Organisations around the world are shifting their focus to artificial intelligence to solve some of their most complex problems. Challenges that directly impact users like product pricing, personalizing communication, and real-time support, all the way to internal efficiency in automating work-load like reporting, forecasting accuracy, and assigning service requests to name a few.
At its core, the promise that AI brings is that it can go through large volumes of data to help us make better, informed decisions. One area where companies are rapidly implementing artificial intelligence solutions for positive gains is no secret, marketing.
While it may seem like a no brainer to implement AI, it's much easier said than done. Stakeholders can often become too dependent on the AI in marketing itself without considering how to interact with it. Knowing the right questions to ask the AI for one, or deploying AI into incompatible systems and internal structures that work in parallel with each other.
For context, one of the most commonly used types of AI in marketing is transaction data. Take pricing for Grab as an example. AI is used to determine that trip you'll take after your spin class on Saturday morning to brunch. Grab needs to collect real-time data on where the drivers are, where you are, and when you request your pick-up. This matching takes place in seconds and sets the price with the most optimal objective for the company.
We can think of it two-fold, systems and predictions, often you'll hear the acronym "AI" thrown around, but decisions are based on machine learning and statistics. Machine learning refers to the tools used to understand the data to provide predictions, whereas the AI will consolidate these findings and predictive output paving a way to integrate them into the systems we use.
A company may decide to adopt or build their own AI systems with internal data science teams, this team will take the data and proceed to draw out predictions based on their expertise. This can vary from user behaviour, like Where will the consumer likely click for a promo? Or where on a page will the user drop off? This is the exciting part for data scientists as it taps into their core skillsets. On the other side of the room, you'll have your marketing team who may be more interested to know which product to launch first, or who receives the promo, and what should it be priced at. This is precisely where misalignment takes place. Two teams, two goals.
This takes precedence when companies are utilising AI to determine churn. Subscription-based companies are obsessed with understanding churn and preventing a customer to go to a competitor. They often focus on retention strategies, spending a lot of time and investment on building sophisticated models that tell them who are the customers who are most likely to or are on the brink of leaving in the next billing cycle. While these predictions can be useful, it may be more beneficial to understand out of my existing customers, who is most likely to be persuaded by this new offer instead.
The AI can tell you who will be your potential customer, and who is your competitor's customer, but that is not the question that is most beneficial to ask, instead it is who is the persuadable customer? The consumer that if targeted by your communication today is most likely to choose your product or service. While the prediction can provide you with a behaviour (what the user likes) ultimately the decision is to change that behaviour, not predict. It all boils down to decision making, dissecting the problem we are trying to solve.
When we adopt new technologies for making predictions it becomes increasingly easier to miss the business objective that we started with. If a marketer decides to target a specific audience based on prior experience or gut feeling they will be taking into account the persuadable user, but if we want to adopt AI we're looking to be more precise about this decision, to leverage on the data available and separating the decision-maker from the prediction maker. When this happens and neither speak to each other, this is where misalignment peaks. The data scientist will proceed to analyse and predict, on the other side of the room the marketer will be thinking in what way can we change the behaviour. The AI is not the problem here, structure and cross-collaboration are.
There will always be uncertainty when testing new technologies, what is important for us to understand is what is the potential outcomes of these uncertainties, what is the potential cost if a decision-making framework is not properly integrated with those who will ultimately be running it.
Now that we've explored AI prediction and decision making, how do both of these areas play a role in digital bias? Up next we'll be dissecting how organisations should not overlook the importance of bias in order to drive brand impact. Schedule a chat with us to learn more.
