Telco use-cases for AI: A simple categorisation model

The coming years will see the application of AI technology across all sectors of the economy and life. The telecoms industry is no different. Although I’ve been commenting on telco-sector AI in the context of “TelcoFuturism” for some time (link), and co-ran a workshop on it in May 2017 (link), the last few months have seen a notable upswing in interest. I’d say that the public use-cases now seem to be significantly in advance of those for blockchain, in terms of potentially-transformational technologies.

That said, it can still be hard for many executives to grasp exactly what is likely to change, and when, for AI/telecoms combinations. This is highlighted by the surge in AI-related panels, presentations and even complete streams at industry conferences – although sometimes I see more interest from generalist AI people about the telecoms vertical, versus telecom specialists looking at what’s new. 

Both sides of the equation have large volumes of obscure acronyms, multi-layered technology stacks, and complex volume chains – which can mean that mutual understanding is often confined to narrow niches. AI covers machine- and deep-learning, language processing, machine-vision and much more. Telecoms includes vast realms of internal systems and processes that are unknown to most who are not insiders – domains like core networks, OSS/BSS, network optimisation, toll fraud and service-assurance are alien to those not steeped in the industry.

One of the ways I’ve been using to “set the scene” for describing AI/telecoms intersections is to simplify and categorise the use-case areas. I count three, possibly four, large “buckets” into which a variety of telecom AI impacts will fit. These buckets are not based on either specific AI or telecoms technology slices, but more on understandable business functions and roles:

• Dealing with customers
• Managing operations
• Creating new services
• (External risks)

 Within each of these areas, there are many, many sub-sectors – and also some overlap.

“Dealing with customers” can include everything from voice/text chatbots for customer-service, through to predictions of which customers are least-happy and may “churn” to competitors. Where telcos have retail outlets, it could incorporate various in-store technologies, or it could be about smarter web-consoles for B2B customers running complex managed services.

“Managing operations” is even more diverse – it could be fault prediction for network elements, optimising the 100s of configuration variables for radio networks, spotting fraudulent traffic to international premium-rate numbers, allocating engineering resources more productively, or protecting against hackers and malware. There are hundreds of possible uses here, which mostly overlay on top of existing operational/business support systems (OSS/BSS) See also my recent post (link)

“New services” also spans a range of areas, but broadly splits between AI-enabled and AI-enabling services. An AI-enabled service could be a local-language voice assistant added to a cable operator’s set-top box or remote control. Or it could be the provision of integrated “smart city” solutions including video-cameras and security analytics. AI-enablement could include offering “edge” servers for hosting local processing, milliseconds transport-time away from a device, or it could be the provision of anonymised bulk data for others to apply algorithms to. Telco opportunities with IoT+AI include both enablement and enabled services, in numerous manifestations.

The “risks” category includes a diffuse set of possibilities by which AI might harm the telecom industry, or dampen demand for services. Smarter devices (eg autonomous vehicles) will be able to host their own offline image processing & route-planning locally, rather than needing realtime connectivity at 5G speeds/latencies. Another threat could be customers’ smart assistants renegotiating price-plans on their behalf – after crowdsourcing millions of conversations to deduce how best to game the retention staff’s scripts and objections. (Of course in the latter example, the customer-retention team could themselves be bots). Numerous types of automated “least-cost X” and arbitrage engines are likely to emerge. Various security risks are also probable here too.

Clearly, using just these four "buckets" misses much of the fine-grained detail. But I find it helpful as a starting point, as most top-level industry issues apply differently to each. 

Consider input data, for example – for both customer management and operations, telcos have abundant historical records and ongoing data collection that may generate terabytes per day. But for the former, privacy considerations often come to the fore in terms of regulation and risk, while this is far less of a concern for internal operational data, for example on how the network is running. For new services, almost by definition the focus is on collecting/processing/transporting new data, rather than deriving conclusions from existing sets. 

This four-way framework is also useful for thinking about different types of ROI model - split broadly between impacting existing revenues, existing costs, new revenues and potential changes to underlying assumptions. 

I'll be covering these topics in more depth in various upcoming presentations and reports, as well as looking at other areas of telco-linked innovation such as blockchain, 5G and enterprise verticals. Please get in touch if you would like more detail, or are interested in internal workshops, external support through events or white papers, or are seeking ongoing strategic advisory support.

Machine-learning & operations for telcos; and a discount-code for AI World

A lot of discussion about deep/machine learning in the telecoms industry focuses on either customer management, or applications in various enterprise verticals. Typical use-cases are around assisted self-care for end-users (for example, diagnosing WiFi configuration problems), or spotting customer behaviours that suggest imminent churn (or fraud). 

Some telcos are looking at chatbots or voice assistants, for example for connected-home applications. Then there are offers around image recognition, perhaps for public-safety use, where telecoms operators have traditionally had a major role in many countries.

All this remains very important, but recently, I've been seeing a lot more "behind-the-scenes" use-cases being discussed, as both mobile and fixed operators look to improve their operational effectiveness and cost-base. Many of these are less-glamorous, and less-likely to highlight in non-telecoms articles, but they are nonetheless important. 

A few examples have been:

• BT has talked about using ML to improve efficiency of maintenance staff schedules, depending on particular tasks and base locations.
• Vodafone talked at the recent Huawei Mobile Broadband Forum about AI being used in radio networks for predictive load-balancing, predicting patterns of users/usage, and optimising configurations and parameters for better coverage and throughput. It also referenced using ML to help distinguish real from fake alarms
• KT in Korea, talking about collecting 50TB per day of operational data from its fibre network, and using it to optimise performance, improve security and predict faults. (It also realised that it accidentally created a huge realtime seismic detector, if earthquakes - or maybe North Korean nuclear detonations - flex the fibres)
• Telefonica working with Facebook's Telecom Infra Project initiative to map population density (from satellite images) to network usage data, to work out coverage gaps (see here)

As well as traditional telecom operators, new breeds of Internet-based communications providers are also looking at instrumenting their services to collect data, and optimise for multiple parameters. For example, Facebook (which is a "new telco") is improving its voice/video Messenger app, by collecting data from its 400 million users. This involves not just call "quality", but maps codec use to battery levels on mobile devices, and various other measureables. Potentially this allows a much broader type of optimisation than just network-centric, by considering the other trade-offs for users such as length of call vs. power consumption vs. video quality.
The key for all of this is collection of operational data in the first place, whether that is from network elements, end-user devices - or even external data sources like weather or traffic information.

I'll be digging into this in various future posts - but I'll also be speaking at various conferences and panel sessions about Telecoms & AI in coming months.

In particular, I'm on a Mobile & AI Panel at AI World in Boston, which runs from Dec 11-13. Details are at https://aiworld.com/ - and if you want to attend, I have a code for a $200 discount for 2 and 3-day VIP Conference Passes: AIW200DA

In January, I'll also be covering AI at the PTC'18 event in Honolulu from Jan 21-24 (link here).

And in April, I'll be at the AI Net event in Paris (link here) moderating a panel and also talking about AI in smartphones.

Overall - I'm expecting a huge #TelcoFuturism push around all aspects of AI in telecoms in 2018, but it's especially the operational and network-management functions that I think will make a big difference. It also coincides with the arrival of both 5G and large-scale NFV, and the intersection points will have a further multiplicative effect.

Huawei Connect: IT services, Enterprise Cellular, video analytics, AI and more

I spent most of last week in Shanghai, attending Huawei's Connect conference and trade show. It was a good chance to get a deep-dive into the company's enterprise activities, as well as get my head around China's broader trends and influences around the technology sector.

I normally engage with Huawei through its analyst relations function, but this trip was organised by a different team. The company apparently considers me a "KOL" ("key opinion leader"), which is a rather diffuse bucket used for a mix of outspoken independent analysts, public-facing academics, video/social bloggers and assorted others. I'm not sure I set out to lead opinions, but I'm certainly happy to voice my own.

(Unlike the analyst events I usually attend, the KOL group isn't really made up of direct competitors, so there's a more collegiate atmosphere - and a very lively WeChat group, partly with logistics about meeting times/locations but also sharing photos or thoughts about the event).

Connect is mostly driven by Huawei's enterprise business unit, which is growing fast (about $6bn revenues in 2016, up 47% [link]), and focuses on cloud and big "infrastructure-led" IT and networking projects. So sectors like smart cities, advanced manufacturing, oil and gas IoT, systems for transport sectors like rail and ports and so on. There's a heavy emphasis on IoT platforms and networks, cloud and storage, video/image surveillance analysis and a lot of AI. 

It clearly intends to be a very significant player in its chosen sectors, using its existing high IT profile in China, plus its global telecom footprint, as a springboard for other international ICT theatres. Unlike Europe, North America and India, China has few global-scale IT companies, especially in systems integration or outsourcing. The closest to a "Chinese version of IBM" is probably ChinaSoft, which has a deep partnership with Huawei anyway, and in which Huawei owns a significant shareholding.

Thinking more about technology-sector comparables, very few have a similar blend of infrastructure/network/telecom expertise, systems integration/services scale and cloud capabilities. Given Ericsson's recent announcements of pulling back on direct enterprise-related initiatives to focus on CSPs and its Cisco partnership as channels (a strategic error, I feel), it's only really Nokia and maybe NEC that have the scope to push the same big-infrastructure enterprise "ICT" vision, although even it doesn't have the full-scale IT services business that Huawei does. Perhaps there's yet more scope for consolidation between traditional IT companies and networks. (Ericsson+IBM? Nokia+HP? NEC+Tata? Who knows....)

One other thing stood out about the event: there was very little spoken about telco networks, Huawei's main business, or the synergies between that business unit and its faster-growing enterprise sibling. 

There was much more about robots and face-recognition than network-slicing and NFV. The main mention of IMS that I saw was in the context of critical communications for public safety, eg push-to-talk. The X-Labs group assessing possible future 5G use-cases was talking about connected drones, or cloud-integrated video-enabled helmets for the blind. There was a "carrier" section in the vertical-industries show hall, but that seemed mostly focused on cloud solutions for telcos.

Conspicuously, there was almost no reference to delivery models for network or IoT capabilities for enterprises. There was no assumption that everything would be provided "as a service", or in particular, delivered by a CSP. There was tacit recognition that some organisations want to own their own infrastructure / private clouds, some may go to a specialist integrator (eg an automation/IoT specialist like Honeywell or GE), and some might use an arm of a telco. For example, T-Systems, Deutsche Telekom's IT unit, was there talking about a Huawei-based storage cloud, deployed for CERN, the leading nuclear and particular research institution on the Swiss/French border.

Huawei also offers its own cloud services, but is quite self-effacing about it, only wishing to become "one of the top 5 clouds" (presumably along with Amazon, Google, IBM and maybe Microsoft - which it also partners) and saying that "1% is enough for us". I don't think Jeff Bezos is going to have too many sleepless nights, although Alibaba, Cisco and Oracle may have different opinions on the top tier's members, the former especially in China itself.

In terms of specific takeouts on my normal coverage areas, a few things stood out:

• Enterprise Cellular: This was everywhere at the event, under the brand eLTE. This is a sort of pre-cursor to a MuLTEfire / CBRS model of non-carrier cellular networks. There's a quite large eLTE ecosystem, especially around public-safety networks but also manufacturing, transport and other verticals. There was a demo of a robot connected with private cellular. There are 3 variants:
• An unlicensed LTE-U version that doesn't need a licensed "anchor" like LAA, so can be deployed by any organisation
• A licensed-band version, where organisations (such as law-enforcement or utilities) can manage to get dedicated spectrum by one means or another
• A narrowband version, which is essentiially NB-IoT in unlicensed bands such as ISM spectrum (which in China, is in the 500MHz range, or 900MHz in the US)
• All of these were targeted at industry verticals. There wasn't any mention of other use-cases like neutral-host providers, hybrid MNO/MVNOs, mesh networks, or consumer-oriented plays. 
• There wasn't any explicit mention of shared-spectrum models like CBRS, but it seems to fit under the second category.
• This all fits nicely with the recent work I've done on private/enterprise cellular. It will be an ongoing theme as it is clearly "happening", including presentations at a few upcoming regulatory conferences, and another workshop with Caroline Gabriel in London on Dec 1 (link)

 

•  IoT networks: There was a huge emphasis on NB-IoT around the event, as well as broadband 4.5G/5G options for drones, connected vehicles and more demanding applications. I didn't see an mention of LoRA, SigFox, or even LTE-M or Cat1 though, but WiFi and ZigBee cropped up on various slides. Some interesting examples of NB-IoT deployments, notably for cities, or specific OEM-led integrations such as China's booming shared-bicycle sector.
• Video and facial networks/analytics: This was a huge theme, as it bridges Huawei's key domains of mobile broadband, cloud services and AI. A major focus is "safe cities", especially using networked video cameras to manage traffic, enforce public safety - and track/spot individual people, whether that is missing children, criminals, or attendees at a trade show. (I joked on Twitter that Huawei had probably been tracking people around the event itself - only for the next slide to reveal that it had been doing exactly that). Missing from most of the material was much mention of privacy - which appears to be less of a concern in China than it would be in much of Europe. That said, we may be fighting a losing battle on that front, as this week's Economist cover & feature articles on face-recognition point out (link).

 
 
 

• AI: Beyond video-analysis, a central thrust of the event was around machine-learning, graph analysis, image-recognition and other forms of AI.  I didn't get a chance to go into too much depth on this, but it's pretty clear this is central to Huawei's cloud ambitions, and probably will link into carrier-domain services like smart-home / personal voice assistants as well as "big data" corporate applications
• We also had a briefing with the handset unit, which discussed the new Kirin AI-oriented chip which includes a neural processing element, as well as CPU, GPU and DSP. This should enable better and more power-efficient local classification of images, without the need to send all data to the cloud. This fits into my ongoing debate on whether 5G's low-latency business case might be undermined by more edge-processing. (link)
• WiFi: Although not as big an emphasis as 4G/5G, Huawei nevertheless had a fair bit of WiFi on display, particularly for large-scale deployments in cities or large public venues like sports stadia. It also had an interesting hybrid WiFi / IoT networking unit, which for now focuses on Bluetooth, RFID and ZigBee but I guess could incorporate NB-IoT (or its eLTE variant), or even LoRa if a client wanted.
• UC/UCaaS: Although not a major focus of the event (itself quite telling) there was a fair bit of unified communications, conferencing and even cPaaS around the show. There was a Broadsoft-style UC platform for operators, and various tools for multi-party meetings. It's not obvious that Huawei is aiming to be a Twilio / Tokbox-style platform provider though, although it does have APIs (including WebRTC) for embedding communications in apps and websites. I didn't see any signs of a Slack/Spark/HipChat rival. Notably, Huawei is partnering Microsoft on Office365, so may not launch its own full UcaaS direct-to-enterprise product. 
• I liked one partner booth in particular "Call Cloud", which uses a crowd-style / sharing economy approach to sourcing customer-service reps, with in-app video. It apparently has 7 million (!) people signed up as potential providers of informal information or support.

Overall, an interesting few days for me, exploring a side to Huawei I hadn't seen before. It's always hard to get a full perspective from a single-vendor event, but it struck me as one of the only real, fully-encompassing examples I've seen of an acronym I normally dislike - ICT. That said, some more candour about positioning vs. competitors would have been welcome. We all know who they are - so descriptions of differentiation would have been useful, even if rose-tinted.

It's also brought home to me how important it is to have a captive market to drive scale, which can then improve adoption rates (and prices) elsewhere. Amazon does it with AWS - its own huge retail business is an "anchor tenant" which helps create traffic volumes that then became reinforced by third parties' cloud usage. Huawei appears to do something similar with domestic government and enterprise business - millions of CCTV cameras, or large-scale city networks, or local IoT uses are helping it exploit pre-existing scale and experience, and then apply elsewhere. There is also a sensible approach to partnering, for example around IoT, with the likes of GE collaborating on distinct parts of the market.

One final comment: the layout of the trade show was excellent. One hall was organised per-vertical, with sections on Manufacturing, Public Safety, Oil & Gas, Finance etc. The other hall was per-technology, with sections on Cloud, eLTE, WiFi, NB-IoT, Developers and so on. I wish other events were similarly well-structured.