Sergey Nivens - Fotolia
Traditional relational databases are good at managing clear and simple relationships: This invoice contains these line items. This loan application contains this set of documents. This department has these employees, and each employee has these working hours.
A traditional database query against these kinds of data sets is based on those predefined relationships -- start with this department, filter for these types of employees. Traditional databases fall short when those relationships become convoluted or cannot be rearranged without rebuilding the whole data structure.
To solve this problem, enterprises are increasingly turning to graph analytics instead, which are often built on top of graph databases.
According to Gartner, 30% of companies will use graph analytics by 2023 for faster and better decision-making. Graph analytics can help companies find hidden relationships in their data, which can help identify cybersecurity attacks, network vulnerabilities, money laundering or even recommend new products for customers.
With the increased use of artificial intelligence and machine learning, graph analytics becomes even more important. According to Gartner, organizations that use graph techniques will use five times as many AI models by 2023 as those that don't.
"In traditional business intelligence, the value is in the facts," said Jim Hare, vice president and analyst at Gartner. "In graph analysis, it's the relationships between the objects themselves."
As a result, he said, graph adoption will double by 2022 to enable more complex and adaptive data science. These are the top enterprise graph analytics use cases to watch.
Identifying unexpected connections
Graph analytics can help identify connections that shouldn't be there or connections that should be there but are missing.
"How is it that this person is in two different locations at the same time?" Hare said. "Or this particular doctor seems to be involved in a lot of these insurance cases."
Networking firm Cato Networks is one company using graph analytics for cybersecurity.
"My team and I use the Neo4j graph platform to correlate various observables and classify them into malware groups," said Elad Menahem, director of security at Cato Networks.
If a particular type of malware uses a certain IP address for its command and control server and the same IP address shows up in another context, there may be a relationship to the malware, he said. Identifying these relationships helps Cato improve its cybersecurity intelligence and ability to detect malicious capabilities, he said.
Optimum and unexpected paths
Finding the shortest path, the fastest path, the cheapest path or an unexpected path between two points isn't just for navigation. One of the top graph analytics use cases is in mapping tools that provide turn-by-turn directions to drivers or plan delivery routes.
"Sometimes the optimal route is not the one that's most obvious," Hare said. "Graph analytics can highlight those kinds of nonintuitive examples."
Television detectives often have case boards with strings of red yarn connecting suspects, locations and clues. That's an analog example of graph analytics.
At Snoop, a British personal finance app, linked connections come into play when tracking user access permissions. Where personal financial data is concerned, access control is important. Only certain employees should have access to data.
Jim HareVice president and analyst, Gartner
A traditional database might provide a basic list of the permissions associated with a user account. But in real life, it gets more complicated. An employee might have access to a specific account that has access to another system, and that system can provide an entry point into another data store.
"Manually, it would take days for two to three people to work out what's happening," said Andy Makings, head of DevSecOps at Snoop.
To track those connections, the app uses graph analytics from Sonrai Security, a cloud security company. "It does all the heavy lifting for us," Makings said.
One common problem area is when an application is moved from a development environment to a testing environment, said Sandy Bird, co-founder and CTO at Sonrai Security.
When that happens, developers and test users might wind up with access to real data they shouldn't have, and those access rights might hang around even after original users have left the company. If hackers can get their hands on those credentials, they can do serious damage via access paths that aren't readily apparent.
In some cases, those access rights are further obscured when an account gives the user the rights to create access to another account, and that new account would have its own set of additional rights. That can lead to privilege escalation attacks that can be hard to detect.
"We use the same approach for network paths as well," Bird said. "You can see how this piece of compute is attached to this interface through this network component. The customers get to see it visually -- see how everything is connected to everything else. "
Clusters and communities
Graph analytics can also be used to find clusters of nodes that are alike, possibly in unexpected ways. A traditional relational database can do simple groupings such as organizing customers by ZIP code.
Graph analytics might expose deeper relationships between those customers. Those who have the longest commutes might be the ones who buy the most audiobooks. Customers can also be group based on previous purchase histories, or many other factors.
This is one of the biggest graph analytics use cases, said Karen Panetta, IEEE fellow and dean of graduate engineering education at Tufts University.
"Any time you go anywhere, you get lots and lots of ads for something else," she said. "It can also make recommendations based on who my circle of friends is, who my connections are."
Social networks are big users of graph analytics, said Doug Henschen, vice president and principal analyst at Constellation Research.
"It's at the core of Facebook's ability to reveal relationships among people," he said.
In graphs, not all nodes are created equal. In a pandemic, some would be considered superspreader patients or events. In social media, they would be the influencers.
And in computer networking terms, it could be a single connection that everything goes through -- your big point of failure. Graph analytics can help you pinpoint that critical node and avoid complications associated with that node's failure.
"If you know the most important node in a network, it can be a source of vulnerability if the network goes down," Hare said.
The most common place companies begin using graph analytics is to get a complete view of their customers, said Michael Moore, managing director at EY.
Multichannel retailers, organizations that acquired other companies and large businesses in general often have customer data scattered through disparate systems with different formats and field names, making it hard to pull everything together.
"Graphs are really good at being able to assess the quality of data from different systems," Moore said. "It's easy to join the data and compare values."
Similarly, if a company has a large, complicated project with many vendors, contracts, equipment, materials and supplies, graph analytics can bring everything together into one place and offer a complete view of the project's total costs.
And in the pharmaceutical industry, the same approach can be used to organize all the compounds that go into making a drug -- and everything that goes into creating those compounds.
"Graph analytics lets you look deep into the data and identify quality issues or supplier optimization opportunities," he said.
This approach can also be used for regulatory compliance, he said.
"It's not unusual for companies to have to classify the lineage of where their data is coming from," he said.
Used both by large and small companies
Some of the biggest users of graph analytics are companies with extremely large data sets with lots of connections and links between them, said Mayank Kejriwal, IEEE member and research lead at the Information Sciences Institute at the University of Southern California.
"The prototypical company that immediately comes to mind is an e-commerce company with data on products, customers and third-party sellers," he said.
But smaller companies and startups can also take advantage of the technology.
"Forward-facing -- and less conservative -- businesses that see themselves as disruptors also favor graph analytics," he said. "They can apply novel machine learning tools and methodologies to them -- including deep learning."
And the technology is getting easier to deploy with support from major cloud providers, ready-to-use commercial tools or embedded inside enterprise products.
"Many vendors have made community versions of their tools free to use, allowing more people to try it out, use it and spread the word organically," he said. "It is safe to say that this is not a niche market anymore."