A guide for the people who own the decision: what an operational data platform has to do, what OT/IT integration involves, why data quality decides whether the platform gets used, and how to choose between building your own, buying from the enterprise incumbents and running open source.
The starting point
Most industrial operations already record everything. The historian holds ten years of sensor readings, the control system logs every alarm, maintenance lives in one system and production plans in another, and the corrections that make the monthly report come out right sit in a spreadsheet only two people understand.
What is missing is not data. It is one place where a reading knows which pump it came from, which process that pump serves, and which reported numbers depend on it. Until that place exists, every question that spans systems is a manual job for your most experienced people.
A KPI in a board report, an emissions figure for the regulator. Tracing it back through spreadsheets and exports takes days, and the person who assembles it is the only audit trail you have.
The answer sits in three systems and in one engineer's memory: the trend in the historian, the work order in the maintenance system, the setpoint change nobody logged.
When a meter is replaced or recalibrated, somebody has to know every report that depends on it. In most plants that knowledge exists only as experience.
A dashboard dies the day it shows its first wrong number. Winning the trust back costs far more than building the dashboard did.
The job description
Vendors in this category disagree about almost everything, but the work itself is stable. Whatever you buy or build has to do five things, and a weakness in any one of them shows up later as a limit on all the others.
Historians, control systems, ERP exports, lab results, files on a share. Connecting a new source has to be routine work, not a project with its own steering group.
Assets, processes, responsibilities and the relationships between them, described once in a shared model, often called an ontology or a knowledge graph. This is the part that turns a reading into a fact about a specific pump in a specific process.
Industrial time-series data is enormous but highly compressible. Our platform routinely reaches around 100x compression on time-series and around 25x on event data, where 2 to 4x is the norm. The engine underneath is built for the job: ClickHouse in our setup reads over 10 billion rows per second. Storage economics decide whether ten years of second-level data is a rounding error or a budget line.
Every reported number should be traceable back through each calculation to the raw signals it came from, with data-quality flags intact. For regulated figures, that trace is your audit trail.
Engineers, analysts and AI agents should all read from the same model without knowing which source system owns which table. A platform nobody can query without a specialist is a new silo with better marketing.
OT/IT integration
OT is the control systems, the sensors and the historians: equipment that measures physical reality, built for uptime and safety, often decades old. IT is the business systems: ERP, maintenance, planning and finance, the systems that describe what the operation means commercially.
Nearly every question that matters to a decision maker touches both. What did the shutdown cost? Is energy per produced unit rising? Which orders are at risk if this compressor keeps vibrating? Answering them means joining a sensor reading to a work order, a production plan and a price, and that join is what OT/IT integration actually is.
Done properly, the integration is one-directional and safe. The platform reads from the control side and never writes back into it, and in sensitive environments the connection can run through your own network zones or with no internet connection at all.
Data quality
The expensive way a data platform fails is not downtime. It is a platform that runs flawlessly and is quietly ignored, because the people who know the plant have stopped trusting the numbers it shows.
Quality has to be measured, per signal and per dimension, and the flags have to travel with the data into every calculation. Then a wrong number becomes something the platform reports, not something a sharp-eyed operator discovers. We measure six dimensions:
Does the value reflect what actually happened in the plant?
Are there gaps, and do the calculations downstream know about them?
Do two systems agree about the same fact?
Is the data fresh enough for the decision it feeds?
Is each event recorded once, not three times through three routes?
Is the value physically and logically possible at all?
Build, buy or wait
This is a decision about years, not quarters. It deserves plain speech about all three options, including when ours is the wrong one.
A pipeline is a weekend. A platform is not. The asset model, the lineage, the quality flags, the access control and the storage economics each look small, and together they claim two engineers indefinitely. That is fine if data infrastructure is the business you want to be in.
Right when your needs are genuinely unusual and you have a standing platform team. Expensive in every other case, just invisibly so at first.
The category leaders are capable and proven, and for some organizations they are the right call. The price of that safety is a seven-figure engagement, an implementation measured in years, and storage that is hard to leave.
Right when you need a large vendor organization behind you, with named around-the-clock support and formal certifications, and budget is not your binding constraint.
Our platform is the same category with a different structure. Open source under AGPL-3.0, free to self-host, standard components throughout, such as PostgreSQL, ClickHouse, Neo4j and Apache Pulsar, so an exit stays an export job rather than a negotiation.
Built and run by four people in Stavanger: one senior engineer, two junior engineers and a marketing lead. You talk to the people who wrote the code, and the support capacity is real but bounded.
Sovereignty
For Norwegian and European operators of critical infrastructure, procurement now asks a sharper question than region names can answer: which foreign law can reach your operational data? A European region operated by a US-controlled cloud provider does not close that gap.
We answer it structurally, not with a hosting label. You choose where the platform runs, and every layer of what runs is open to inspection.
The platform runs on hardware we own in Stavanger: 40 servers, 2 PB of HDD, 400 TB of SSD and 20 TB of RAM, operated by the same team that builds the software.
A Norwegian company, an all-Norwegian engineering team, and no US entity anywhere in the chain. Data hosted with us is subject to Norwegian and EEA law, nothing else.
Self-host in your own datacenter or cloud, or run fully air-gapped. The platform is open source, so what you run is what you can read.
The evaluation
Including us. A vendor who answers these eight without flinching is worth a pilot. A vendor who answers them with a slide deck is telling you something too.
A platform you cannot touch until after procurement is a platform whose weaknesses you will find after procurement. Insist on running it, preferably on your own data.
If the asset model mainly exists in a consultant's head or on a project wiki, it leaves when they do. The model belongs in the platform, versioned and queryable.
Ask for the storage cost of second-level data at your tag count over a decade. Compression and storage engineering decide whether the honest answer is negligible or frightening.
Pick one KPI and ask to trace it back to the raw signals, with the quality flags of every input visible. This is the capability auditors and regulators quietly assume you have.
Ask what an export looks like after five years: formats, tooling, and the model itself. Open standard components keep the answer short. Proprietary storage turns it into a negotiation.
Not which region, but which jurisdiction controls the operator. If sovereignty matters in your sector, ask about corporate structure, not about datacenters.
Ask who picks up the case and how far from the code they sit. Honest answers about support capacity beat impressive answers about support process.
A credible plan starts with one process contextualized end to end, producing an answer somebody is waiting for. A plan that starts with a company-wide rollout is a plan for year two of an implementation.
Frequently asked
A data platform gathers data that lives in separate systems into one place where it can be combined, governed and queried. An operational data platform does this for industrial operations: it connects sensor readings, events and business records to a model of the physical assets they describe, so both people and software can ask questions across all of it.
A data warehouse stores tables for reporting and assumes the questions are known in advance. An operational data platform also stores the model of the operation, the assets, the processes and the relationships between them, and keeps time-series and events connected to that model, so new questions can be asked without redesigning the schema.
The enterprise incumbents typically price in the millions per year, with implementations that take years. Our platform is open source. Self-hosting is free, and you pay for hosting or support if you want them. The honest total cost is dominated by your own integration and modeling work, whichever vendor you choose.
Yes. Self-host in your own datacenter or cloud, run it fully air-gapped with no outbound connectivity, or let us host it on infrastructure we own in Stavanger. The code is open under AGPL-3.0, so what you run can be read line by line.
Connecting operational technology, the sensors, control systems and historians that run the plant, with business systems such as ERP, maintenance and planning, so questions that touch both can be answered in one place. Done safely, the integration reads from the control side and never writes back into it.
The platform is open source, so you can stand it up in your own environment and load real data without talking to anyone first. A serious pilot typically contextualizes one process in weeks, not months. Company-wide rollouts take as long as you give them, which is why we advise against starting there.
With our managed hosting it does: the infrastructure stands in Stavanger, owned and operated by a Norwegian company under Norwegian and EEA law. If you self-host, the data lives wherever you choose to put it, which some operators prefer for air-gapped environments.
Bring one question your team answers by hand today. We will tell you honestly whether an operational data platform is the right tool for answering it, and whether ours is.