I was lucky enough to accompany my fellow Cisco IT EMEAR experts to this year’s Gartner IT Symposium/Xpo in Barcelona. At the show I had the opportunity to meet with customers and share lessons around cutting edge IT technology. I also got to present on a number of examples of Cisco’s own internal use of Artificial Intelligence and Machine Learning (AI/ML).

AI presents an extremely useful range of tools that can help us turn enormous data pools into information we can use to make organisations more secure, networks easier to manage, and businesses more profitable. In recent years we have seen AI deployed in many new offerings across the market. Just about every one of the 130+ vendors at the Gartner ITXpo were offering AI-powered tools somewhere in their portfolio.

Cisco too increasingly uses AI and ML across its customer-facing portfolio. For example, we’ve added Encrypted Traffic Analytics to Stealthwatch to find day-zero malware inside encrypted data streams. We’ve revolutionized network operations with our LAN and WAN DNA Assurance, which uses AI to automate provisioning, identify potential problems, and implement or suggest fixes. We’ve extended AI into the Data Center with UCS Intersight and Network Assurance Engine, Tetration Analytics and AppDynamics. Cisco has even come out with a new AI-ready UCS server featuring Nvidia GPU chips that makes AI work much faster. And much else besides.

Most organisations recognise that AI can be used to improve their internal processes. But most struggle to determine exactly how to get started on this journey. In my recent talk I focused on some of the home-grown AI projects Cisco has been using to address its own business and process concerns (‘Cisco on Cisco’). I hope these examples can help others to see new business opportunities for AI in their own companies.

Like many companies, Cisco still has customers who fax product orders into Cisco, each with their own custom entry forms, none of which are even close to Cisco standard product order forms! Like anyone would, our employees go through the (sometimes painstaking) process of transferring these custom orders into our standard process rather than turn down orders. By implementing an AI/ML process we have been able to increase the speed and accuracy of this process, augmenting our staff to cut the manual workload by about 65% – and it’s constantly improving.

Looking to another example, intellectual property management is important to any company, and Cisco is no exception. Cisco Infosec has developed an AI/ML tool that looks at several thousand documents a second as they move across the network and identifies unusual traffic patterns intended to spot any potentially inappropriate movement of classified information.

Looking at other examples:

  • Our infrastructure networking team found that Machine Learning can identify change requests that are more likely to cause problems in a network, just by looking at the text of the request.
  • Our sales teams make a lot of customer calls. We use AI tools to recommend who they should call first. Other sales teams respond to customers who are interested in Cisco products – but how? Send a brochure or a tech document? Text, call, email, IM, start a video session? AI can suggest the best way to respond, depending on the customer.
  • Our huge supply chain (spanning 700 interconnected suppliers) needed AI help in multiple areas, and Cisco built some custom tools to lend a hand. New AI tools speed up product quality testing. Others make better use of our existing inventory to save costs.

So how can you get started to think of problem areas within your business that could be solved with AI and Machine Learning? Well first, ask yourself if any of your staff are searching though papers or data:

  1. Looking for unusual events or patterns?
  2. Grouping / sorting / classifying data?
  3. Searching for information, or advice?
  4. Trying to predict upcoming issues, or supplier or customer behavior?
  5. Trying to understand what’s important and what’s not?
  6. Trying to “translate” one type of information to another?

These are all classic examples of where AI can be applied to improve processes in a scalable way, and drive efficiencies.

The fundamental steps you should take thereafter, while clearly demanding significant resource allocation, are simple in principle:

  1. Assemble a team of AI/ML developers: Given the skills shortage in the industry, this may be the hardest part! Flexibility and imagination over your resource use can be important. Look to internal experts, but consider retraining or use of external parties.
  2. Start with the Data: Data, as we all know, is the lifeblood of AI systems. Whether you need to draw on internal or external data, locate, cleanse, and use. Again, this can be a highly time-consuming stage of the project, but important to get right.
  3. Build and test different algorithms to address your problem: And continue to try combinations of options. There is a fast-growing ecosystem of AI / ML tools available, and no one knows all of them. With a large and diverse team, you should be able to cover most of the relevant options, but discovering and selecting the right tools is still somewhat of an art form. We’ve provided a few helpful suggestions below.*
  4. Build continuous learning into the system: Continue to automatically pull data updates into the tool, and make sure that topic experts are on hand to continue to train the ML systems by identifying new items of interest among the highlighted new data events found.


Learn more about how AI is applied within Cisco products, and how it might help you.


* While clearly not an exhaustive list, I have found that investigation and use of some of the following algorithms can represent a particularly good ‘starting point’ for those beginning on the journey of an AI/ML development project.

  • Autoregressive and ARMA
  • CART
  • CIR++
  • Compression Nets
  • Decision Trees
  • Discrete Time Survival Analysis
  • D-Optimality
  • Epsilon Greedy
  • Factor Analysis
  • Gaussian Mixture Model
  • Genetic Algorithm
  • Gradient Boosted Trees
  • Hierarchical Clustering
  • Kalman Filter
  • K-Means
  • KNN
  • Linear Regression
  • Logistic Regression
  • Maximum Likelihood estimation
  • Monte Carlo Simulation
  • Multinomial Logistic Regression
  • Neural Networks
  • LP/IP/NLP Optimization
  • Poisson Mixture Model
  • Random Forests
  • Restricted Boltzmann Machine
  • SVD, A-SVD, SVD++
  • SVM