Some 80% of marketing leaders believe AI will revolutionise marketing, but only a quarter say they have a confident understanding of it. ‘Artificial intelligence’ is a broad term for a series of different concepts, including machine learning, robotics, vision and natural language processing, amongst others.
Nearly 70 years ago, Alan Turing outlined the basis of machine ‘intelligence’: the ability for computers to convincingly mimic human conversation, so much so that they become indistinguishable from us. This test was declared to have been passed in 2014. One popular AI prediction is that of “superintelligence”, a state in which machines can surpass human intellect in every field of learning. This is no longer science fiction: 50% of data scientists think superintelligence will be here by 2040.
How exactly does this relate to marketing? Professor Ray Kurzweil addressed this at Cannes Lions 2017, explaining that computer intelligence can help brands move away from targeting mass audiences with the same message, and instead treat consumers as individuals. Marketing leaders have recognised that AI is the future, however deciphering fact from fiction is a challenge in itself. With more companies adopting Artificial Intelligence as part of their offering, it’s important to understand more about their services and the questions you should ask to assess their competencies.
AI systems and models
There are two types of AI systems, and you should find out which is the basis for your providers’ tech. The first is known as a ‘rule-based’ system. This is a simpler form of reasoning, where it follows a series of rules set by human programmers. It is effectively static in its level of intelligence, and is only as effective as the limits of its programmers. The second is known as a ‘model-based’ system. Model-based reasoning is more sophisticated, and is what you ideally want to be working with. These models learn from data which continually improves performance.
You should also ask your provider about the data they have used to build the AI model. Do they use full customer history, or a sub-set of customer data? A sub-set is a more limited approach, and is only designed to work for a very niche market. Broadly speaking, full customer history is the preferable dataset for model-based AI. Additionally, AI models need to be updated frequently. Anything less than daily will lead to a reduction of the model’s performance over time. Be sure to ask your provider how frequently their models are updated.
The performance of models should be measured effectively, so find out how your provider does this. The gold standard of measurement is the use of concurrent, persistent control group – ideally 10-20% of all campaign impressions. The AI should learn from these unoptimised impressions, and this will also give you a means to measure campaign uplift. If
your provider bases performance on past campaigns, their measurement will be totally unreliable.
Algorithms and targeting
Algorithms are the DNA of artificial intelligence. Find out from your provider which algorithms they use, as this largely determines their overall AI methodology. There are a range of learning algorithms, and while I wouldn’t get too concerned with which is better, your provider needs to be prepared to share the details.
Ask your provider whether their AI targets segments or individuals. Whilst targeting segments has its uses, not everyone within the segment is identical – plenty of people don’t fit their stereotype! Marketing technology is moving towards targeting individuals more effectively, so your AI provider should be working to achieve this as far as possible.
Lastly, you need to know how your AI provider deals with data. As a start, check the experience of their data science team. AI may well be a relatively new discipline in digital advertising, but it’s not a new sector: it’s worth knowing if their AI solutions are being managed by experts in data science.
AI needs to be built on and trained by very large, high quality sets of data, otherwise it is essentially useless. Be sure to find out whether your provider has a DMP which can contain and manage the data. Location, behavioural and demographic data can all be used to build more accurate customer profiles, and this will allow the AI to make better decisions as it sees the bigger picture.
Marketers need not get bogged down with the finer detail of data science and algorithms. However, a rudimentary understanding of what good AI practice is will give you the tools to accurately assess your providers and their solutions.
Stephen Upstone is the chief executive of LoopMe