Just like everything else you buy, data-based products (such as segments, attribution models, targeted media campaigns, or PMPs) don’t just suddenly show up on the shelf at the store. Getting there is a complicated process. But, if you know how it works and what bumps and pitfalls to look out for, you can get the quality you need.

Using the creation of a segment as an example, Jake Moskowitz from the Emodo Institute explains each of the seven steps in the process and shows you how to tell if companies are being responsible with the data that goes into their products.

Step 1: Occur

A data point starts as a moment of occurrence or creation, for example, when a device is seen inside a brick-and-mortar store. But before you can trust that this data point is correct — that this person was truly in a specific store — you need to understand its potential imperfections. For instance, the GPS might not be accurate, like in a mini-mall where stores are incredibly close together. Or, since phones are not tracking users every second, they could walk into a different store or drive away, and the last known location would be wrong.

Step 2: Categorize

The data occurrence that’s collected is raw and doesn’t mean much until it’s categorized. For the occurrence in the brick-and-mortar store, the data point would be lat/long, not the actual name of the store. One must know which store corresponds to a particular lat/long. So. it’s equally important to verify the quality and accuracy of both the data point and the POI database, or whatever database is being used to categorize data points.

Step 3: Define

Someone has to decide who actually qualifies as a shopper in that brick-and-mortar store. Is it someone who’s been in a shop once in six months, twice in a week or something else? You need to understand how stringent the definition is and what qualifies a device as being located in a segment.

Step 4: Expand

Using look-alike models to expand the size of a segment once it’s defined can help drive scale and value. However, if the look-alike parameters are too loose, you’ll end up with too many devices that have never been inside the store. If they’re too stringent, you may not achieve the scale you’re looking for.

Step 5: Match

The company creating the segment has to get it loaded into a place where you can use it. It can be a multi-step matching process, and match rates aren’t all the same. You want to minimize loss and inaccuracies from matches.

Step 6: Cross-platform

Very often you’ll need to access your consumers across mobile, web, in-app, CTV and other platforms, but any time data runs through a cross-platform database you end up with inaccuracies and data loss.

Step 7: Use

The data point’s journey ends with a human deciding which segments get used and in what amounts, and what percentage of the media is targeted to different segments. If you or a partner uses the segments in a non-optimal way, just to achieve scale, it will compromise the quality and accuracy of the campaign and increase the data waste.