One brain to rule them all: do telcos need centralized AI?

AI brain
Image credit: Orla / Shutterstock.com

Many telcos are adopting AI one department at a time, which is resulting in AI silos. Amdocs:next has an idea: a centralized cloud-based AI brain running everything.

Telcos are increasingly dipping their toes in the AI waters, with different departments exploring different use cases for artificial intelligence and machine learning (ML). The problem is that this is resulting in AI silos  – which is not only inefficient, costly and difficult to scale, but also goes against the grain of the current push towards digital transformation.

One proposed solution: create a common AI brain that everyone can use.

Hillel Geiger, head of marketing at amdocs:next (the division of Amdocs focused on next-gen tech related to AI/data, IoT, cybersecurity and payments/financial services), says that telcos are adopting AI technology for things like intelligent catalogs or predictive engines for CRM. But their experience so far has generally been less than satisfactory, in no small part because the data – and thus the AI being implemented to analyze it – is still siloed away from each other.

“If I had to give an analogy to the human body, the situation is like having a brain for each one of our separate organs – like, my leg has a brain and my hand has a brain, and many of these brains are not communicating with one another, which obviously is not the best way of handling this,” Geiger explains.

What’s needed, he says, is a centralized ‘AI on top’ approach to analytics that – once you strip away all the marketspeak – amounts to (1) building a cloud-based data hub with a data model that can be used for multiple use cases, (2) making sure everyone in the organization can access that data, and (3) leveraging the same AI capabilities across different business functions and use cases.

Geiger says Amdocs:next has developed a solution that does just that – enable one overarching, automated AI “brain” in the cloud to serve different domains.

“Let’s say a service provider needs to match a system for journey management to allow them to introduce AI in the different places of the customer journey and do different kinds of offerings along that journey,” he says. “We’ll implement it in a way so we can later on use it also for the intelligent catalog, or use it not only for CRM, but also for different network capabilities. So this is having a common AI brain.”

Geiger adds that getting this to actually work requires an ecosystem approach, which is why amdocs:next is partnering with companies like AWS, Microsoft Azure and Snowflake to help develop the cloud and AI/ML side of the equation, while Amdocs provides its own expertise with B/OSS and analytics, as well as data management capabilities.

Another key to making a common AI brain work is not just developing the right logical data model, but making sure the data itself is structured properly so that the model can crunch it accurately regardless of which domain is accessing it, and be able to make decisions based on – say – whether the customer whose behaviour you’re analysing is a consumer or an enterprise customer.

“It’s about looking at the data when it’s first created, and understanding whether you need to store this data – is it bringing anything new, is it duplicate data?” Geiger says. “We manage a lot of this and automate a lot of these processes, and handle the data structure it in a way that makes it much easier to process and utilize at the later stages.”

One crucial challenge to this approach is AI training. Typically, AI/ML models have to be trained to ensure that they understand the data they’re processing to ensure the eventual decision the AI makes is correct and unbiased. That’s challenging enough when the AI is focused on a single domain using a particular dataset, let alone multiple domains or applications.

Howver, Geiger says the amdocs:next ‘brain’ is designed to accommodate this.

“Although few data sources can be leveraged to solve multiple problems, the brain enables specific model training pipelines and related business logic to be deployed and configured, with the goal to optimize the solution for a department-specific business problem, whether it’s related to marketing, customer care or operations,” he says. “In cases where cross-organizational KPIs need to be optimized, the brain will include these domain-specific KPIs in the decision making process.”

One interesting aspect of this centralized approach – at least for amdocs:next – is that as more CSPs adopt this solution, whatever their central brains learn as they go along can be applied to other CSPs who adopt it later.

“We have the capability to take the learnings from different service providers and train our algorithms in a way that would be much more scalable, and allow them to learn much faster,” Geiger says.

Be the first to comment

What do you think?

This site uses Akismet to reduce spam. Learn how your comment data is processed.