Skip to main content

Marketers already use data science to understand how consumers respond to existing campaigns. With machine learning, they can go further: they can test campaigns before they even go out. MEF experts debated the implications on a MEF webinar.

Did you know machine learning can make you smell nice?

Apparently, perfume houses are racing against each other to see how the tech can help them develop new fragrances.

They are using machine learning (ML) to analyse millions of archived formulas and thousands of raw ingredients. They compare them against client guidelines in order to devise new formulas.

Watch the MEF Webinar in full

AI in perfumery? It just goes to show how new self-learning tech is seeping into every conceivable commercial sector.

Mobile, of course, was an early adopter. MNOs are already using ML to better understand the messaging traffic flowing across their networks. Products such as RealNetworks’ Kontxt look for patterns in the data in order to classify messages into groups.

Machine learning helps you load the dice. As a marketer you might know how to make a message contextual, and you might know when is the right time to send…Now, you can hit that last variable of sentiment and emotion.”

Using Natural Language processing (NLP), Kontxt can study the words in a message to tell what kind of communication it is (P2P or A2P, promotional alerts or two-factor authentication code etc), when it was sent, what network it traversed (local, international, grey etc) and so on.

Just as important, Kontxt can analyse characters to draw deeper conclusions. For example, it can look for misspellings that might indicate fraudulent traffic.

So MNOs are now discovering how ML can complement – and maybe even replace – existing anti-spam tools such as rules and firewalls. It begs the question: if ML can improve the classification of existing messages, can it work at the creation level too? Can it help marketers to formulate more effective campaigns? Insiders think it can.

In the latest MEF webinar ‘Using NLP – natural language processing – to create value in messaging’, we invited two of them to share their knowledge of the topic. The two speakers were:

  • Michael Bordash, GM & CTO for Machine Learning & Messaging at Realnetworks
  • Andy Gladwin, Senior Director, Global Mobile at Cheetah Digital

Here are their key insights.

Traditional data science analyses historic patterns. ML does this analysis in real time.

Michael Bordash: “Data scientists have been looking at analytical data and predicting usage. They have that down cold. Now, we’re looking at what are the characteristics of the text in a campaign before it even goes out. We can ask questions such as: how do I craft content that resonates more? How can machine learning help me pick the right tone, the correct intent and the best word choices?”

NLP can help marketers keep to MNO guidelines and legal regulation

Andy Gladwin: “Brands definitely face a challenge with regulation. They will not put their brand at risk… In the US especially, there are lots of rules – with different regulatory requirements among different operators. These are rules that can end up blocking messages or even fining the brands that send them. Brands can use ML tools adhere to policies. Before they send a message, they can put it into a template to ensure they are compliant. It lowers the risk considerably.”

Every model starts with a hypothesis… and that starts with data and testing. You might need a small army to label that data according to your desired outcome. But when you put that model into production, the benefits (of self-learning) will vastly outweigh having to craft all these rules beforehand.”

You might know what to say. ML lets you know how to say it.

Andy Gladwin: “Machine learning helps you load the dice. As a marketer you might know how to make a message contextual, and you might know when is the right time to send…Now, you can hit that last variable of sentiment and emotion. Maybe you’re sending a Black Friday alert. You can make sure it has the right level of joy and urgency. Or if it’s a bank fraud alert, you can give it the right amount of emotional fear to drive response.”

Humans train the model at first. Then the model trains itself.

Michael Bordash: “Every model starts with a hypothesis… and that starts with data and testing. You might need a small army to label that data according to your desired outcome. But when you put that model into production, the benefits (of self-learning) will vastly outweigh having to craft all these rules beforehand. That was how we did things 10 years ago. But if you stick with this new model, it will pay back 10 fold.”

Providers can ease marketers into the world of ML with existing models

Andy Gladwin: “Most enterprises don’t have the in-house expertise to build this. Starting from a blank is a huge leap. So I think they will turn to providers who already have models. Starting from the mature position that this offers is less scary.”

The customer insight you get from ML can become a brand asset

Michael Bordash: “The resonance you get with customers (thanks to ML) becomes part of your IP. You can use it to look outwards and identify other consumers who might be similar. You can use APIs from Facebook and others to find people who might believe in your brand in the same way.”

ML should be easy to use – offering constant friendly advice

Andy Gladwin: “A machine learning tool should be like an assistant, advising which words resonate more.  Consumers see these kinds of tools all the time – whether it’s help with writing emails better or suggestions of people you might want to include in a message.”

When ML is applied to RCS, the value it delivers around messaging will scale up again

Michael Bordash: “Push notifications give marketers access to incredible amounts of data that goes well beyond what you can get from messaging. RCS will level that playing field.”

ML is great for privacy. No human sees the message content.

Michael Bordash: “In macro terms, machine learning is ideal for handling privacy concerns. There’s no human looking at the data. Also, you can pull out all the personally identifiable information from the data and tokenise it.”

Brands interested in using NLP to gain more customer insight and value can get in touch to be in with a chance of winning one of three pilot programs run by RealNetworks and Cheetah Digital. – Find out more

Tim Green

Features Editor, MEF Minute

  

MEF