The well had a history. Somewhere among decades of drilling reports, production records and intervention logs were the clues an engineer needed: what happened at a particular depth, when it happened, and whether the same trouble might appear in the next well. The usual route to an answer ran through several systems and, quite possibly, a colleague who remembered which report mattered. MAANA’s ambition was to make that memory searchable.
- MAANA connected industrial data and expert reasoning in a computational knowledge graph.
- Its customers included Chevron, Shell, Aramco and Airbus; engineers and operations teams were the intended users.
- A WellLine demonstration drew on records for more than 54,000 Gulf of Mexico wells.
- The company moved from on-premises software toward Azure, then was acquired by SparkCognition in 2021.
There is a comic imbalance in modern industry. A company can track a pump by the second and still struggle to learn why an experienced engineer distrusts its next scheduled repair. Sensor readings live in one system, maintenance notes in another, and the engineer’s reasoning may never be recorded at all. MAANA, founded in 2012 by Babur Ozden and Donald Thompson, was built around that missing connection. Its Knowledge Platform gathered facts, relationships and decisions into a graph that software could query and specialists could inspect.
A search box is an insufficient witness
At its 2015 public launch, MAANA looked like an enterprise search company. Ozden said conventional information retrieval gave unsatisfying answers across new industrial data sets. Search might find a well report containing the word “kick.” It would not necessarily connect that event to a depth, a neighboring well, a later intervention, or the expert who understood its cause. MAANA’s proposed improvement was to turn those relationships into first-class information.
The distinction sounds technical until you ask a practical question. Which wells had a similar incident at a similar depth? A traditional database may keep the event under each well’s record. MAANA’s graph treated the well, depth, event and document as linked entities. The difference is the difference between finding a file and reconstructing a sequence of events. For the engineer making a drilling plan, sequence is everything.
WellLine made the theory visible. Led by Jeff Dalgliesh, a MAANA director who had spent 18 years at Chevron, a three-person team assembled the application in less than a summer. Its demonstration loaded public Bureau of Ocean Energy Management material covering more than 54,000 Gulf of Mexico wells. On screen, a well’s notable events appeared as a chronology; engineers could search across documents and systems instead of opening them one by one. The team reused React and GraphQL, software associated with Facebook, alongside natural language processing from Stanford. The cleverness lay in applying familiar tools to a domain whose abbreviations can defeat an outsider: POOH and RIH mean something quite specific to a driller.
The other database had a pulse
Ozden later described the company’s founding insight more plainly. Industrial firms were moving data from isolated systems into central platforms, yet “the most important data in AI sits in people’s heads.” MAANA wanted to bring that know-how into the same working model as documents and measurements. Its platform offered visual tools for subject-matter experts to encode decision processes, then collaborate with data scientists and developers on applications. That is a more demanding task than making a prettier dashboard. It requires people who understand the operation to say what the relationships mean.
Shell offered a revealing test. Its teams examined whether MAANA could help connect a crude-oil choice to corrosion risk in refinery equipment. Another proposed application examined health, safety and environmental incidents, including near misses whose outcomes could have been worse. The raw material included database entries, technical papers, reports and emails. MAANA’s DocAssist could extract information from unstructured documents, while experts shaped it into knowledge models. The aim was to give engineers a better basis for decisions about maintenance and risk. Public accounts describe pilots and applications approaching implementation; they do not establish a measured reduction in corrosion or incidents.
“The most important data in AI sits in people’s heads.”Babur Ozden, in a 2021 interview
The company’s research followed the same principle. A 2019 paper on MAANA’s Meta-learning Service argued that machine-learning projects often lose the reasoning behind a model after the model is delivered. The service recorded choices made while searching pipelines and tuning parameters, leaving a trail that another data scientist could inspect. Even here, the product’s distinctive interest was less in an algorithm’s score than in preserving the judgment around it.
A customer can be a better investor than a stranger
The capital story is unusually instructive. Ozden said traditional venture firms were difficult to persuade: heavy industry and its long sales cycles sat outside their familiar investment pattern. Corporate venture groups reacted differently because their colleagues lived with the problem. Chevron, GE, Intel, Shell and Saudi Aramco affiliates joined the backing; Aramco’s venture arm led a $26 million Series B in 2016. Public reports put the 2017 Series C at about $28 million. These were large checks, but Ozden emphasized something harder to buy: introductions to the people who could authorize trials, pilots and contracts.
The arrangement also says something about MAANA’s market. This was enterprise software for large, complex operators, sold through substantial contracts and tailored projects. At launch, customers generally ran it in their own data centers under annual licenses. By the 2021 acquisition, its buyer described multi-year agreements with Chevron, Shell, Aramco and Airbus. Customer pricing and the acquisition price were not disclosed. The real cost to copy MAANA’s approach therefore cannot be reduced to a software bill: it also includes integration work and time from the people whose expertise gives the graph meaning.
Pick one expensive decision first. List the evidence scattered across systems, then sit with the experts who can explain its context. Model those relationships before promising an enterprise-wide answer.
When the infrastructure moved, so did the graph
MAANA’s own architecture had to adapt. Ozden said the original platform was designed for on-premises deployment. As industrial customers shifted their digital plans toward cloud systems, MAANA rebuilt in 2016 with Microsoft Azure in mind. A 2019 collaboration connected the Knowledge Platform to Azure services and development tools. Ozden said more than 80 percent of customers and prospective clients were then using Azure for digitization. The statement is a company figure, but the change of direction is clear: a knowledge layer has to meet data where its users are willing to keep it.
That position placed MAANA between several markets. Enterprise search helped people find documents; data platforms stored and processed them; industrial AI tools optimized particular assets. MAANA tried to make a connective layer that carried business context from the first two into the third. Its breadth could be an advantage for a firm with many repeat decisions and committed domain experts. It could also demand considerable modeling work where records are poor, experts disagree, or the prospective decision is too small to repay the effort. A graph does not settle a dispute merely by drawing a line between its participants.
In July 2021, SparkCognition acquired MAANA. It cited the computational knowledge graph, industrial expertise and customers as the attraction. MAANA’s corporate independence ended, but its central question has only become more pressing as companies generate more data and more AI outputs: what does the organization actually know, and can its people inspect the path from evidence to decision? The answer, MAANA suggested, begins with a well’s history. Somewhere in that history is the fact an engineer needs. The difficult part is teaching software why it matters.