Example of Machine Learning algorithm recognizing shingles. White crosses mean that square of 25X25 pixels is recognized as part of shingle roof with high probability. Developed by author, Dmitry Golubev, 2017.

It’s just slightly customized image recognition and recommendation engine applied to picture of house. You could see dozens of similar on-line services from Google, Netflix and Amazon, etc. They recognize, classify and recommend using standard ML technics.

Machine Learning is set of algorithms which analyse and classify data in accordance with some criteria or predict outcome based on some previous experience. In this case, we can apply ML to classify some parts of picture as roof, to recognize $5Million house and to predicts from previous statistics that new line of architectural shingles provides the best chance to get a deal.

What is the reason that roofing industry is still far from the functionality? There is no specific problem, and answer depends on person’s position and department.

I like the answer that it’s low in the list of strategic priorities. So, there are no immediate investments into people and technology. And for some, there is a false hope that, may be, one day, some vendor will bring everything ready to use.

But, why is competence in ML strategic?

First, ML routinely helps people to make right predictions in many business areas. It gives competitive advantage in reducing costs/ finding new opportunities and helps to grow expert knowledge in key business areas. It is relevant to sales, marketing, lunching new products, optimization of internal processes, quality control, IT, manufacturing process control, maintenance, etc.

Second, it takes several years to gather and process right data to get working ML with high precision. By nature, it is heavy R&D based on company own data. So, it’s impossible to repeat as almost every step should be re-invented based on totally different data set.

Third, once acquired, it can be successfully combined with other competitive advantages such as advanced quality control, process automation, unique marketing and sales strategies.

Can roofing industry use shared Big Data and Machine Learning servers/network infrastructure?

From my opinion and experience, for purpose of R&D and model training stage, there should be dedicated equipment which is not colliding with other infrastructure. One of practitioners made clear comparison that one GPU with 11GB Memory is equivalent of 4 servers with generic CPUs and tones of Memory.

Another consideration is that ML task running on Big Data cluster prevent other tasks from using most valuable resources. It’s true for Cloud solutions as well. Amazon, Data Bricks and others have specially designed types of cluster for algorithm training process.

It makes sense to use separate environment for ML and ad-hoc analysis from money saving and operational continuity perspective. After ML model is trained, it can be put to Big Data cluster as algorithm for anomaly detection, or finding next best investment. But, model training process should be separated from Big Data operational use.

All applications of ML can be roughly represented by next diagram connecting decision making process and company’s internals and externals.

What is Machine Learning (ML)? Why ML expertise is getting critical for roofing industry?

Before answering the questions, let’s check next example. Some of shingle vendors provide on-line function of “trying” different shingles on a picture of a potential buyer’s house. A client uploads picture, and with simple graphic tool sets a contour of the roof. After a bit painful activity, he/she can start changing colours and types of shingles.

So, what is wrong here? It’s 2017 and ML algorithms work pretty good. In modern scenario, after customer uploaded picture, it should appear with best shingles fitting to color of walls, light, landscape and season. And it should invite client to try several most viable recommendations. Of course, as everything based on algorithms, something can be a little bit off, and a client can correct results with couple clicks for better look.

Here is the example of working algorithm for the problem using Machine Learning approach.

Roofing Industry. Applications of Machine Learning

Previous Page    Next Page

Implementation of ML in roofing company requires huge cultural shift. Modern manufacturing is all about reducing costs as everyone produces similar products. It provides not much space for trying creative ideas and having separate laboratory for innovative initiatives. But, in time of pivotal changes, it’s going to be different. To survive under pressure of new technologies, company need to embrace and integrate innovations into portfolio of projects. Resolving the challenges is impossible without energetic and creative people; getting them in right time and in right combination should be another strategic priority.

Machine Learning is going to change several aspects of roofing industry: it’s going to be much technological for customers and contractors, and it’s going to be much more cost saving for manufacturer. It has only one draw back: to make it happened company need to gather a lot of very specific data, so specific that there should be solutions implemented before starting gathering the data. To understand what are these solutions, company should have a team knowing roofing company Technological and Data Architecture. Member of team and R&D Committee should have a vision about application of new technologies to specific business areas and understand requirements for new generation of Machine Learning algorithms.

Unfortunately, in this specific case, external consulting companies can sell only technology and software. It's just Lego building blocks to assemble the solutions. It means that future of roofing company depends on internal team of bright professionals having shared vision and dedication to developing unrepeatable competitive advantages very specific for roofing industry.

Original Image