Who does what
A DataHub rollout needs five roles. Only one of them is technical.
The role that matters most, and is most often assigned wrongly, is the domain expert who says what the things in your operation actually are. That cannot be outsourced to a supplier or a data team, because it is the part only your own people know.
The five roles
| Role | Accountable for | Typically |
|---|---|---|
| Sponsor | Choosing the first question; unblocking access to source systems | Operations director, plant manager, COO |
| Domain expert | What the things are and how they relate, the model itself | Process engineer, maintenance lead, senior operator |
| Data steward | Naming conventions, data set boundaries, model quality over time | Engineering or asset-information lead |
| Administrator | Users, permissions, retention, cost, uptime | IT / OT operations |
| Integrator | Getting data out of source systems and into DataHub | Automation engineer, systems integrator, or a developer |
Small organisations routinely combine these, one person can be steward and administrator, and the sponsor is often also the domain expert. What does not work is having none of them, particularly the domain expert.
The mistake worth avoiding
The most common failure pattern in this category is this: the platform is bought by IT, the model is built by a supplier or an internal data team, and the engineers who actually know the plant are consulted at the review meeting.
The result is a model that is technically correct and operationally useless. It uses the categories a data modeller would choose rather than the ones the plant uses, so the people who were supposed to benefit look at it and do not recognise their own operation.
The model must be built by people who know the operation, using words the operation already uses. The platform's job, and DataHub's specific design goal, is to make that possible without those people needing to write code.
What each role does, week by week
Weeks 1–2 · Frame it
- Sponsor picks one question the organisation genuinely wants answered, and names the people who will answer it. Choosing a first question →
- Administrator stands up the platform (or gets it stood up) and creates accounts. Installation → · Identity providers →
- Integrator confirms which source systems can be read, and how.
Weeks 3–6 · Model and land data
- Domain expert runs the modelling sessions: what exists, what it is a kind of, how it connects. Building your model →
- Data steward sets the naming conventions before anything is created at volume, and decides the data set boundaries. Naming and standards → · What governance asks of a steward →
- Integrator lands the first two or three sources as time series and events.
- Administrator grants read and write on the right data sets. Permissions →
Weeks 7–12 · Answer the question, then widen
- Domain expert and analysts answer the original question against the model, and document how.
- Sponsor reviews the answer, and decides the second question.
- Data steward captures what the first pass got wrong about the model and fixes it, this always happens and is a sign of a healthy project, not a failed one. Grown, not built →
- Administrator sets retention and reviews cost. Data lifecycle →
How much time this actually takes
Realistic expectations, for a first project scoped to one site or one process area:
The integration work is the variable that dominates. A source system with an open API is a day; one that only exports CSV on request is a project of its own. Establish this early, it is the single biggest driver of how long a first project takes. Increasingly, some of this work is done by agents rather than people. Agents that build and run integrations →
When you need outside help
You do not need a supplier to run DataHub, and you should not need one to build your model. Outside help is genuinely useful for two things:
- Connecting difficult source systems, particularly legacy control and historian systems with proprietary interfaces.
- Facilitating the first modelling sessions, if your organisation has never done structured modelling before. The facilitator should be teaching your experts to do it, not doing it for them.
If a proposal has a supplier building your model and handing it over, that is the pattern described above, and it tends to end the same way.
- Where to start: choosing a first project that finishes
- Building your model: how to run the modelling sessions
- The business case: what to promise, and what not to
- Administration overview: what the administrator role involves