Though there are actually five phases to a construction project – concept, planning, execution, performance monitoring and closeout – in reality, most projects consider just two: planning and execution. People theorize a lot about the value of lessons learned and archiving during closeout, but the truth is, this rarely gets done on real projects. I think the reason for it is that, to date, there has been little benefit associated with re-using lessons-learned and historical as-built performance and worse, capturing this in a meaningful way is really hard to understand for future project success. Therefore, closeout has largely been seen as a ‘check the box’ type exercise if it’s even done at all.
That then poses a bit of a problem when it comes to recycling through the ‘concept’ and ‘planning’ phases on new projects as we tend to then start from scratch each and every time. How often do you waste valuable time searching for and then trying to understand the context of a previous project?
For those who believe projects are so unique that we can’t really benefit from lessons learned or historical productivity rates from prior projects, I would challenge that while scope is often unique, the how do we build something aspect is highly repetitive across projects. It’s about time we did a better job of re-using our past project experiences to help forecast future project success.
The Benefits of Knowledge-Driven Planning
Many organizations are undergoing a ‘digital transformation,’ rethinking the way expertise and competencies could be captured and reused for the betterment of the organization and the projects they take on. The organizations that will come out ahead will find a pragmatic approach to applying new technologies in support of their digital initiatives.
We have been developing an AI Planning Assistant and home/repository for project knowledge. That is, a solution that assists in the development of project schedules and cost estimates. At a high level, this program makes informed suggestions to the planner as to what the duration, costs and sequence of work should be based on the context of what you are building. Tell the program you are building a commercial office building and it will come back with the scope, timelines, activities and costs typically associated with building such a project. This all sounds like magic, right? Actually, it’s just very clever computing using a machine learning AI technique. It gets its ‘knowledge’ by understanding what happened in the past on previous projects.
Project Knowledge Types
Today, the program stores multiple sources of project information in a knowledge library. It uses these multiple sources to then make informed suggestions. The more sources you interrogate, the higher the chance your research is going to be sound. The same is true when it comes to interrogating the knowledge library. Sources of stored knowledge include:
- Historical as-built CPM schedules. Very powerful for helping with the development of high level (levels 1-3) timelines as well as templating and submitting a detailed CPM schedule
- Project cost estimates. Typically, deliverable and quantity-based
- Project risk registers. Useful for tracking common project bottlenecks and problem areas
- Productivity rate tables. Ideal for building up detailed plans based on quantities, rates and crew sizes.
Today, more and more organizations are turning to the program to help them build better project schedules that are highly realistic and achievable with a lower degree of risk than they would have otherwise carried using a traditional CPM-type scheduling program alone. The real challenge though has been feeding valid ‘project knowledge’ into the program to make it smart enough to then give useful suggestions. There is, however, a remedy for this.Â
Knowledge Cleansing
Historical as-built schedules, of course, tell us a lot about how we performed on a project, but they don’t always represent the norm in terms of typical timelines and estimates. The historical project in question may have incurred delays due to unforeseen risks or poorly defined scope, for example.
To overcome this, we have developed a technique called Knowledge Cleansing enabling you to take your historical as-built schedules and load them into your knowledge library with the confidence that they will be used appropriately when making suggestions.
So how do we do this? The program carries out a cleansing technique that adjusts and normalizes the durations and costs associated with the historical information captured in the knowledge library. The cleverness of this normalization is that you, as the owner of the knowledge, get to define what factors you want to consider as part of the normalization.
- Did geographical location impact productivity?
- Were your construction durations longer than expected due to specific risks?
- Was engineering over budget due to poorly defined scope in specific areas of the project?
By identifying these factors and then tagging the relevant parts of your project(s) with these, the program is able to normalize the knowledge down to a more typical or standard suggestion. For example, if our fabrication was carried out in China, then the costs would perhaps be 20% less than the standard cost rates but the durations might be 10% greater due to lack of expertise. By enhancing the knowledge library with these multiple normalization factors, the suggestions that are then made by the program are automatically adjusted accounting for these discrepancies.
The net result is that you don’t have to spend time and money sanitizing the data that you load into your knowledge library. This cleansing is done in the program itself.
One Size Really Does Fit All
In a similar way to CPM schedules being cleansed in the knowledge library using the factoring approach described above, we can also normalize suggestions based upon attributes other than project performance.
By providing context you can factor by attributes such as project location, type of sub-contractor, execution location, the complexity of the scope, uniqueness of the project, etc., the program can make the necessary adjustments to its suggestions. For example, say the standard pipelay productivity rate is $30K per linear km but based on the fact the project is in Northern Alberta, this rate increases by 10%. In addition, if it takes one day to lay 3 km of pipeline, that productivity rate can be factored based on the location of Northern Alberta and adjusted accordingly when suggestions are made to the planner.
By adopting a standard productivity rate table and then applying influencers or factors, it eliminates the need for and complexity of having to store multiple rate tables for all possible combinations of location, complexity, or other attributes.
Learning from Mistakes
While we believe this AI-driven approach to project planning is the smart way to build project plans and cost estimates, we are not pretending that it is perfect. It is not, but it does have the capacity to learn and continuously improve.Â
In the case where the program makes an erroneous suggestion that perhaps isn’t relevant to the scope in question, then it can learn from its mistake. When the planner or cost estimator tells the program that the suggestion was bad, it learns from this so that next time it doesn’t make the same mistake again. This machine-learning approach to planning is one of the first times we are actually not only taking on board suggestions from stored knowledge but also finessing that knowledge as we move forward. This is a powerful concept to achieve project success in more cases, more often.Â
Though closeout has largely been seen as a ‘check the box’ type exercise in the past, we can now see that the how do we build something aspect is truly repetitive across projects. With the benefits of knowledge-driven planning, we can do a much better job of re-using our past project experiences to help forecast more future project successes, more often. Request a demo of InEight Software and see how we can make a positive difference in your construction project performance.
