How to Build Trust with Data Products

For organizations to operate efficiently, decision makers should be able to make good data driven decisions. It’s fair to say that trust in data is paramount in today’s data-driven organization As AI application becomes mainstream, data quality becomes more critical. If we mistakenly train AI models with bad data, poor outcomes will inevitably result. If AI models interacting with clients perform poorly, it can have a significant impact on business and long-term brand equity. You must be able to trust your data as strongly as you trust your employees.

While trust in data is vital, surprisingly it is low. In 2020, KPMG found that only 35% of decision-makers trusted their enterprise data. Another study in 2023 by Precisely and the LeBow School of Business reflected some improvement; nonetheless only 46% of respondents had high or very high trust in their data.

Organizations make several mistakes that lead to poor data quality and the breakdown of trust:

• Publishing outdated data
• Publishing inaccurate data
• Publishing incomplete data
• Not providing the context to ensure data is well understood

These mistakes lead to bad decisions and poor business performance. For example, if an employee makes a commitment to a customer based on flawed data and is unable to fulfil that promise, the company risks losing not just the customer but also its reputation.

Losing trust is easy but, once lost, gaining it back is very difficult. Once executives consume bad data or review reports with errors, their confidence in future data is shaken.

Why Data Products are More Trustworthy than Data Pipelines

Improving data trust requires a fundamental shift in the way we access and use data. Traditional process of accessing data has been enabled by launching individual projects and building ETL pipelines. The success of these projects is measured on how much code is developed and how much data is delivered. Data quality, while important, is just part of the equation. In a project-based mindset, output quantity is the measure of success.

However, when we shift the objective to business outcomes, data quality becomes the core measure of success. Trust is built when the goals and motivations of data practitioners align with those of data users. A successful data product is not measured by how much data it provides but by how well it meets the needs of its users. The genesis of any data product comes from a desired business outcome. Even if data products provide access to vast amounts of data very quickly, they will not be used, if they don't meet users' needs thus limiting their ability to deliver positive business outcomes. For a data product to be successful, users must trust it enough to base their decisions on its output.

Greater Visibility and Collaboration Reduces Errors and Builds Trust

The success of data products is driven by a diverse team with different functions working together to deliver superior business outcomes. This process needs to be built on trust, transparency, visibility, and collaboration among colleagues. Trust among data product teams translates to data that users can trust.

With project based ETL processes, collaboration, visibility, and transparency are a challenge. Typically, these projects originate with a data request from a user, however once the pipeline is built and data is delivered, the user may not have any visibility into how or when the data was collected.

Similarly, data engineers don’t always know how the data they deliver is being used downstream. They may be unaware how their actions will influence downstream analysis of AI models. For example, if a data engineer changes the schema of a data set, it could break some analyses or dashboards that rely on that data set.

Additionally, there is a lack of process for providing feedback. As these processes are hard-coded, it’s difficult to change and incorporate feedback, even if there is more visibility into their use.

Building Trust with Data Products

Successful data product strategies are built on a standard data stack or platform that enables greater collaboration, visibility, and transparency.

Collaboration

Enhanced communication and collaboration always build more trust and are therefore should be an integral aspect of a data product strategy.

Collaborating around data access and sharing responsibility to safeguard data creates a common understanding among data teams. Federated governance strategies, where central IT teams and domain managers share responsibility for data governance, are a core feature of data strategies built on products. The benefit of this approach is the people most familiar with the data, domain managers, are more involved in data governance. This supports a more nuanced and practical approach to data governance. Sharing data governance responsibility enables greater scalability and agility, as not every governance policy needs to be run through the IT department.

For this strategy to work, domains and IT authorities must collaborate and trust each other. Different domains and the IT department need to collaborate to define who is responsible for the governance of which data sets, where IT control ends, and where domains begin. Clear understanding and communication avoid confusion. This results in flexible and adaptable data products with high data quality that users can trust.

Data producers and consumers can also build trust through collaboration. Instead of a project-based approach where data is tossed over the proverbial “fence," data products are constantly evolving. Data product creators can regularly incorporate feedback from users. This exchange of feedback builds trust between creators and users, ensuring that creators have the business outcome of users at the center of the development process.

Users also collaborate with other users to build validity. Typically, data product marketplaces will allow users to rate data products to provide validation to data products' quality and usefulness.

Providing a platform to align participants and enable collaboration and feedback throughout the process supports greater confidence among the team and users' trust in their output.

Visibility and Transparency

Visibility and transparency are crucial for delivering reliable data products. Lack of visibility leads to errors typically in linear waterfall-type approaches which are often used to develop ETL pipelines, providing very little visibility between the data users and the engineer developing the process. If an error in the transformation process occurs, the data user may not know there is an issue and will continue to use outdated or erroneous data in their analysis. On the other hand, data engineers typically don’t have visibility into how changes they incorporate in the backend affects analysts' models.

Data products, data federation, and centralized metadata management act as a bridge between data engineers and data user. Data product managers and producers facilitate better understanding of requirements, needs, and concerns between data engineers and data consumers.

A data product production platform acts as a central place for collaboration, gaining information about the health of data products and their application to use cases or various models. The key to this transparency is a federated governance platform that tracks and manages governance policies across domains. All participants can provide inputs and gather data about data products and the data that constitute them, such as:

• Completeness: count of the number of records with incomplete or null data values.
• Validity: if data reflects reality or what you would expect to see.
• Timeliness: how up to date the data is.
• Lineage: source of the data and its trustworthiness.
• Accuracy: measures of data accuracy.
• Uniqueness: frequency of repetition of a value.

Metadata control plane is at the core of this platform providing transparency into enterprise-wide metadata. A federated data platform consolidates data from across an organization to provide greater visibility into data quality and lineage. This central repository also tracks metadata changes at the source to ensure data analysts can adapt their models and analyses accordingly. Automated alerts are delivered to the subscribed users informing them of the changes and the health of the data with this centralized tracking system.

Context

High quality data is fundamental for creating trust in data, and so is providing the correct context around this data. Business terminology is not always uniform across domains. If data users are confused about what a term means or how KPIs are calculated, errors will occur, and users will lose trust. Data glossaries are extremely helpful in ensuring that users understand the meaning of the data they are working with to avoid confusion and mistakes.

Discoverability

The ability to find the right data set for an analytics project also helps to build trust in your data assets. Evaluating health and quality metrics across all the data sets from a single pane of glass allows analysts to consider multiple variables before adding a data set to their analysis. Also, a data product marketplace that leverages AI can recommend the best data products for users. This capability builds trust that these platforms are designed to deliver not just better data, but better business outcomes.

In the age of data and AI, we will be increasingly reliant on the data we collect and base our decisions on. The ability to trust the quality of this data will have profound effects on business outcomes. Those who succeed will make data quality and integrity a top priority.

To learn about building trustworthy data products with the Avrio platform schedule a demo.