As much as we like to think that the evolution of artificial intelligence into sentient, self-thinking entities will forever be confined to the pages of an Isaac Asimov novel, that reality is far closer than we think.
Artificial Neural Networks (ANN), which are in essence, computing systems modeled on our very own biological neurological systems, have made the concept of a self-thinking AI entity a reality, or a close approximation of it, rather.
Now, by saying that neural networks emulate our own neurological process, we don’t mean to evoke images of walking, talking humanoid robots that could be planning to take over our planet. That is, thankfully, still contained within the wide, sometimes terrifying, world of science fiction. For now.
What we mean is that artificial neural networks are progressively learning systems that continuously improve their function over time. The network is made of hundreds, thousands or even millions of neurons called units arranged in three interconnected layers:
- Input units, which receive information and data from an external source that the network needs to process or learn about.
- Output units, which produce a response to the information processed or learned by the network.
- Hidden units, which sit between the input and output units and form the bulk of the network that processes or learns the tasks it’s supposed to perform.
There are two ways in which an ANN learns:
Here the ANN is “taught” by giving it both input variables and expected outputs for those inputs. Neural networks learn through a process known as “back-propagation” – where they compare their actual output to the expected output and then self-correct to narrow the margin of error between the actual and expected output. Once they have reached the point where they can no longer be trained, they can then be fed a new set of data, and can be expected to produce the expected outcomes with accuracy.
A good example of a supervised ANN is face-recognition software: the software is taught how to recognize distinctive features of the face through a series of images, until it can identify faces.
The network is only fed a set of input variables, but no expected outputs. It is left to identify the structure and relationships between the different inputs. From there, the network can group the data into inherent groups based on the similarities in certain characteristics. It can also be used to generate associative relationships between the data.
ANN’s have already found a wide range of fields in the real world: in engineering, management, medical and industrial applications. And even games: take this Rubik’s cube ANN developed by Jeremy Pinto, a graduate from the Waterloo University in Canada. Using Google’s machine learning library, Tensor Flow, he developed a neural network that could solve a Rubik’s cube after it was shuffled six or seven times away from the completed solve.
So, now we come to the crux of this article: How do artificial neural networks and marketing connect?
In many ways…
However, with the influx of information pouring in in the form of big data, making inferences, predictions and decisions is now beyond the scope of human capability. Since big data is available through multiple channels, in multiple forms, there’s a need to organize it into a format that’s relevant to a particular business’s needs and goals.
Artificial neural networks can be effective in gathering and extracting the right information from big data, identify trends, relationships and connections between the data, and then rely on the past outcomes and behaviors to help marketers identify and implement the best tactics and strategies.
Predictive analytics is a confluence of two statistical methodologies, data mining and predictive modeling, which can be augmented by the machine learning capabilities of neural networks. By learning to recognize the current and past trends and behaviors, artificial neural networks can make predictions on future outcomes within a campaign.
For instance, Microsoft used the BrainMaker neural network to fine-tune its direct mailing campaign, increasing its mail response rate from 4.9% to 8.2%. How? The network analyzed data associated with 25 variables such as the recent product purchased by a customer, the date of first product purchase and the time elapsed between the release of a new product and the purchase of the product.
By analyzing behavioral patterns associated with each of these purchases, the NN was made to score each of the users according to the likelihood of them opening a mailer. This allowed Microsoft to incisively target only those users with a higher likelihood of opening a second mailer from them, and thereby increase their mail response rate.
Segmentation and micro-targeting are key tactics in any marketing campaign; marketers need to be able to single out the customers that will respond positively to a product or service. A customer’s response, as we know, is influenced by a number of factors, including specific characteristics associated with them, such as their demographics, socio-economic status and geographic location, and more importantly, by their attitude and emotions at any given time.
Neural networks can be used effectively to segment the audience into distinct groups based on the above-mentioned qualifications. Unsupervised neural networks can be used to segment the data based on similarities in broader characteristics like their demographic or geographic locations; whereas, supervised neural networks can be used to create more incisive segments based on very precise conditions like their buying behavior or actions they took on marketing content, such as a download of a particular e-book or signing up for a newsletter.
Estimating a business’s future performance, both long and short-term, based on historical data, competitor and industry analysis, and economic trends is essential to its success. Insights drawn from sales forecasting can help a business make informed marketing decisions pertaining to their growth and increase in their sales revenue.
Neural networks trained with the back-propagation algorithm can be used to predict future sales performances, product demand, and inventory control.
Gelling with ANN’s and Marketing
Does this mean that human involvement in marketing and sales activities will eventually disappear? The short answer: no. At least, not yet.
This fear isn’t without any basis, however. Technological progress in the field of ANN’s is now at the point where they can learn on their own with no human input: Google’s AlphaGo, the latest incarnation of the Go-playing AI, started devising its own moves to win the game within the space of three days – all without any human intervention. This has led the team at Deep Mind, Google’s AI group, to start applying it to real-world problems.
Likewise, in the field of content marketing, advances in neural networks have taken over nearly every aspect of it, right from content creation to content distribution.
So back to our question then, will the use of AI and ANNs completely knock human involvement out? According to a study done by the Future of Humanity Institute at Oxford University, the extent of AI involvement will depend on the specialty and precision required in the field. That means translational tasks – such as those undertaken by marketing and sales reps – will be largely replaced or augmented by AI-driven assistants by the year 2024.
How is that a good thing, you ask? In terms of the business’ success, it certainly is, for one. But, instead of putting marketers out of business, these advances would actually make their job easier, allowing them to focus their skillsets on coming up with newer, more effective strategies instead of number-crunching and data analysis. As we previously noted in our article on AI and Content Marketing, the advances in AI will mostly mean a shift in the skillsets required to handle the job.
Artificial intelligence isn’t going to yank the carpet from under marketers’ feet; if anything, it will help them grow into a more efficient version of themselves.