Fear and Loathing in the Data - Why We Can't Have Nice Things...

We spend a lot of time talking about the future and the potential of data. Exciting times - imagine we were alive to witness the day that the promise of data as the "new oil" was really a thing. But we may have forgotten something. In the middle of all the hoopla about how we can do an AI this, a machine learning that, and a customer data platform these, we left out the basics. While we are spending time thinking about that which we shall consume we may be ignoring what we need to put in. Being functional at a basic level in the art and science of data we need to consider the full path.

Return to Basics

Far from the promised land of using data to make the right business decisions and inform the next steps is the grim meathook reality of the data we use to communicate with our customers. The first step: Before we understand how our customers understand our products and how we will change how we treat our customers we need to establish consistency in our customer data. What does a complete customer data profile look like? How is it collected, and how is it reviewed, cleaned, deduplicated, and managed? The second step: What are we saying about our products? Have we created a consistent data structure around our product catalog that allows us to communicate features, benefits, inventory, pricing, and technical detail to customers and prospects in the right format at the right time? Pretty simple, right?

Setting Expecatations

We expect a lot from the data. Specifically, we refer to these expectations with the word "analytics". But consider how your customers perceive your products throughout the full course of the customer journey and you can see how easily it might be to get distracted by all the wrong things. All the shiny pennies of pre-fab dashboards that juke and jive around things like bounce rate and time on site, etc...are hooking us on the junk of leading and lagging indicators that may not be accurate unless our customer data and product data is complete and accurate. Let's unpack: We see a marketing team at the altar of analytics considering a new button placement, running an A/B test on a new marquee image, or doing anything to get that engagement metric they are looking at to go up. Meanwhile, product data is either incomplete, out of date, or difficult to find. An anonymous user fails to take the next step in becoming a prospect or customer because the product catalog is not getting the same level of attention that the UX/UI is getting. This is not to say that UX/UI considerations aren't important - but their value increases exponentially when data is complete and accurate.

From a customer data perspective, we may be getting a vibe that our list is "fatigued" or that our most enthusiastic customers are no longer interested. But which version of the customer are we imagining here? Are we imagining the customer that was at Company A 10 years ago, or the customer that we have kept up with kept their data clean, and either flushed them out of the system when they stopped buying or found a way to recover them by finding and engaging them in their new location, or simply reminding them that "we are still here with great products that are exactly what you might be looking for". How we maintain customer data is a critical piece in being able to use data to take action. Nothing we can engage the customer with will be the right thing if our record of the customer is out of date, duplicated , or inaccurate. Creating standards around simple but intense functions like deduplication means making decisions and taking some risks. Leadership may not want to hear about list attrition. But they may want to hear about increases in the right kinds of engagement.

Getting Started

It can be difficult to know where to start - so here are a few steps to truing up these two pillars of your data stack.

1. Identify systems of record and respect those systems of record - customer and product data should have one authoritative source of truth for critical data points. For products, it will often be enterprise resource planning or product information management. For customers enterprise resource planning and customer relationship management. These are the only places where this data should be modified. A myriad of spreadsheets floating around email threads is a good leading indicator that we have strayed from the systems of record and have now gone rogue with critical information. Pick the source of truth and stick to it - respect it.
2. Define complete records - establish what you really need for a complete picture of a product or a customer. Dump the rest. One field, one value. Eliminate "multi-purpose" fields that mean different things to different stakeholders. You either need it, or you don't. Normalize it into the record to establish it as a requirement.
3. Identify the channels and formats that the data will need to be consumed by - allow knowledge of the channels you will publish data in to drive decisions you make around what "complete" means for each record. Then decide how much of a complete record belongs in each channel. The whole thing will always be there in your system of record. You are publishing and syndicating the right elements at the right time.
4. Standardize and sanitize - to judge the completeness and accuracy of data, there need to be standards for what the data needs to contain. The details are critical. Capitalization, spelling, and consistency of formatting all count here. Define that schema and sanitize all data to that standard. If a customer fills out a form as "christopher" instead of "Christopher" correct it - capitalize that first letter - you will reap the benefits when christopher realizes that you care about his data and aren't just parroting back what he put in the form. You won't catch everything, and you certainly won't be able to help your relationship with test123 by calling them Test123 - but hey that one is on them. Decide what characters and formats are allowed/not allowed and enforce them. Avoid allowing free text strings in places where controlled vocabulary or numbers should be the value. Be ruthless in enforcing the standards. Remember, spacing, capitalization, formatting, removal of special characters, all matter.
5. De-duplicate - establish a standard for what constitutes a unique record, and then de-duplicate based on that. You will lighten your overhead and have a more accurate picture of performance - especially with customer data. Duplicates hide problems and opportunities and sometimes the decisions are difficult. Assess the risk of being wrong, decide, and then commit.

Summing Up

Doing the work and putting in the effort will pay off in understanding what your analytics really mean. After all if your assumptions about your own products and customers are not accurate, what meaning do even the fanciest analytics actually have?