5 Top Things to Consider When Evaluating a CDP Series
Customer Data Platforms (CDPs) have been incredibly popular for companies looking to get more out of their data. It’s a platform that consumes and unifies first-party data building a single view of each customer and making behavioral, transactional and demographic data accessible for activation. There are also many different functions a CDP offers, and every CDP is unique in their own way. With this 5-part blog series I will share my insight from my experience evaluating CDPs. With evaluating and finding the right CDP there are a few important factors not to be overlooked. I will begin with my in depth POV of deterministic and probabilistic matching.
Deterministic vs. Probabilistic Matching
When evaluating a CDP, it is important to evaluate what approach the CDP has with matching. Is it deterministic, probabilistic or mix of both? First, I will explain what deterministic and probabilistic matching is, then compare both approaches side by side, and then ending with a few questions to include in your next evaluation.
Deterministic Matching
Deterministic matching is using customer data as a unique ID to find an exact match between records from various sources (e.g., order history, lead forms, social media platforms). It requires a set of business rules to be configured, and as more data elements are brought in the more complex the rules become. Deterministic matching works great when the focus of an organization is retaining and retargeting the customer base.
Probabilistic Matching
Probabilistic matching, sometimes referred to as “fuzzy” matching, matches records based on the degree of similarity between two or more datasets using fragments of information (e.g., addresses, phone numbers, email addresses). Typically, the matched data is derived from a third-party data provider which is used to enhance customer/prospect records with demographic data. It uses a statistical approach in measuring the probability that two records belong to the same individual. There are typically three buckets; records are divided into either a match, not a match, or likely a match. Probabilistic matching is very useful for companies looking to expand their customer base acquiring new customers through targeting prospects.
Here is a side-by-side comparison of the two approaches:
| Deterministic | Probabilistic | |
|---|---|---|
| Accuracy | High degree of accuracy (>80%) | Lower degree of accuracy (<80%) |
| False Positives | Low | Moderate |
| Scale | Lacks in scale being limited to user authentication on their device | Expanded targeting reach with the inclusion of anonymous traffic |
| Matching Rules | Higher Complexity | Lower Complexity |
| Data Reliability | High degree of reliability and accuracy | Low degree of reliability and accuracy |
| Cross-Device Relationships | Lower | Higher |
| Known & Unknown Users | Works well for known users (customers) | Works well for identifying anonymous users |
| Alignment with Customer Data | High | Low |
| Marketing Strategy | Retention | Acquisition |
Closing Thoughts
Now that you understand both approaches, remember you will encounter CDPs having only deterministic matching or a combination of both deterministic and probabilistic matching. This is a core functionality impacting your goals and strategic plans. If your marketing goals include turning unknown prospects into new customers and you go with a CDP with only deterministic matching, this will not enrich anonymous user records. It is defeating the purpose of having a CDP by limiting the audiences you can activate to only your customer base. On the other hand, if retaining the customer base matters, and not reaching unknown prospects, then it may be a running contender.
I will end with a few questions to include in your next evaluation questionnaire.
Preview of the next series:
CDP & Current Technology Stack
Finding the right CDP with the best fit around the current technology stack is crucial. The selection of one CDP over another could have significant differences on what resources are needed and whether additions to or deprecation of existing platforms are needed. How well does the CDP fit in with the current technology stack? Is an orchestration layer needed, does the DMP or DSP need to be replaced? Is there support for business intelligence needs?
Ingestion & Extraction of Data
The availability of data is just as important as having access to the data. An evaluation of the ingestion and extraction to and from the CDP is important to look at thoroughly to evaluate if the company can continue business as usual. For each integration the CDP should be able to provide whether are using APIs or batch files to ingest/extract data, the formats they accept, whether it is real-time data transfer, does it support self-service, and what delays are expected.
The Use Cases
Another very important piece with evaluating a CDP, is making sure to have use cases that test a wide breadth of the CDP. Part of this is also having a good understanding of the entire customer journey to make sure the use cases cover the functionality needed. Look for potential use cases for today, and what the company is looking to do in the future. Some areas to include would be ingestion and extraction to critical platforms, basic to complex segmentation, evaluate orchestrated journeys with triggered automations to critical platforms, for probabilistic matches ensure the data is available and can be activated on, just to name a few.
People, Resources & Services
After a CDP is in place, teams across the organization will have access to actionable data and where previously siloes prevented this access to teams. There will be non-technical and technical people who need to extract and send the data to different destinations. Some CDP platforms will require more technical users with the lack of self-service tools that are easy to use. What is the self-service documentation like? Does the CDP have a good track history of great customer support outside of implementation? Is the CDP fully self-service where technical support on the CDPs behalf be needed on rare occasions?
