Safe Software Built a $100 Million Business Translating the Data Nobody Else Wanted to Touch

The least romantic sentence in enterprise technology may be this one: the file will not open. The coordinate system is wrong. The utility database calls a pipe one thing, the mapping system calls it another, and the contractor has emailed a CAD drawing from 2009. Somewhere, a very qualified person begins copying rows by hand. Safe Software built a business around preventing that moment.

Its product, FME, connects data sources, changes their structure, checks their quality and sends the result somewhere useful. A city can sync service requests with work orders. A utility can combine weather, vegetation and asset maps to assess wildfire exposure. An engineering team can move terabytes of design files into the cloud while preserving metadata. The work is technical, consequential and mostly invisible when it succeeds.

That invisibility has paid. Safe says more than 25,000 organizations and 200,000 people use FME across at least 125 countries. In fiscal 2025, the privately held Canadian company passed $100 million in annual revenue, up nearly 20 percent, and added 345 customers. It now employs roughly 350 people. The odd part is not that enterprise data integration became valuable. The odd part is that Safe reached this scale by beginning with the fussiest data of all: maps.

The nuisance worth owning

Don Murray and Dale Lutz founded Safe Software in 1993 in the spare room of a suburban British Columbia home. The name was a wink at SAIF, the Spatial Archive and Interchange Format. Their early product was called the Universal Spatial Data Translator. It did what the label promised: moved geographic information between incompatible formats without requiring customers to rebuild it by hand.

Spatial data was an excellent wedge because it was harder than ordinary tables. Geometry, projections, rasters, CAD drawings and location attributes all had to survive translation. A merely adequate connector could produce a clean-looking mistake. Once Safe learned to handle those edge cases, ordinary databases, business applications, cloud stores and real-time feeds looked like adjacent territory.

The first international customer arrived in 1997: Finland's National Land Survey. Esri Sverige became an early reseller the same year. Those two events foreshadowed the operating model. Safe would build the product and a network of specialists would sell, teach and implement it. The founders had spotted a trap early: consulting revenue rewards more billable time, while software improves when the same solution can be reused. Today the partner ecosystem exceeds 140 companies.

“Data should work for you, not against you.”Safe Software's operating belief

A visual factory for data

FME's interface resembles a factory floor drawn as a flowchart. A user drops in a source, connects transformation blocks and routes the finished data to a destination. The blocks can rename fields, repair geometry, filter records, call an API, apply an AI model or trigger another system. Users can pause the line and inspect what is happening mid-process. The point is not to abolish technical thinking. It is to let the person who understands the roads, pipes or permits express that thinking without commissioning a fresh application.

The current platform has three layers. FME Form is the authoring environment. FME Flow deploys workflows as scheduled jobs, event-driven automations, streams, web apps and API endpoints. FME Realize extends the stack into augmented reality for field work. Flow can run on premises, in Safe's hosted cloud, across a hybrid setup, and now through AWS Marketplace. In late 2025, FME also began executing inside Snowflake, bringing transformation to the data rather than hauling sensitive or enormous datasets elsewhere.

Customers buy fewer handoffs

Safe's strongest customers live where location and operations collide: local government, utilities, transport, telecom, construction, engineering and natural resources. Its public customer pages name McDonald's, Chevron, HERE, Deutsche Telekom, Transport for London and the U.S. Army Corps of Engineers. The breadth matters less than the recurring problem. Each organization owns valuable information marooned in specialist systems.

In Naperville, Illinois, one employee used FME to automate more than 3,300 service requests and work orders in the first year - work Safe says would otherwise have required three specialized staff. Auckland Transport compressed a planning process from years to months. FortisBC automated wildfire-threat assessment around customers and assets. These are not generic “digital transformation” claims. They are chores with nouns, queues and anxious residents attached.

The product works best when a workflow repeats, the source systems will not disappear, and the person closest to the problem can define the rules. A single conversion can be useful, but automation is where the economics compound: fewer handoffs, less re-keying, consistent validation and an audit trail when something looks wrong.

The moat is accumulated annoyance

FME competes with broad integration suites such as Informatica, MuleSoft, Boomi and SnapLogic; visual analytics products such as Alteryx; Esri's own tools; and the universal alternative, a determined engineer writing Python and SQL. Safe's advantage is not that no competitor can draw boxes and arrows. It is the library beneath them: thousands of transformations, more than 1,000 integrations and three decades of weird format behavior.

Geospatial depth is the sharp end of that library. FME can reason about geometry and coordinate systems while also moving ordinary enterprise data. Esri's Data Interoperability product is itself built on FME technology, a revealing combination of competitor and distribution partner. For a GIS-heavy organization, that heritage reduces the number of moments when a general-purpose integration tool shrugs at a polygon.

There are limits. Safe has been placed in the “Niche Player” quadrant of Gartner's data-integration market for several years. The company's own response is refreshingly direct: it is not yet the most comprehensive option for every enterprise scenario. Global buyers may prefer a larger suite already bundled into their stack. Developer-led teams may value versionable code over a visual canvas. A moat built from edge cases is deep, but it is not automatically wide.

The price of becoming a platform

Safe was profitable from inception, according to the announcement of JMI Equity's strategic growth investment in January 2024. The amount and valuation were not disclosed. That sequence is notable: outside capital arrived after 30 years of customers, not before product-market fit. The stated target is now $250 million in annual revenue by 2028.

The broader platform remains custom-quoted. Safe says more than 40 percent of revenue goes back into development and frames pricing as a long-term relationship. That promise met friction in 2025, when legacy Desktop and Server products moved to Form and Flow maintenance pricing after more than 20 years without an adjustment. Some community members reported steep increases and questioned the lack of public prices.

What failed first was not the software. It was the old bargain between a specialist vendor and customers accustomed to old maintenance economics. Safe changed its model because the product had become a broader platform; customers judged the change against the narrower tool they originally bought. Both views can be rational. The practical lesson is less comfortable: if trust is part of the product, a pricing migration spends it.

AI needs clean plumbing, too

Safe's AI pitch avoids asking customers to choose one model. FME can call external AI services inside workflows, while AI Assist helps users build them. In 2026, Safe added Model Context Protocol support to FME Flow, positioning it as a governed layer through which AI agents can reach approved enterprise tools and data. The selling point is control: an agent should not need a free-roaming credential simply to look up a work order.

This is a credible extension because AI inherits every integration problem beneath it. A model cannot reason reliably over data it cannot find, interpret or trust. Yet the “Any-AI” label will earn its keep only if FME makes governance and orchestration materially easier than coding directly against each model and system. Connectors create permission to enter the conversation; they do not settle it.

FME Realize offers a smaller example of how Safe changes its mind. The AR project began as an experiment. It worked, but preparing and packaging the data was messy. WebView support in 2024 let the team combine HTML interfaces, live databases, IoT feeds, maps and voice input. Then data virtualization removed more preparation. The product evolved when a demo stopped being a self-contained scene and became another view onto the same live platform.

What to copy - and when not to

The conditions matter. This model works when the problem repeats across many customers but appears custom to each one. It requires a patient appetite for maintenance, backward compatibility and uncelebrated edge cases. It also benefits from buyers who value reliability more than the lowest initial price - governments and utilities are natural territory.

It would not work for a startup chasing a quick category flip, or for a product team unwilling to support yesterday's file after tomorrow's release. FME's value comes from continuity. That is also why a price shock can hurt: customers are not buying a disposable app. They are placing durable operational knowledge inside it.

Safe Software's story is less about maps than about respecting the unglamorous boundary between systems. The company made that boundary visible, teachable and reusable. Now it is trying to do the same for enterprise AI. The outcome will depend on whether “all data” remains a precise engineering promise rather than a loose marketing phrase. After 33 years of translating stubborn things, Safe has earned the chance to make the argument.