Recent surveys show that although many enterprises have embraced AI in a quest to improve productivity, many of those projects have failed. Building AI that is able to reliably complete complex tasks is proving more difficult than originally anticipated. It seems however, the reason is not a new one, but rather an old one. Data quality has forever been the Achilles heel of customer analysis. AI doesn’t make data quality problems go away; it simply amplifies them.

Recently, organisations, knowingly or unknowingly, have skirted most data quality issues by focusing on channel-based decisioning. By using channel data, e.g. web data in isolation of offline data, you are able to avoid most of the data issues associated with data consolidation yet come up with a “sensible” recommendation.

True customer marketing, however, is not the sum of channel-based decisions. Creating a reliable single customer view from disparate product and channel-based systems is by far the most complex problem to solve within the suite of problems you need to solve when consolidating data. Why is it so difficult?

Product and channel systems were often not designed around customers, so they create, manage and delete data based on product and channel requirements, not on customer ones. This makes the creation of a single customer view based on their data brittle.

For example, if you update your address in a credit card system but not the savings account, the link between the two accounts might break. If you upgrade your gym membership and a new record is created, it could look like you have churned your old membership and there will be two customer records instead of one.

Online systems are built around email addresses, and offline systems are built around physical addresses and phone numbers. People have more than one phone number and they change them. Customers often intentionally do not put accurate data in each system. For example, I usually do not put the day of my birthday; the month and the year are correct, but unless it is an FI or telco, I will usually use the first of the month.

So why is a Single Customer View so important? Because if you want to use AI to create insights and personalisation, it needs to have a reliable, consistent and complete view of your customer and their transactions. If it doesn’t, then it might hallucinate to fill in the gaps or make appropriate recommendations based on a fragment of data. You can create synthetic data to train models, but you can also use synthetic data for real life decisioning.

For that, you need real data and you need a golden customer record… a customer record that stitches together all of your customer interactions into a single view that AI can use to make reliable recommendations.