Skip to main content

Solving the Multiple Versions of Truth Problem with Semantics

Every enterprise struggles with the ‘multiple versions of truth’ problem. These meetings are expensive pressure cookers where semantics are argued instead of recommendations and actions. As someone responsible for data and analytics, I’ve felt the pain of these moments — especially when the source of confusion is avoidable.

Even in mature organizations, simple issues like mismatched timestamps, unclear filters, or inconsistent metric names cause costly misunderstandings. Sometimes the root cause is competing reports or poorly aligned definitions. Other times, it’s just a lack of shared understanding.

This article explores the role of semantics in enterprise reporting and offers practical solutions to eliminate conflicting truths, starting with the simplest fixes and building toward a sustainable approach.

Multiple versions of Truth Caused by Semantics Disconnect

Solving enterprise semantics problems with a data-first mindset is incredibly difficult. It’s like trying to solve a Rubik’s cube where the colors change constantly. There are plenty of smart folks out there taking a fresh look at this problem. I refer to this bottom up (data first) perspective where the objective function is connect data to meaning. This meaning helps describe how to manipulate the data into a business consumable format that is consist and theoretically reusable. However, I am skeptical applying data semantics without re-thinking how semantics are actually managed and applied over time, is creating the same problem in a new place hoping for a different result. It’s why I have been poking at some of the AI semantics commentary recently.

multiple versions of truth

Technology-centered perspective: For a long time, data and analytics platforms vendors have taken a self-centered view of the semantic layer. As a result, semantics are scattered and disjoined across self service reporting business applications, and now data platforms. There are of course, universal semantic layers like At-Scale that could have the right solution at the right time.

My time years ago working at BusinessObjects taught me a lot about what works and doesn’t work with universal semantic layers. Conceptually, BusinessObjects had the solution, great execution, and brilliant people. It ultimately lost market momentum while operating inside of the SAP machine.

Why Semantics Matter

Business semantics is the general corporate language or business vernacular used day to day. A customer, revenue, and onboarded are 3 terms commonly where meaning can vary based purely on who, where, and what you are disusing. This unstructured understanding of how things are understood and managed and can easily be structured into taxonomies, glossaries or more advanced structures. You can walk into 2 different companies that compete in the same industry segment and experience different semantics. Business semantics are typically disconnected from business and analytics technology.

Data/ analytics semantics aka “semantic layers” include the technical definitions, relationships, and rules that translate raw data into metrics, dimensions, and facts. This structured layer often defines how data is joined, filtered, aggregated, and labeled. This makes it possible for consistent reporting across BI tools and business functions. As the business intelligence function became decentralized and organizations purchased multiple tools, some enterprises struggle with multiple versions of truth purely as a function of disparate reporting platforms accessing the same data.

7 Simple Causes: Not Semantic Layer Created

The presence of semantic layers, even when executed flawlessly, will not always solve the multiple versions of truth problem. I will take it one step further… If you have 1 system of record, 1 semantic layer, 1 report… You can still end up with multiple versions of truth! Here are the common real-world instances that occur regularly with ultra simple solutions!

#1 Differing date/time dimensions: Example: Cohorts vs Activities.

Someone pulls a cohort based fact and someone the same fact from another tab in the same report citing a period of time. Both parties are correct!

Solution: Improve the title and apply a detailed description and disclaimer in reports.

In geek speak… One metric or fact can be expressed in 1 model. However, when reported that same fact can have multiple date/time dimensions, thus changing it’s semantic meaning.

#2 Same numbers pulled at different point in time.

Same report pulled at different point in time. In this case both parties are correct.

Solution: Put an “as of date” on the report. I still see way too many reports and dashboards missing this to omit it from the my list.

#3 Two flavors of same metric (from same self service report)

Someone looks at facts through a different lens/perspective (data is filtered differently) and two versions of truth are presented without details how it was filtered.

Solution: This is a scenario where data semantics does matter. This is alignment between people and a need for governance to capture that alignment. In my approach, this is context data that surrounds the use of a metric.

#4 Aliases and naming conventions

A power point has a key fact or slide title where the name of a metric differs from what is presented in the report.

Solution: This is a semantics issue but goes back to human understanding and alignment. Sometimes this becomes an alias (different name for the same metric), or it could be a simple one-off error.

#6 Change management

Something changed somewhere in the applications / data / analytics pipeline. If this is systemic and continuous you have to dig and discover the root cause:

  • Team capacity or training / upskilling needed.
  • Governance processes limited or nonexistent.
  • Deficient or nonexistent SD/LC processes.
  • Technology debt, sophistication, or bloat.

Solution: Not every solution here is simple after all! Every customer has its own issues delivering consistent and reliable decision support. I am always happy to help, or make connections with experts that deal with enterprises large and small.

#7 Excel exports and manipulation

Data is exported, transformed, and blended into Excel and presented side by side with other reports.

Solution: Alignment and evaluating Excel logic is typically required to get to a solution.

I have captured these + a dozen other scenarios, existing as signals to help prevent the problem.

Solve Multiple Versions of Truth with Semantics Layer Alignment

1: PEOPLE: Creating a semantic layer is the starting line and not the finish line

Teams need to find a way to close the gap between information consumers and the semantic layer creators. Rarely does a business consumer’s understanding of metrics start from the creation of a semantic layer. Newly created semantics and metrics that shed new light on the business process require even more alignment and care.

2. PROCESS: Semantic layer management done right is a governance process and NOT a technology requirement gathering process

Create a recurring governance process to align on the organization’s most important semantics, specifically metrics, KPIs, facts, and dimensions. Effective governance is not about tools or data for that matter.. It’s about structured conversations, shared understanding, and formalized communication pathways, and accountability.

3. TECH: Capturing when, how, why and where changes happen that distorts facts?

Use metadata tools to audit usage, monitor schema drift, and understand where facts get distorted. Technology alone won’t solve the problem, but it can shine a light on root causes.

4. AI: The hype is justified, but the reality is messy

Semantics of all types will continue to be a hot topic as tech/service companies fight to become your preferred AI solution. The aspiration to have an in-house AI analyst ready to answer questions with data 24×7 is an exciting one… There is a lot of innovation happening here. Most of the practical testing I have done, I would describe the AI analyst as a technically sound analyst on day 7 of their job slight amnesia.

That statement comes from someone that has pumped data semantics, business semantics, context data, report and dashboard meta data, and plenty of observations and logs from previous sessions… I will let the AI experts tell me there is a better way to test. So far in practical terms this is the picture I have in mind introducing my AI analyst to a business stakeholder who brings their AI analyst insights to a meeting only to find out it’s way off:

Semantic layer gone wrong

How do you set yourself on the right path to create a useful and correct semantic layer with current technology solutions?

If you are a data or analytics professional getting started and wanting to implement a semantic layer and want to maximize adoption, I recommend the following advice:

  1. Get out of the database and sit in on meetings where metrics are presented and debated.
  2. Conduct a metrics governance alignment meetings between business stakeholders.
  3. Put emphasis on standardizing and organizing curated reports for executive facing meetings.
  4. Create internal naming conventions, tooltips, and glossary tags for reuse.

Have more advice or experiences that you want to share? Let me know!

More info on how I am solving semantics problems at DataTools Pro

In 2023, I set out to automate the discovery, alignment, and change tracking of metrics and KPIs. The goal was simple: speed up onboarding and improve trust in metrics definitions. What we built, and continue to refine, is a metrics-first approach to semantic alignment. We released metrics analyst in early 2024 and have continued to refine in the spirit of “release early and often.” Shaped by customer feedback, failure, and iteration, we are releasing version 2 this summer!

I’m excited to solve at least one side of the Rubik’s Cube… even if the colors keep changing. Feel free to schedule a call to learn more!

author avatar
Ryan Goodman Founder
Ryan Goodman has been in the business of data and analytics for 20 years as a practitioner, executive, and technology entrepreneur. Ryan recently created DataTools Pro after 4 years working in small business lending as VP of Analytics and BI. There he implanted an analytics strategy and competency center for modern data stack, data sciences and governance. From his recent experiences as a customer and now running DataTools Pro full time, Ryan writes regularly for Salesforce Ben and Pact on the topics of Salesforce, Snowflake, analytics and AI.