Is Your Marketing Organization Ready for AI?
Use AI and machine learning to get the insights you need from your data to deliver personalized experiences in real time.
Your customers leave a digital footprint wherever they go. They’re on mobile, surfing the web on desktop, downloading apps, visiting you in store, calling your customer care center, and using connected TV or even emerging technologies like augmented and virtual reality.
But understanding their behavior across all these platforms and channels is a herculean task for any marketer, and results in so much customer data that often you’re left wondering how to make it actionable and how to sift through it to see patterns or combine it for a full picture of your customer.
In recent years, we’ve been able to deliver insights into data using “big data” and statistical approaches, but there’s opportunity to do even more — thanks to artificial intelligence (AI).
AI and machine learning can help you delve into huge amounts of data to uncover insights you need to know about your customers, enabling you to identify patterns in your data at speed and scale. But to leverage this technology effectively in your marketing organization, you must develop a well-thought-out plan to guide your progress. Here’s a framework you can use to drive the next level of digital experiences for your customers.
#1: Define the business problem you’re trying to solve
Einstein once said, “If I only had one hour to solve a problem, I would spend 55 minutes defining the problem and the remaining five minutes solving it.”
Before rushing to apply AI to your data, first determine the business issue you want to solve. Be articulate in defining the problem, visualize the insights and how would you consume and act on it. Thinking about this upfront will help the entire process of getting from data to insights. Too often, people start at the wrong end of the problem. They think that since they have the data, they should make use of it — when in reality they need to define the business problem first.
# 2: Understand what data you have and how it can solve the problem
Once you identify the data sources, it’s important to focus on the fidelity of the data. Any machine learning or AI technology you use will rely on historical data. You need to make sure the data is labeled correctly, de-duplicated, unbiased, and is broad enough to cover the extremes of your business.
This is an area on which you will have to continually focus. You’ll need to think about how you can add your own domain knowledge and create new features that can help power your insights. Feature engineering is the process of using domain knowledge to extract new variables from raw data that make machine-learning algorithms work. Data on its own at times is of no value, but the insights (features) you can derive from it are priceless. For example, a customer support complaint has no significance for a machine learning or AI algorithm, but the sentiment of the complaint could be of significance.
The other major consideration consists of governance and ensuring that the data you have can be used for the purposes you have in mind. This is especially critical with the General Data Protection Regulation (GDPR) recently going into effect. The new law has global implications beyond Europe for how we protect personal data and customer privacy, so machine-learning technologies will be critical to modernizing data capture processes and ensuring customer data only is used for specific business objectives. Remember, just because you have access to data doesn’t mean you can use it.
Lastly, as you go through cleaning and transforming your data, it’s important to maintain an audit trail so you can go back and track the lineage of your data.
On the Adobe Experience Platform, we provide access to an ecosystem of extract, transform, and load (ETL) data vendors to help with data cleansing and transformations. We also provide standardized and normalized schemas for the marketing domain via Adobe’s Experience Data Model (XDM) and have built the data governance layer into the platform to label and enforce policies on your data that conform to your enterprise policies.
Businesses are complex, and you may have multiple sub-brands with different personas associated with each, so in some cases you may need to stitch data together before you’re even ready to analyze it. In our world, a profile is a graph that you’ll need to stitch together to drive insights. The unified profile on the Adobe Experience Platform helps you do just that.
#3: Apply the right technology
Once your organization has taken the steps to unify its data, then you’re in the best position to take advantage of AI and machine-learning-based technologies and services to solve your business challenges.
Most business challenges usually will fall into the following categories:
- Regression, where you predict future numeric values (e.g. predicting revenue).
- Classification, where you assign a class/category to your data based on historical learnings (e.g. predicting customer purchases, or yes/no).
- Clustering (unsupervised learning), where you identify common patterns to form homogeneous groups based on your data (e.g. creating customer segments based on their browsing behavior).
- Reinforcement learning, where you learn from the environment and are trying to maximize a reward (e.g. maximize conversion on a website).
On the technology front, neural networks (or deep learning) have come a long way and, in most cases, provide better accuracy and precision in predictions and at scale compared to more conventional algorithms. However, they come at a cost of lack of transparency (i.e. why the algorithm is making the recommendation), difficulty in tuning, and the need for more computing power and resources. You need to weigh accuracy or consistency vs. transparency when you decide on the technologies. In regulated industries (insurance and retail banking, for example), it’s critical to have transparency into model predictions. At Adobe, we are heavily investing in research to give our customers more transparency into AI.
Once you have selected the technology to use, focus on training and experimentation with different data-sets. At this stage, humans need to play their role. They need to measure the efficacy of their models, look for biases, and set-up thresholds to ensure their models perform at the same level of accuracy when unleashed for production usage.
Data Science Workspace on the Adobe Experience Platform gives you access to either use pre-built recipes that Adobe has developed to solve specific business problems, build your own algorithm from scratch, or import your own model on the platform. It offers a full-fledged machine learning and AI framework to develop, train, measure, evaluate, and deploy machine learning and AI-based intelligent services.
Using prebuilt or custom machine-learning recipes, data scientists don’t have to build models from scratch, and they can accelerate their organization’s progress into AI-driven experiences. And with pre-built recipes, our customers don’t need to focus on writing code. Instead, they can focus on their business data and features, and use them to drive insights.
However, in addition to answering the question of what technology you’ll use, you’ll also have to consider who will actually implement it. The end-to-end process is challenging. Some people in your organization may have strong machine-learning skills, but less expertise in data management. That’s where the Data Science Workspace can help you go from data to insights seamlessly, and cater to the different level of expertise your organization may have with AI and machine learning.
#4: Decide how to take action on insights from your data
The insights generated from machine-learning models — whether they’re prebuilt or custom — need to be consumed or acted on by your solutions, leaving you with the question of how best to use all this new information to improve customer growth, revenue, or other core metrics. If you’ve followed the framework I’ve outlined, this part should be easy. You would have a plan to consume the insights. On the Adobe Experience Platform, any machine learning or AI-intelligent service, especially if it relates to your customers, will automatically enrich those customer profiles with custom attributes that you can use to act on. You can use these attributes to segment customers, and then use the right meaningful experience to delight your customers.
For example, if you predict propensity scores for how likely new customers are to buy a product, you then can use this information to create segments based on levels of propensities. For each segment, you can further predict the best channel to reach those customers, predict the best time of day to reach them, or, when they come to your website, prescribe the best experience to show them. You can then orchestrate the entire experience end-to-end using AI and machine learning.
The promise of AI
AI and machine learning can be vital parts of your organization’s digital transformation. By democratizing the data you have and by making it easier to understand and draw insights from, you also can close the gap between departmental silos, marketers, and data scientists. Eventually, AI may bridge the technical gap between these two groups. Non-technical marketers may become “citizen data scientists” who are able to follow an easy-to-use interface to create prebuilt models, enabling their organizations to derive even more value from their customer data.
But this transformation all has to begin with the framework I’ve outlined. By starting small, being extremely focused, and generating small wins along the way, you’ll eventually get to a place where your organization isn’t simply just a marketing organization, but an AI-driven, experience-focused business.