In part one of this two-part blog series we addressed the power of AI and machine intelligence and their applications in market research and data gathering. While these tools certainly have a lot of potential, they do have limitations. In part 2 we address these limitations and the best method of overcoming them.
As mentioned in part one, the first hurdle is that the AI must be trained; or in other words, the machine must learn. The ‘supervised’ machine learning method obviously requires a significant time and human resource investment in order to create the training data sets. Comparatively, the ‘unsupervised’ machine learning method would seem as though it requires little in the way of resources, however, this is not the case.
Unsupervised machine learning also requires a significant investment of both time and computing power. It takes time to process the massive amounts of data required in machine learning. Of course, this process can be sped up with a greater investment in computing power. This dynamic can be summed up with a slight modification to the old adage: Money = Time = Computing Power. Furthermore, there is a human resource investment because there should still be some level of monitoring as small errors can quickly have a snowball effect when fully automated. A developer will also be required to customize any generic AI building tools. We cannot forget either that the data itself also requires a significant investment in storage space.
But the most important limitation to keep in mind is that the AI results will only be as good as the data fed into it. The whole idea behind machine learning is that it should mimic the way in which a human brain develops and even that can easily be corrupted when the input is inaccurate. For example, imagine teaching a child that the color green was actually called 'red' – if left uncorrected this could greatly affect the way that child processes color-related instructions (like that red means “stop”). Likewise, if the data fed into the machine is somehow flawed it can skew an algorithm’s output.
So, what does this means for customer intelligence? Good customer intelligence is built on three component parts: big data, social data, and primary research. AI is capable of quantitatively managing these three components. However, this data-driven intelligence must work with human intelligence for true customer intelligence. Qualitative insights are crucial for not just confirming AI findings, but to spot things that might not show up in the data like the subtext of dialogue, relations an algorithm can't account for, and the "why" behind the quantitative facts.
AI may be the future of customer intelligence gathering but it needs a human touch.
Heather Carpenter is Marketing Manager of Silicon Valley Research Group, a global market research and strategy development firm focused on the needs of technology companies.
Do you have questions, or would you like to hear from Silicon Valley Research Group on a specific topic in a future blog publication? Contact us.
If you'd like to start a dialogue about issues you want to address with market research, click the Find Out More button. We will provide you with a free quote on research solutions tailored to your needs.