, and you need to make sure you thought of all the rules.
AI takes a data driven approach. AI will create the rules based on what it finds in your data. And AI will update itself if you train it with more data. AI can find relationships in data that you possibly can’t.
The following table will help you understand the difference between dashboards and AI. Then we will look at AI use cases.
| Dimension | Dashboards/Reports | AI |
| Purpose | Display and summarize existing data for monitoring and reporting. | Analyse, predict, and optimise outcomes, potentially generating new insights and recommendations. |
| Complexity of Insights Required | Low to moderate. Typically used for tracking KPIs, milestones, and status updates. | High. Suitable for complex problem-solving, scenario analysis, and identifying patterns not easily visible. |
| Data Dynamics | Static or near real-time. Reflects the current state or historical data. | Dynamic and adaptive. Capable of processing real-time data and learning from new inputs. |
| Actionability | Provides information for manual decision-making and human interpretation. | Directly supports automated decision-making, offering prescriptive recommendations or automated actions. |
| User Interaction | User-driven, so the user must run the dashboard. | AI-driven interaction, where the system might proactively suggest actions or surface hidden insights. |
| Scalability of Use Cases | Suitable for standard use cases with predefined metrics and goals. | Ideal for scalable, complex, and evolving scenarios where continuous learning and adaptation are needed. |
| Required Data Structure | Requires well-structured, clean data, typically in tabular formats. | Can work with unstructured, semi-structured, and structured data; also capable of handling data variability. |
| Customisation | High degree of customisation for specific metrics and visualisations. | Customisable through training, tuning, and model selection; requires expertise to tailor effectively. |
| Predictive Capabilities | Limited to trend lines and basic projections. | Strong predictive capabilities using advanced algorithms to forecast future outcomes based on historical and real-time data. |
| Adaptability to Changes | Less adaptable; requires manual updates to reflect new data or changes in metrics. | Highly adaptable; can learn from new data and adjust predictions or recommendations over time. |
| Resource Requirements | Lower technical expertise required for setup and maintenance. | Higher expertise required for setup, training, and maintenance; may also require more computational resources. |
| Decision Timeframe | Best for short-term, real-time monitoring. | Suitable for long-term planning and strategic decision-making where predictions and optimisations are needed. |
Types of use cases
There are many articles listing out potential use cases for AI, but these are often a dump of random ideas. As there are many approaches to AI it is important to think about the different types of AI, and what type of use case is relevant.
Using natural language as an interface (Generative AI)
Using a language model in a ChatGPT type interface to interact with your project data in your PPM solution, such as to query which projects should be focussed on, is often the first step in using AI in a PPM tool. We added this as a solution in Altus to give the PMO and management a simple and quick way to gain insights into the portfolio without needing to go to a report, or to individually approach all the project managers. This is about a natural language interface and is made possible by language models with integration to Altus.
Using natural language to assist with experience and knowledge (Generative AI)
Continuing with natural language is the use of the language model for its content and knowledge, not only its ability to understand natural language. Project managers are often coming and going due to contracts, and no one can remember the lessons learned from years back. Sensei has created an AI solution on the Altus PPM platform that reviews your old projects and provides advice on past lessons learned and benefits achieved that you should consider when planning new work. In this way we can use AI to encapsulate past knowledge, providing an accessible and intuitive interface that reduces the need to try and seek this knowledge from the current team members who may not know.
Augmenting GenAI with your own knowledge for help and training
Not all your data and knowledge is in your PPM tool. GenAI can also use your other files and documents. At Sensei we have created an AI that assists users with all relevant information within and outside the Altus platform and guides them through your own Altus processes. This brings people up to speed quickly and helps them to use Altus efficiently without having to draw heavily on other team members.
Prediction and decision support based on past knowledge (machine learning)
Not everything is language based. Before generative AI and language models made AI so high profile, we had machine learning. There are many types of machine learning, but we won’t go into the details here. For the purpose of this article, machine learning works by feeding data into an algorithm that creates a formula to explain the data. So, you can feed it past variances on tasks, and it will learn from that to create a prediction of new variances. Typical use cases include predicting project variance, cost estimation, resource variance, etc.
Rules
Often, I get asked about AI but what the person really wants is for some rules and logic to be added to their PPM solution to enable a process. Rules can be used for example to approve work, changes or resources. Where you want to apply specific rules, you need workflow and maybe a rules engine but not usually AI, and definitely not GenAI. AI might be useful if you want to learn from a human which rules apply when, as opposed to coding the rules. The trap here is that you have not kept your PPM solution aligned with your processes and now are using AI to close the gap. Only use AI where you need to apply learning that cannot be coded into rules. And watch out for people calling rules AI.
Optimisation (algorithms and machine learning)
Working out how to effectively use your limited resources, people and money, to do the work you need to do to achieve your strategic goals is what bridging the strategy-execution gap is all about. AI can definitely help with this.
At Sensei we have built a model to give you the most efficient portfolio of projects within a budget, based on the benefits of those projects. In this case you want to use a mathematical model, and this will require numerical input and a specific algorithm to be selected.
This is not typically true machine learning as it optimises once based on constraints, without learning from the past. However, where this becomes powerful is if you do create an ensemble of models which include optimisation based on past performance. This is not a GenAI model.
Using the matrix to help you find your AI
To help make sense of the above types of AI and to find the most appropriate use case, we have developed the following matrix. The strength of each approach to solve that use case is rated as high, medium or low. This is not an exhaustive list of use cases, but it will help you understand the type of functionality in relation to the type of AI available.
| Use Case Types | Natural language | Machine learning | Visualisation & dashboards | Rules | Optimisation |
| Querying a number of projects using natural language | High | Low | Medium | Low | Low |
| Assisting in creating risks and issues | High | Low | Low | Low | Low |
| Adding experience to creating risks, benefits etc. | High | Medium | Low | Low | Low |
| Prediction of costs, variance etc. | Low | High | Medium | Low | Low |
| Selecting optimal projects for a budget or resources | Low | Medium | Low | Low | High |
| Generating tasks | High | Low | Low | Low | Low |
| Creating status reports | High | High | Low | Low | Low |
Rules of thumb
As you think about applying AI, or you view AI based solutions, here are some heuristics to guide you past the sales hype:
- Horses for courses. Not all AI is equal, and GenAI does not equal all of AI.
- Don’t use GenAI on knowledge that doesn’t exist in the world. If you want to predict the variance on projects, this is not based on knowledge that a language model would have. It is based on your own data. Therefore, GenAI may not be appropriate.
- If you feed in numbers and expect numbers to come out, you need a mathematical AI. This could be machine learning and is not likely to be GenAI. Remember just because everyone is finding GenAI easy doesn’t mean it equals all of AI.
- If you are using machine learning it must learn from something to be relevant, and that needs to include your own data. Don’t believe the sales hype, ensure there is time and the ability to train on your data. And make sure it still works then too.
- If you don’t prepare your data, don’t expect slapping AI on top of it to work. 90% of your time should be spent cleaning and structuring your data. This is even important with text.
Don’t forget the whole system and the people
Task versus systems thinking
AI can either be bolted on, or part of the overall process. For portfolio and project management, you need to think of the PPM solution, the process and the users. This makes up the total system. Many tools are simply adding a chat panel on the side. However, that is something a user must choose to interact with. This is task thinking and the AI serves to support one specific task only, and only if a user bothers to go to it. It isn’t part of the process in a seamless manner.
Systems thinking makes the AI part of the overall process. For example, when I enter a risk, Altus can automatically provide me with information on past risks and lessons learned. Systems thinking goes further than this though and is really about focussing on the user. How will the user receive the output of the AI? For example, we use AI to predict a potential failure of a project. When this is shown as a red light on a dashboard it can be confronting as trust has to be built with the AI. Rather, we need to ensure that we use AI to augment the user and give them a chance to drill into the details and make up their own mind.
If you would like to discuss how to practically apply AI across your portfolio and project management work or you have a problem you think could be solved with AI please reach out to us. Sensei is investing time and resource into harnessing AI in the project and portfolio management space and would love to talk to you.
Get in touch to find out more.
