IBM Watson won’t solve PLM platform problems

August 17, 2015


Last year, my attention was caught by CIMdata article – IBM Forms New Watson Group to Meet Growing Demand for Cognitive Innovations. The interesting for cognitive computing is growing these days and you can get tons of interesting materials about that on IBM Watson website.

Cognitive computing is the simulation of human thought processes in a computerized model. Cognitive computing involves self-learning systems that use data mining, pattern recognition and natural language processing to mimic the way the human brain works.

The following passage from CIMdata article caught my attention:

IBM Watson Analytics allows users to explore Big Data insights through visual representations, without the need for advanced analytics training. The service removes common impediments in the data discovery process, enabling business users to quickly and independently uncover new insights in their data. Guided by sophisticated analytics and a natural language interface, Watson Analytics automatically prepares the data, surfaces the most important relationships and presents the results in an easy to interpret interactive visual format.

Data discovery is a tricky topic. As I mentioned in my earlier blog last week – PLM cannot drain product data swamps. The problem of PLM is in fact related to limitations of data modeling and ability to capture large scales of organizational data. In a long run it limits ability to create an environment for product innovation. So, maybe IBM Watson is here to help?

Over the weekend, my attention was caught by The Platform article “The Real Trouble With Cognitive Computing” and the troubles IBM has trying to figure out what they are going to do with the Watson supercomputer. The article explains that IBM folks came up 8,000 potential experiments for Watson to do, but only 20 percent of them.

The discussion about single information model in Watson is something PLM folks can benefit when thinking about future of PLM platformization. Here is my favorite passage about Watson:

“The non-messy way to develop would be to create one big knowledge model, as with the semantic web, and have a neat way to query it,” Pesenti tells The Platform. “But that would not be flexible enough and not provide enough coverage. So we’re left with the messy way. Instead of taking data and structuring it in one place, it’s a matter of keeping data sources as they are—there is no silver bullet algorithm to use in this case either. All has to be combined, from natural language processing, machine learning, knowledge representation. And then meshed as some kind of distributed infrastructure.”

What is my conclusion? The odds are Watson won’t be a pragmatic technology for PLM vendors to rely on and build a future of PLM platform innovation. However, the giant knowledge model Watson failed to build can be an alert for PLM architects trying to create a holistic model of the future PLM platforms. It might not work… The reality is much messy than you think. This is a note to folks taking strategic decisions and PLM innovators. Just my thoughts…

Best, Oleg

picture credit IBM Watson



How PLM can ride big data trend in 2015

December 22, 2014


Few month ago, I shared the story of True & co – company actively experimenting and leveraging data science to improve design and customer experience. You can catch up by navigating on the following link – PLM and Big Data Driven Product Design. One of the most interesting pieces of True & Co experience I’ve learned was the ability to gather a massive amount of data about their customers and turn in into a information to improve product design process.

Earlie this week the article What’s next for big data prediction for 2015 caught my attention. I know… it is end of the year “prediction madness”. Nevertheless, I found the following passage interesting. It speaks about emerging trend of Information as a service. Read this.

The popularity of “as-a-Service” delivery models is only going to increase in the years ahead. On the heels of the success of software as a service models, I believe Information-as-a-Service (IaaS) or Expertise-as-a-Service delivery models are likely the next step in the evolution. The tutoring industry provides a good blueprint for how this might look. Unlike traditional IT contractors, tutors are not necessarily hired to accomplish any one specific task, but are instead paid for short periods of time to share expertise and information.

Now imagine a similar model within the context of data analytics. The shortfall most often discussed with regard to analytics is not in tooling but in expertise. In that sense, it’s not hard to imagine a world where companies express an interest in “renting” expertise from vendors. It could be in the form of human expertise, but it could also be in the form of algorithmic expertise, whereby analytics vendors develop delivery models through which companies rent algorithms for use and application within in their own applications. Regardless of what form it takes in terms of its actual delivery, the notion of information or expertise as a service is an inevitability, and 2015 might just be the year IT vendors start to embrace it.

It made me think about how PLM can shift a role from being only “documenting and managing data and processes” towards providing services to improve it by capturing and crunching large amount of data in organization. Let speak about product configurations – one of the most complicated element of engineering and manufacturing. Mass production model is a think in a past. We are moving towards mass customization. How manufacturing companies will be able to get down cost of products and keep up with a demand for mass customization? Intelligent PLM analytics as a service will be able to help here.

What is my conclusion? Data is a new oil. Whoever will have an access to a most accurate data will have a power to optimize processes, cut cost and deliver product faster. PLM companies should take a note and think how to move from “documenting” data about design and processes towards analytical application and actionable data. Just my thoughts…

Best, Oleg

PLM: from sync to link

October 17, 2014


Data has an important place in our life. Shopping lists, calendars, emails, websites, family photos, trip videos, documents, etc. We want our data to be well organized and easy to find. Marketing folks like to use the term – data at your fingertips. However, the reality is just opposite. Data is messy. We store it in multiple places, we forget names of documents and we can hardly control it.

Everything I said above applies to manufacturing companies too. But, it gets even more complicated. Departments, contractors, suppliers, multiple locations and multiple systems. So, data lives in silos – databases, network drives, databases, multiple enterprise systems. In my article – PLM One Big Silo, I’ve been talking about organizational and application silos. The data landscape in every manufacturing company is very complex. Software vendors are trying to crush silos by introducing large platforms that can help to integrate and connect information. It takes time and huge cost to implement such system in a real world organization. Which makes it almost a dream for many companies.

In my view, openness will play a key role in the ability of system to integrate and interconnect. It will help to get access to information across the silos and it leads to one of the key problem of data sharing and identity. To manage data in silos is a complex tasks. It takes time to organize data, to figure out how to interconnect data, organize data reporting and to support data consistency. I covered it more in my PLM implementations: nuts and bolts of data silos article.

Joe Barkai’s article Design Reuse: Reusing vs. Cloning and Owning speaks about the problem of data re-use. In my view, data reuse problem is real and connected directly to the issue of data silos. I liked the following passage from Joe’s article:

If commonly used and shared parts and subsystems carry separate identities, then the ability to share lifecycle information across products and with suppliers is highly diminished, especially when products are in different phases of their lifecycle. In fact, the value of knowledge sharing can be greater when it’s done out of sync with lifecycle phase. Imagine, for example, the value of knowing the manufacturing ramp up experience of a subsystem and the engineering change orders (ECOs) that have been implemented to correct them before a new design is frozen. In an organization that practices “cloning and owning”, it’s highly likely that this kind of knowledge is common knowledge and is available outside that product line.

An effective design reuse strategy must be built upon a centralized repository of reusable objects. Each object—a part, a design, a best practice—should be associated with its lifecycle experience: quality reports, ECOs, supplier incoming inspections, reliability, warranty claims, and all other representations of organizational knowledge that is conducive and critical to making better design, manufacturing and service related decisions.

Unfortunately, the way most of companies and software vendors are solving this problem today is just data sync. Yes, data is syncing between multiple systems. Brutally. Without thinking multiple times. In the race to control information, software vendors and implementing companies are batch-syncing data between multiple databases and applications. Parts, bill of materials, documents, specifications, etc. Data is moving from engineering applications to manufacturing databases back and forth. Specifications and design information is syncing between OEM controlled databases and suppliers’ systems. This data synchronization is leading to lot of inefficiency and complexity.

It must be a better way to handle information. To allow efficient data reuse, we need to think more about how to link data together and not synchronize it between applications and databases. This is not a simple task. Industry that years was taking "sync" as a universal way to solve problem of data integration cannot shift overnight and work differently. But here is a good news. For the last two decades, web companies accumulated lot of experience related to management of huge volumes of interconnected data. The move towards cloud services is creating an opportunity to work with data differently. It will provide new technologies of data integration and data management. It also can open new ways to access data across silos. As a system that manage product data, PLM can introduce a new way of linking information and help to reuse data between applications.

What is my conclusion? There is an opportunity to move from sync to link of data. It will allow to simplify data management and will help to reuse data. It requires conceptual rethink of how problems of data integrations are solved between vendors. By providing "link to data" instead of actually "syncing data", we can help company to streamline processes and improve quality of products. Just my thoughts…

Best, Oleg

IoT data will blow up traditional PLM databases

September 23, 2014


IoT is one of the recent popular buzzwords in business and consumer world. I’ve been touching it in few of my previous posts – 3 Things PLM can do with IoT tomorrow and PLM and IoT common trajectories . Things changed from 2009 when I first time touched PLM and Internet of Things. The opportunity to connect huge amount of devices to the internet looks like a reality we are going to face tomorrow. Gartner’s Hype Cycle Special Report for 2014 report put Internet of Things on the top of the curve indicating – Peak of Inflated Expectations.


So, what can we do about that? My attention was caught by Joe Barkai’s article – The Emergence of Application-Specific IoT. Read the article and draw your opinion. My favorite passage from the article was about "Content and IoT" – a tremendous value of content coming from connected devices.

The business potential is not in the conduit, or the “plumbing” of the IoT; it is in the content. However, the data streams of disparate devices cannot be simply connected. Nor can the portfolio of dissimilar applications communicating with these devices be simply daisy-chained as depicted by those futuristic scenarios. Data interoperability is critical to harvest the potential value of the IoT’s content and to enable new meaningful business models that utilize its data. That means not only compatible and interoperable data protocols, but also, more critically, data models and common semantics, so that disparate devices and services can be linked, aggregated and harmonized to form an IoT solution.

The last thing about linking, aggregating and harmonizing information coming from connected devices contains huge potential for design and product lifecycle management. We can certainly dream about how to bring all information into PLM applications. However, here is the thing… Will existing PLM data architecture is capable to handle such data scale? Remember my "Will PLM data reach yottabytes?" article.


I want to come back to my thoughts about future of data management and PLM. Existing data architectures were developed back 10-20 years ago. It looks like vendors will have to re-tool PLM platforms with new technologies and capabilities coming from a modern PLM data management landscape.

What is my conclusion? IoT is going to change the scale of data that modern product development and manufacturing companies are consuming. It is not about managing of documents or even bill of materials. The tremendous amount of data will be collected from connected devices and transformed into consumable information assets. I don’t think somebody can predict where this process can take us today. However, it is coming, for sure. It is an alarm for PLM architects, data scientists and IT managers in manufacturing companies. Just my thoughts…

Best, Oleg

How to visualize future PLM data?

August 12, 2014

collective experience of empathetic data systems

I have a special passion for data and data visualization. We do it every day in our life. Simple data, complex data, fast data, contextual data… These days, we are surrounded by data as never before. Think about typical engineer 50-60 years ago. Blueprints, some physical models… Not much information. Nowadays the situation is completely different. Multiple design and engineering data, historical data about product use, history of design revisions, social information, data about how product is performing coming in real time from sensors, etc. Our ability to discover and use data becomes very important.

The ways we present data for decision making can influence a lot and change our ability to design in context of right data. To present data for engineers and designers these days can become as important as presenting right information to airplane pilots before. Five years ago, I posted about Visual Search Engines on 3D perspective blog. I found the article is still alive. Navigate your browser here to have a read. What I liked in the idea of visual search is to present information in the way people can easy understand.

Few days ago, my attention was caught by TechCrunch article about Collective Experience of Empathetic Data Systems (CEEDS) project developed in Europe.

[The project ]… involves a consortium of 16 different research partners across nine European countries: Finland, France, Germany, Greece, Hungary, Italy, Spain, the Netherlands and the UK. The “immersive multi-modal environment” where the data sets are displayed, as pictured above — called an eXperience Induction Machine (XIM) — is located at Pompeu Fabra University, Barcelona.

Read the article, watch video and draw your conclusion. It made me think about the potential of data visualization for design. Here is my favorite passage from the article explaining the approach:

“We are integrating virtual reality and mixed reality platforms to allow us to screen information in an immersive way. We also have systems to help us extract information from these platforms. We use tracking systems to understand how a person moves within a given space. We also have various physiological sensors (heart rate, breathing etc.) that capture signals produced by the user – both conscious and subconscious. Our main challenge is how to integrate all this information coherently.”

Here is the thing. The challenge is how to integrated all the information coherently. Different data can be presented differently – 3D geometry, 2D schema, 2D drawings, graphics, tables, graphs, lists. In many situations we can get this information presented separately using different design and visualization tools. However, the efficiency is questionable. Many data can be lost during visualization. However, what I learned from CEEDS project materials, data can be also lost during the process of understanding. Blindspotting. Our brain will miss the data even we (think) that we present it in a best way.

What is my conclusion? Visualization of data for better understanding will play an increased role in the future. We just in the beginning of the process of data collection. We understand the power of data and therefore collect an increased amount of data every day. However, to process of data and visualizing for better design can be an interesting topic to work for coming years. Just my thoughts…

Best, Oleg

Will public clouds help enterprises to crunch engineering data?

August 6, 2014


The scale and complexity of the data is growing tremendously these days. If you go back 20 years, the challenge for PDM / PLM companies was how to manage revisions CAD files. Now we have much more data coming into engineering department. Data about simulations and analysis, information about supply chain, online catalog parts and lot of other things. Product requirements are transformed from simple word file into complex data with information about customers and their needs. Companies are starting to capture information about how customers are using products. Sensors and other monitoring systems are everywhere. The ability to monitor products in real life creates additional opportunities – how to fix problems and optimize design and manufacturing.

Here is the problem… Despite strong trend towards cheaper computing resources, when it comes to the need to apply brute computing force, it still doesn’t come for free. Services like Amazon S3 are relatively cheap. However, if we you want to crunch and make analysis and/or processing of large sets of data, you will need to pay. Another aspect is related to performance. People are expecting software to work at a speed of user thinking process. Imagine, you want to produce design alternatives for your future product. In many situations, to wait few hours won’t be acceptable. It will be destructing users and they won’t use such system after all.

Manufacturing leadership article Google’s Big Data IoT Play For Manufacturing speaks exactly about that. What if the power of web giants like Google can be used to process engineering and manufacturing data. I found explanation provided by Tom Howe, Google’s senior enterprise consultant for manufacturing quite interesting. Here is the passage explaining Google’s approach.

Google’s approach, said Howe, is to focus on three key enabling platforms for the future: 1/ Cloud networks that are global, scalable and pervasive; 2/ Analytics and collection tools that allow companies to get answers to big data questions in 10 minutes, not 10 days; 3/ And a team of experts that understands what questions to ask and how to extract meaningful results from a deluge of data. At Google, he explained, there are analytics teams assigned to every functional area of the company. “There’s no such thing as a gut decision at Google,” said Howe.

It sounds to me like viable approach. However, it made me think about what will make Google and similar computing power holders to sell it to enterprise companies. Google ‘s biggest value is not to selling computing resources. Google business is selling ads… based on data. My hunch there are two potential reasons for Google to support manufacturing data inititatives – potential to develop Google platform for manufacturing apps and value of data. The first one is straightforward – Google wants more companies in their eco-system. I found the second one more interesting. What if manufacturing companies and Google will find a way to get an insight from engineering data useful for their business? Or even more – improving their core business.

What is my conclusion? I’m sure in the future data will become the next oil. The value of getting access to the data can be huge. The challenge to get that access is significant. Companies won’t allow Google as well as PLM companies simply use the data. Companies are very concerned about IP protection and security. To balance between accessing data, providing value proposition and gleaning insight and additional information from data can be an interesting play. For all parties involved… Just my thoughts..

Best, Oleg

Photo courtesy of Google Inc.

The end of single PLM database architecture is coming

August 5, 2014


The complexity of PLM implementations is growing. We have more data to manage. We need to process information faster. In addition to that, cloud solutions are changing the underlining technological landscape. PLM vendors are not building software to be distributed on CD-ROMs and installed by IT on corporate servers anymore. Vendors are moving towards different types of cloud (private and public) and selling subscriptions (not perpetual licenses). For vendors it means operating data centers, optimize data flow, cost and maintenance.

How to implement future cloud architecture? This question is coming to the focus and, obviously, raising lots of debates. Infoworld cloud computing article The right cloud for the job: multi-cloud database processing speaks about how cloud computing is influencing what is the core of every PDM and PLM system – database technology. Main message is to move towards distributed database architecture. What does it mean? I’m sure you are familiar with MapReduce approach. So, simply put, the opportunity of cloud infrastructure to bring multiple servers and run parallel queries is real these days. The following passage speaks about the idea of how to optimize data processing workload by leveraging cloud infrastructure:

In the emerging multicloud approach, the data-processing workloads run on the cloud services that best match the needs of the workload. That current push toward multicloud architectures provides the ability to place workloads on the public or private cloud services that best fit the needs of the workloads. This also provides the ability to run the workload on the cloud service that is most cost-efficient.

For example, when processing a query, the client that launches the database query may reside on a managed service provider. However, it may make the request to many server instances on the Amazon Web Services public cloud service. It could also manage a transactional database on the Microsoft Azure cloud. Moreover, it could store the results of the database request on a local OpenStack private cloud. You get the idea.

However, not so fast and not so simple. What works for web giants might not work for enterprise data management solutions. The absolute majority of PLM systems are leveraging single RDBMS architecture. This is fundamental underlining architectural approach. Most of these solutions are using "scale up" architecture to achieve data capacity and performance level. Horizontal scale of PLM solutions today is mostly limited to leverage database replication tech. PLM implementations are mission critical for many companies. To change that would be not so simple.

So, why PLM vendors might consider to make a change and to think about new database architectures? I can see few reasons – the amount of data is growing; companies are getting even more distributed; design anywhere, build anywhere philosophy comes into real life. The cost of infrastructure and data services becomes very important. In the same time for all companies performance is an absolute imperative – slow enterprise data management solutions is a thing in the past. To optimize workload and data processing is an opportunity for large PLM vendors as well as small startups.

What is my conclusion? Today, large PLM implementations are signaling about reaching technological and product limits. It means existing platforms are achieving a possible peak of complexity, scale and cost. To make the next leap, PLM vendors will have to re-think underlining architecture, to manage data differently and optimize cost of infrastructure. Data management architecture is the first to be considered. Which means end of existing "single database" architectures. Just my thoughts…

Best, Oleg


Get every new post delivered to your Inbox.

Join 286 other followers