Philadelphia -- If customer experience is indeed the most important product, then it’s probably time to move machine learning up on the priority list.
That was the gist of Monday’s kickoff workshop on machine learning and network operations, where executives detailed how the techniques can be applied to network optimization and customer care.
“Right now, we throw away more data than we use,” said Tom Cloonan, CTO for Arris's Cloud & Network Solutions division and the moderator of the session. “We can do better, in terms of the decisions we make about how to modify plant, split a node or not, replace an amplifier, and ultimately, know whether customers are happy or not.”
Jason Schnitzer, founder of Applied Broadband, described his work to optimize DOCSIS 3.1-based networks using the three metrics of optimization: Modulation Error Ratio (MER), Codeword Error Ratio (CER), and downstream receive power -- no small feat, given that DOCSIS 3.1 essentially obviates traditional 6 MHz channel widths, instead spreading 7,680 OFDM (Orthogonal Frequency Division Multiplexing) subcarriers, between 24 MHz and 192 MHz of spectrum.
“If you’re collecting data from tens of millions of devices, doing 7,680 samples, four or so times a day, you quickly get into very large-scale data,” Schnitzer said. Ultimately, DOCSIS 3.1 will enable operators approach the Shannon Limit by enabling multiple modulation profiles across the entire population of cable modems. “So, in a sense, optimization is necessary, for 3.1 to be successful.”
Chris Menier, VP of products and marketing for Guavus, said machine learning is growing out of the industry’s early steps into “big data” -- which began in departmental silos. “Network operations, field operations, billing, care, they all had their own tools, reports, and dashboards,” he said.
Next came data warehousing and federation, then automation use cases, he said. “Now, it’s about enriching and correlating the data -- where did it happen on the network? To what type of device? What was it playing when it failed?”
By detecting anomalies, and correlating them with additional data, operators can identify and address problems before they impact customers. And if care agents are armed with enriched data, in a timely manner, “you can turn someone into a promoter, from a detractor, in NPS [Net Promoter Score] terms,” Menier said.
By applying machine intelligence, anomalies can be detected, correlated and even automatically repaired. Does it work? “It sure does,” Menier said, to the tune of $70 million in savings for an unspecified operator.
