As marketers begin to seek practical applications of AI, they are encountering the first challenges. However, the biggest challenge of all is data. Almost all AIs require “training”, this means preparing historical data so that it can learn what has happened in the past to predict what might happen in the future. This “predictive model” can then be applied to a live engagement to determine a likelihood of an event occurring.
To visualise this in action, think about purchasing a new bed. A predictive model could calculate that customers visiting a mattress company site who came from a Google query about a bad back will go to the back-support capabilities of the mattress page and there is a 78% chance they will purchase after reading this page with no further discounts required. Sales messages can then be tailored to focus on how the mattress will cure a bad back for the current interaction.
At the time of making a decision, the AI will require all the historical data to apply the predictive models and determine what the customer context is for this engagement. Only a few data items are actually used from the current interaction – i.e. real-time data.
Even the biggest companies find it a massive challenge to set up and format the data so that it is suitable for use by an AI like this example. This is not just a problem of sourcing the data and redesigning/re-formatting it to be appropriate, it is also a problem of training data people on how to do this and how to set up not just the initial data but to keep it flowing. Therefore, it is also a problem of organisation, process and change management. The example above is relatively simple on the face of it but there are at least four different data sources just in this made up example!
Then add to this the poor marketer who just wants to sell more mattresses. The change of processes that must happen to implement an AI must take into account how the marketing strategy and objectives are included in the development of the decisions that the AI is making and the messages that are subsequently delivered. Marketers need to understand what can be predicted and therefore what they could do with that prediction. It has the potential to get messy.
Some AIs, of course, have “self-learning” algorithms that when pointed at the particular interactions can train models for making predictions. However, that still takes time and still requires that your marketer knows what you need to predict. It requires a lot of collaboration between data people and marketing which is far from easy.
But what if you could determine the customer’s context without historical data?
To achieve this you would need a single planning tool that can map all the content against customer needs. In addition, you would need an AI that can serve the right content to the right audience at the right time across all the channels based on the maps and customer needs.
By taking this approach you would not need to create new data structures, warehouses and feeds and you would not need a large technical project to get up and running.
You would also not need to make wholesale changes across the whole organisation to implement new processes for running the new system.
Plus, all the data you need is “real-time”, so you don’t need to fetch data or fire models to make predictions.
And lastly, marketing is in complete control as you could manage the customer’s journey through the content that you create.