You’ve no doubt seen the relatable AI memes travelling across the universe of cyberspace right now. “To replace creatives with AI, clients will have to accurately describe what they want. We’re safe, people.”
The terms “creatives” and “clients” have been swapped out for programmers, product managers and anybody else whose job is at threat. But the message is the same: garbage in, garbage out. We cannot use AI to its full potential if we’re not giving it a clear brief.
Baking the cake – understanding what we really want
In 2023, we’re generating more data than we know what to do with. Estimates vary, but figures for generated, captured, copied or consumed data come in at as high as 328.77 million terabytes per day [LINK: https://explodingtopics.com/blog/data-generated-per-day]. It’s a classic case of quantity over quality. In fact, some researchers are even going as far as to describe large language models as “ticking timebombs” due to copies of data.
With this in mind, we have to acknowledge that not all data is valuable, and as such, won’t help us make decisions. We cannot jump on the bandwagon of AI without understanding what we really want. Let’s bake a cake.
To bake a cake worthy of a Hollywood handshake, we need:
The cake: the outcome or answer we want
The ingredients: the data itself
The recipe: the program that helps everything come together.
What we have here, beyond a very worthy Victoria Sponge, is a procedural program. We have raw data – eggs and flour – and we know what we need to do to produce our desired outcome. In isolation, AI doesn’t have the recipe to produce outcomes. Rather, it relies on the outcome you request of it (the cake) and comes up with its own calculations to bake something edible.
As we have seen in the past, this comes with mixed results.
The benefits of using structured data
At Tiger, we rely chiefly on structured data to help our clients make strategic decisions. For example, we look at deterministic data (which is more accurate as it has been provided by customers) such as call length, number of user licences or customer waiting times.
In telecoms, this is far more useful than unstructured data, which is prone to error. Let’s say we’re asking an AI to show us this month’s call data. This could include unstructured data such as speech, which the AI would turn into an SQL query to retrieve information from the database. All it takes is for one misinterpretation of a regional accent and we’ve essentially retrieved gobbledegook.
At best, this is inconvenient. At worst, this could cause us to make misguided decisions with catastrophic results, particularly if we’re working in fields like healthcare or insurance.
Deterministic outcomes
Of course, we can apply human intelligence to unstructured data – for instance, gauging employee satisfaction from their tone of voice and general demeanour in video calls. But this is time-consuming and cannot be done at scale.
So we use structured data to help our clients make key business decisions. Notable use cases include:
Adoption of tools: quantifying user licences shows senior teams where they could cut costs
Call waiting times: identifying peak times helps our clients to allocate staff resource
Preferred channels: counting how many customers interact via telephone versus live chat shows managers where to invest.
Missed calls: a missed call is a missed sale or opportunity.
This gives our clients deterministic outcomes based on the data available. As humans, we offer the roadmap to get from A to B – how can this data guide us in the overall direction of the business?
To get back to basics, this is why we need human intervention before we can leverage AI. It’s all very well having the data, but we need to understand what we want from it so that we can use it to its full potential. Do we want to save costs, improve customer service, retain staff or all three?
The value of AI
As time and technology move on, we should expect to see more use cases for AI in telecoms. For example, while Tiger doesn’t deal directly with network outages, we’re already seeing AI being used to track network congestion and help companies stay operational.
It will also be invaluable when it comes to processing data at speed, in particular, unstructured data. We can use human data mining techniques to make decisions on structured data, but the merits of voice or facial recognition will come into their own in years to come. In practical applications, this could help our clients with fraud detection, from video to voice calls (barring some regional accent hiccups – but we can work it out).
The challenges
However, it’s important not to rely on AI alone or let it be the only guiding principle for business decisions. There are still ethical considerations, such as how natural language interfaces are processing private conversations.
Again, this is pertinent in fields such as insurance, finance or healthcare. We should expect to see more regulation around AI and data privacy in years to come – and we should have answers if our customers start asking questions. Already, two in five consumers are worried about AI [LINK: https://yougov.co.uk/topics/technology/articles-reports/2021/11/18/global-more-people-worried-not-about-artificial-in]
This is why we cannot underestimate human intervention. Letting humans take the reins will assure stakeholders that:
We gain true deterministic outcomes based on knowing what we want from data.
We apply an empathetic approach to unstructured data like video calls to read emotions.
We respect privacy concerns and keep customers informed about how data is being used.
There is no doubt that AI will help us to make better decisions in the future. But just as an AI might not consider gluten-free flour or other data nuances, we cannot overlook human intelligence.
Phillip Smith is Software Director at Tiger, providing data analytics for the unified communications space.