Bad data rarely makes an entrance. It does not crash through a window or pull a fire alarm. It waits in a renamed column, a duplicated record, a stale feed. Then it surfaces three decisions downstream, wearing the respectable clothes of a dashboard. By then, the engineer who caused it may not know, the analyst who found it may not own it, and the executive reading it may have already acted.
Maarten Masschelein has spent most of his working life in that awkward distance between a number and the people who must trust it. He is the co-founder and CEO of Soda, a company that makes software for testing, monitoring, and managing data quality. The technical vocabulary is precise. His larger argument is plain: responsibility should travel with the data.
That idea gives Soda’s story its shape. It explains why the company opened its tooling to engineers, why it developed a human-readable checks language, why it later emphasized contracts between data producers and consumers, and why Masschelein keeps returning to teams, accountability, and shared purpose when asked about software. Underneath the code sits an organizational nuisance. The person who understands what a field should mean is often not the person who can test it.
A childhood with a before and after
Masschelein grew up in an entrepreneurial family. The household gave him unusually early access to computers, around age seven or eight. He remembers childhood as two periods: before the computer and after it. The second period became a point of no return. He spent hours at the machine, then took that curiosity into software and data at university.
School supplied technical foundations. Work supplied the missing subjects: people and process. He entered the data-management world at Collibra around 2010, when chief data officers were still unusual and large organizations were trying to understand their growing estates of information. He stayed roughly seven years, including a spell in revenue operations. Two of those years were spent in New York.
The experience revealed an industry rich in governance ambitions and poor in everyday ease. Data work crossed departments, yet much of its tooling remained specialized. Rules lived with business experts. Implementation lived with engineers. Problems appeared downstream, where the original context had thinned to a ticket and a red icon.
“Think on an international scale from the get-go.”Masschelein’s advice to Brussels founders
He returned to Brussels in January 2018. In April he established Soda. Tom Baeyens, an engineer with deep open-source and workflow experience, joined as co-founder the following month. Their pairing was usefully asymmetrical: Masschelein brought years inside data governance and its commercial realities; Baeyens brought the instincts of a software builder who had already cultivated developer communities.
An alarm is not an answer
Soda’s early target was the silent issue. Software engineering had normalized unit tests, application monitoring, and incident response. Data teams often discovered breakage when somebody complained about a report. Masschelein and Baeyens wanted data pipelines to reveal their own trouble sooner, before a flawed input became a flawed recommendation or an executive decision.
The distinction becomes clearer through Masschelein’s own frustration. A failed data product once prevented him from making a call. It was not a catastrophe, he recalled, but it was annoying. That small moment contains the whole market: a product can remain visibly online while its useful decision-making layer has quietly disappeared.
Detection, however, creates its own temptation. Build a handsome dashboard, count the broken records, and admire the accuracy of the complaint. Masschelein’s newer writing pushes beyond that posture. If four thousand records fail on Monday and the same four thousand fail on Friday, observation has produced no cure. Someone still needs to find the cause, correct the right data, and check that the correction did not create a fresh problem.
The work after the red light
This is where the product story turns into a management story. A data engineer can detect that a field contains null values. A sales operator may know whether null is acceptable. A finance leader may know the consequence of being wrong. Reliability improves when those people can express, approve, and enforce an expectation together. Soda calls these expectations data contracts. The phrase is sober on purpose. It asks not merely whether a table is healthy, but what the people on either side have promised one another.
Give the rule to the person who knows it
In 2022, Soda introduced Soda Core and SodaCL, a low-code, human-readable language for quality checks. The premise was not that everyone should become an engineer. It was that more people should be able to participate in describing reliable data. A check might specify an acceptable row count, a missing-value threshold, or an expected pattern without hiding the policy inside an obscure system.
The open-source path also matched Baeyens’s background and Masschelein’s instinct for democratization. By 2023, Soda reported nearly 1,400 GitHub stars, 750 teams using its core tooling in production, and a Slack community of 2,000. Numbers date quickly in developer ecosystems, but the direction was the point: invite practitioners into the product’s language and let their use sharpen it.
Masschelein’s preferred formulation is that “data quality is a team sport.” It avoids the familiar corporate trick of making something everyone’s concern and therefore nobody’s task. In his version, shared responsibility requires named roles, clear agreements, and tools appropriate to each participant. Collaboration without ownership is just a crowded meeting.
A common purpose only works when the operating principles survive contact with Monday morning.
Battle scars, good manners, clear purpose
When Masschelein describes a strong team, he begins with the individual: deep skill, experience, a few battle scars, and a balance of intellectual and emotional judgment. He also values people who are pleasant to have around and comfortable with accountability. The dream-team version adds complementary strengths, a clear common purpose, ambitious goals, strong leadership, and operating principles that colleagues can use to hold one another to account.
It is an operator’s answer, not a poster. It makes room for personal development, but it does not pretend talent automatically coordinates itself. Masschelein has watched the same issue play out in modern data organizations, where titles multiply faster than shared understanding. Data engineers, analysts, scientists, product managers, governance leads, and business users can all be capable while the seams between them fail.
He is candid enough to keep the theory human. Before hosting his own interview series for Soda, he told his network he was excited and “somewhat nervous.” Asked to choose music for a data-team conversation, he picked Daft Punk because the songs make him happy with friends. “Harder, Better, Faster, Stronger” also offered an irresistible operating metaphor: deliver quickly, at scale, but with enough control that speed does not turn mistakes into policy.
When the consumer is an agent
Soda’s 2025 acquisition of NannyML extended the company’s view from conventional data monitoring into machine-learning systems. The strategic line is easy to follow. If data products need reliable inputs, models and autonomous agents need them with greater urgency. A person who sees an order total of zero may pause. An automated system can keep going, serenely confident in an impossible number.
Masschelein’s 2026 writing argues that AI does not erase familiar quality gaps. It changes the speed and method of closing them. Rules that once required manual drafting can be proposed from context. Anomalies can be evaluated with richer signals. Stewards can spend less time cleaning records one by one and more time defining what the business means. His aspiration is a platform that helps human expertise become executable without forcing every expert to become a programmer.
There is a neat continuity here. The child absorbed by a computer grew into an executive preoccupied with what computers miss. The governance operator learned that policies are weak when separated from practice. The founder paired a business vocabulary with an engineer’s open-source craft. The product expanded, but the question stayed stable: how can the person who knows what “good” means act before bad data travels?
Masschelein often returns to focus. Pursue the goal; decline what pulls the company off its path. Think internationally from the beginning. Those rules sound almost severe, yet Soda’s central idea is generous. More people should be able to understand, question, and improve the data they use. The quiet failure should have an owner. The contract should be visible. And when the dashboard smiles politely at a bad number, someone - human or machine - should know enough to object.