Program Manager| HARPER SRLSanto Domingo / Distrito Nacional, Dominican Republic
Predictive AI tools promise accuracy, yet they rely on imperfect data and assumptions. Could this lead to overconfidence in forecasts, or can PMs use AI responsibly to improve, not replace, judgment?
Saving Changes...
Sort By:
Luis BrancoCEO| Business Insight, Consultores de Gestão, LdªCarcavelos, Lisboa, Portugal
AI can sharpen scheduling by reducing noise and exposing patterns, but it cannot eliminate uncertainty.
The real risk is mistaking precision for truth.
Forecasts built on incomplete or biased data demand more judgment, not less.
Used responsibly, AI expands scenario thinking, challenges assumptions, and strengthens decision quality.
Used uncritically, it creates an illusion of control.
In the end, value comes not from the model, but from the project leader who interprets it with awareness, discipline, and ethical clarity
Saving Changes...
Sergio Luis ConteHelping to create solutions for everyone| Worldwide based OrganizationsBuenos Aires, Argentina
Predictive AI tools do not promise accuracy. Outcomes in AI are always probabilistic and they are delivered with the associated probability. Human being decides based on that. Saving Changes...
Program Manager| HARPER SRLSanto Domingo / Distrito Nacional, Dominican Republic
Thank you all for the thoughtful insights, I really appreciate the different angles you brought. This confirms what I suspected when I asked the question: AI can strengthen forecasting, but it doesn’t replace uncertainty, or judgment. Saving Changes...
Jennifer TouchtonTechnical PM - PPM & Digital Transformation| Gallagher Benefits ServicesLouisville, KY, United States
In these scenarios, I recommend thinking of the AI tools as an SME. Their experience and knowledge is whatever you give them. Even the best SME can't promise 100% accurate predictions. Saving Changes...
My understanding is that AI systems are built on probabilistic principles. Probabilistic systems handle uncertainty by expressing outcomes as likelihoods rather than fixed results - something that can be done with traditional statistics alone. However, artificial intelligence goes a step further: it uses probabilistic methods to interpret context, learn from data, and make judgments in ways that resemble human cognition.
a) In a probabilistic system, the same input can produce different outputs, and the outcomes are described in terms of probabilities.
b) In a deterministic system, the same input will always produce the same output.
A probabilistic system can generate many possible outcomes, but it often cannot judge, e.g., context or ethical implications. Humans are therefore needed to design the system, interpret uncertain outputs, validate the probabilities, and provide domain expertise.
AI tools can improve accuracy, but with limited precision. I'll try not to use too much statistics terminology, so a simple analogy is that accuracy is shooting a bunch of arrows at a target, and they all hit the target but may be widely spread apart. Presion is the arrows landing close together but perhaps far from the target itself.
AI can sort through a lot of data to calculate things like what is the average time to complete certain tasks as well as the range (e.g. between 10-20 days) and provide probabilistic estimates for individual tasks and large sequences of tasks combined. It can also tell us the uncertainty in those estimates, and what are the most important variables in the estimates. Typically 1 or 2 items have the biggest overall impact. If most tasks are 1 day +/- 1 day, then on average some will come out low and others high but mostly the errors cancel each other out. If you add in one task that is 10-20 days it tells you that is the task has the widest range of probable outcomes that will be the most significant to the overall accuracy of your schedule and needs the most visibility.
Sometimes people get fooled thinking those estimates are accurate, when perhaps they're only precise. With some manufacturing products, we knew that delivering 100 new products a year, the average labor was 10,000 hours so we could predict our total cost very well. The range was very broad though so there were often big surprises when one project went 3x over budget. Saving Changes...
We might need to add a new category for estimates. So, in addition to ROM, Budget, Definitive, and Baseline estimates, we might need something like MLE (Machine-Learned Estimate), either as a new category or appended to the other categories (ROMMMLE?). Then, just like the other estimates, it can be treated like a statement of fact by the people we provide estimates to.
With a large enough dataset including historical task & project data, resource availability & performance data, workload, WIP, contextual factors, project structure & constraints, external factors, real-time execution data, metadata, and classification data, Predictive AI / ML could probably provide more accurate schedules and forecasts, it would be able to synthesize all that data easier/faster than a human could, but it would still be based on known data and subject to change as new data is obtained. Saving Changes...
Community Champion
