Source of Truth

Imagine your address is listed in four different lists at a school: at the front office, with your homeroom teacher, in the library, and at the sports club. You move and only report it in one place. From then on, the four lists contradict each other, and no one knows anymore which one is correct. A source of truth solves this problem by designating a single place as authoritative. Only there is the address changed, and all other lists retrieve the value from that place. The term comes from software development, but refers to a very general principle: for every piece of information, there is exactly one place that is right.

What goes wrong when there are multiple truths

In large companies, data almost always exists in multiple copies. Accounting has its own customer list, sales has its own, and support has yet another. Each list was correct at some point. Over time they drift apart because changes don’t reach everywhere. This is called data inconsistency, meaning contradictory information about the same fact.

The consequences are rarely dramatic, but they are costly. Invoices go to old addresses. Two departments report different revenue figures to management. A customer receives duplicate mail because their name is spelled slightly differently in two systems. Employees then spend hours figuring out which number is correct instead of working with it.

With AI, the problem becomes even more important. A language model like ChatGPT answers based on what it has read, without verifying anything. If you feed it contradictory company data, it confidently passes on the contradictions. A clean source of truth is therefore often the real groundwork that must happen before an AI project can even meaningfully begin.

One leading system and many copies

Technically, one system is designated as the leading system. Usually this is a database, meaning a program that stores data in an organized way and logs changes. Only this system is allowed to modify the information. All other programs read the value or receive a copy of it.

These copies are called caches or replicas. They are permitted and often necessary because accessing the central database takes time. The crucial rule is: a copy must never invent a change on its own. It regularly checks in or is notified whenever a change occurs. If it deviates, the leading system always wins.

A common misconception: source of truth does not mean everything must live in one single giant database. Responsibility is usually split up instead. The HR system is the truth for salaries, the warehouse system for inventory, the customer system for addresses. What matters is only that it is clear who is responsible for each individual piece of information. Likewise: a source of truth does not guarantee correctness. It guarantees uniqueness. If you store a typo centrally, you have one unambiguous typo.

From Git to the customer service chatbot

The principle is most well-known in software itself. Programmers store their code in a shared repository, often using the tool Git. This repository is the source of truth for the code. What a developer has locally on their laptop only counts as real once it has arrived there.

In everyday life, you encounter this principle with online banking. The account balance in the app is just a display. The truth lies in the bank’s system. That’s why a payment may first appear as pending and hours later show differently – the display fetches the authoritative status afterward.

In business news, the term usually comes up in connection with data projects and AI rollouts. When a corporation reports that it is building a central data warehouse, meaning a large collective storage system for analytics, it is almost always about a source of truth. AI assistants in customer service also rely on this: they access a verified knowledge base instead of formulating answers freely. This keeps it traceable where a piece of information came from.