Time is compressing. The attention span of individuals and organizations is becoming ever shorter. There are only brief windows that can be devoted to any given transaction.
Expectations around situational knowledge, updated in real-time, are compounding. Customers expect the applications they interact with know where they are, what they are looking for and what their preferences are in terms of price, value, design and timeliness of delivery. If the proper options aren’t provided immediately, the customer will move on to another provider.
While the time available to an algorithm or application is contracting, the complexity of the tasks for successful interactions is increasing. Digital businesses must do more in ever shorter time windows. Differentiation in the digital world has a lot to do with a deep understanding of the customer based on multiple data streams and driving decisions that provide a deeply personalized experience.
In business, we need to have a current picture of our situation. We build models and analytic dashboards, but these fall short in allowing us a real-time view. The latency of these models may be measured in seconds, minutes, hours, even days and weeks. As latency expands, the picture is not of the moment and that difference defines operating risk. The phrase “time is money” has never been more accurate.
Brokerages require as close to a real-time view of positions allowing them to manage risk more closely, in terms of their aggregate position, based on a transactionally correct view of precisely what they have on their books to respond in real-time to global market changes.
For retailers, it means knowing the content a customer has seen and the environment where they encountered it. Did they just come from a brick and mortar store? What digital journeys have they been on? What online content have they been perusing? Continually adaptive models use the answers to these questions to select active customers and recommend options that address what they are searching for.
The digital shopping experience is composed of multiple linked processes, working in concert to continually understand the customer experience. Different system elements–user profiles, decisioning, recommendation engines and machine learning and AI–require data across different time windows. Some of these systems deal with terabytes of data and some petabytes. Some require absolute consistency and for some eventual consistency is good enough. In many of these systems, there is a demand for hyperscale within tight time windows.
The ability to take in more data within a given time window feeds the currency of the context. It is not enough to have a lot of data, but you also need the most current view of that data. Operating at hyperscale enables an organization to deliver exponentially more data than traditional databases. This makes models more accurate and, by extension, more valuable.
Scale Plus Speed Equals More Time
In some sense the combination of scale with speed creates time, giving you the ability to do more within a given time window. Consider this a dividend, allowing the use of more data or allowing the process to be completed in less time.
The systems that interact directly with either customers, employees or clients are edge computing systems, pulling in and processing data within relatively tight time windows. Edge systems may have expiry windows where data is deleted after a given time window where it is deemed to be stale.
The data collected and processed at the edge may be moved to a system of record or core transactional store and kept for longer periods providing historical context. These data sets may grow to petabytes in size while still needing to support response times measured in milliseconds. Meeting these requirements requires hyperscale capability, where the performance of data access does not degrade as the repository of information grows.
Here’s a real-life example. We must match the information collected regarding devices, identity and patterns that establish context against our database of user profiles. We want to run more sophisticated recommendation engines based on these patterns to ensure we are putting the closest thing to what the customer wants in front of them. We also want to make pricing decisions in real-time, to gain the largest basket at the best margins. We may have established a model that tells us customers with this profile are likely to make additional purchases after this transaction, so our analysis and optimization boundaries go beyond just this transaction. We will want to log this so we can track this behavior to validate the model. We have these processes going on, some in parallel, others in sequence:
- Establish an identity profile of the customer.
- Match that profile against a database of profiles that models likely desires and behaviors.
- Based on that match, determine the products and/or services to recommend.
- Select the content and media types based on the user profile.
- Determine pricing based on the user profile and the transaction profile.
The last element, real-time pricing, is something we want to modify to take into account the margin and profitability of the customer over time. We also know that if this impacts the overall window of the transaction the customer has a higher likelihood of abandoning the transaction and we can lose out on not just this transaction, but a whole set subsequent to this one.
Doing Better for Both the Customer and the Business
By building models based on hyperscale capability, 10 milliseconds are saved here, another 20 milliseconds there. This creates the time bandwidth in the overall process to add the notion of optimizing pricing around not just this transaction, but the value of this transaction profile and the relative margins of the various components in this sale and other subsequent sales that are likely to happen. We have created time to do this additional processing and do better–better for our customers and better for our business. And we know that as we add more data points to our profiles and scale up to include more data overall, we won’t experience performance degradation. These time benefits will remain. This is the real-time innovation dividend.