Imagine you ask your driving app to map the fastest route from Silicon Valley up to wine country in Napa Valley, but it can only get you as far as San Francisco. From there, you either have to figure out the rest of the way on your own — or settle for a wine bar in the city.
That’s the kind of business guidance out-of-the-box artificial intelligence (AI) algorithms often deliver: They can take you in the right direction, but they might not get you to your intended destination.
For brands, the whole promise of AI lies in its ability to discover hidden insights from your unique set of customer data to provide extremely precise, granular guidance. However, the time and cost to wrangle and model data have forced some brands to turn to out-of-the-box AI.
Out-of-the-box AI can make sense for some brands. For example, startups with limited data on which to train models or smaller brands that are using AI for the first time and currently lack the resources required to build models using their own data. In these cases, out-of-the-box AI can help discover net-new value quickly. It can also speed data prep time by focusing on only a limited set of their data and requiring no specialized data science resources to train or refine algorithms.
However, the speed and convenience of out-of-the-box AI come at a cost, and it is important to understand what they are before you check off the AI box forever. Basically, it comes down to this: Pre-baked algorithms trained on another brand’s data may simply not apply to your business.
After all, any two brands have business models and customer bases with unique traits, even when they operate in the same sector. For example, they may have:
• Dissimilar customer bases with widely diverging behaviors, resulting in algorithms that are less precise.
• Differing product and marketing priorities and strategies, making algorithms less relevant.
• Disparate systems that organize and format customer data in different ways, resulting in outputs that are less accurate.
• Potentially valuable data sets that go unused because the algorithm does not consider them, making insights less valuable.
Even if an algorithm does provide value, its usefulness is likely to degrade over time as your brand offerings, business strategy, market position and customer base keep evolving.
As your organization embraces AI, consider these three reasons out-of-the-box AI may only get you part of the way to your goal -- and in some cases actually send you off in the wrong direction.
1. Lowest-Common-Denominator Data
Out-of-the-box AI models must limit themselves to the lowest common denominator in terms of data types -- the ones that every brand is sure to have at hand. Certainly, this can lead to useful findings, but it also means your brand forgoes insight from all the diverse, granular datasets you already possess or might collect in the future.
For example, a prebuilt churn model -- a measurement of the risk a customer will abandon your brand -- might not take into account varying patterns depending on the product that was purchased. A customer who has not bought an everyday item like diapers within the last month may well be at risk of churning. However, a customer who buys a hammer might not be at risk even after six months or a year. Often, generic churn models cannot take into account product-specific data since there is so much potential variation from brand to brand.
Similarly, a brand’s one-time clothing customer may share many of the same demographic and/or behavioral characteristics with that brand’s houseware customers (e.g., age, gender, geography, etc.). However, an out-of-the-box algorithm may not detect these similarities since this is a pattern unique to this brand’s own customers.
By contrast, models built on all your own unique data can detect these valuable patterns in demographic and behavioral data that the pre-built model misses out on entirely.
2. Black-Box Algorithms
Often, out-of-the-box AI also means black-box AI. In other words, marketers have no insight into how models are built or how they weight different data types. As a result, you have no way to evaluate the algorithm, let alone optimize it, by drawing on your own expertise about your brand’s unique customers, offerings and strategic goals.
With insight into a pre-built lookalike algorithm, you may be able to quickly determine that, based on knowledge of your particular brand, the algorithm is likely to provide poor guidance. If it does prove to be effective for your brand, the algorithm could provide a model to build other lookalike models using the same set of features but on other kinds of target audiences.
3. No Learning Or Results-Based Refinement
Even if an out-of-the box-model includes the right set of data types, the relative weighting of each data type can lead to inaccurate conclusions, given the inevitable differences in an individual brand’s offerings, business objectives and/or customer base. For example, it may be very unlikely for one subscription brand to upgrade customers once they have signed up, while another brand may have a range of service levels or other offerings that make upselling a rich opportunity. A single out-of-the-box model may drive net-new value for the second brand, while the first may waste precious resources marketing to customers who, in reality, are low-quality targets.
Ideally, AI algorithms should learn from their own successes and mistakes. With a customizable AI model, you are able to adjust the weighting of data according to your knowledge of your business -- and make further refinements based on the model’s performance.
Yes, out-of-the-box AI can provide value in some cases, but it is critical to understand its limitations and consider embracing AI’s true promise -- insights that grow directly out of the full spectrum of your unique set of data. Solutions that can automatically ingest, validate and organize all relevant customer data types, both historic and in near real time, will provide much stronger results: net-new value driven by AI and happier and more productive data science teams.
