Atlas is Ninety One's programme to make our data trustworthy. It is not a new system you need to learn — it's a way of working that makes sure the data behind our reports, dashboards, and AI tools is accurate, owned by someone, and checked every day.
For years, teams across the firm have each kept their own version of the same data — their own spreadsheets, their own checks, their own fixes. The same numbers get gathered and corrected in multiple places, often with different results. That's manageable when a human double-checks everything by hand. It stops being manageable once AI tools start using that data automatically, because AI doesn't know the data is wrong — it just produces an answer that looks right.
We call this risk "trusted wrong decisions" — an output that looks fine but is built on bad data underneath. Atlas exists to stop that from happening.
80% of our data domains reaching a trusted state by March 2027. A "domain" just means a defined area of business data with one accountable owner — more on that in the next section.
EDP stands for Enterprise Data Platform. Think of it as the one trusted home for the firm's data. Instead of every team pulling numbers from wherever and checking them their own way, the EDP is the single place that holds the checked, governed version — and it's where reports, dashboards, and AI tools should be reading from.
You do not need to know how the EDP works technically. It runs on a platform called Databricks, but unless you're an engineer, you'll never need to open it.
No. Systems like Oracle, Salesforce, and Charles River keep doing what they already do — capturing and running the business. Atlas doesn't touch that. It sits alongside them, pulls the data those systems produce, checks it, and makes a trusted version available. Nothing about your day-to-day system use changes.
Atlas is led by a small central programme team, supported by senior leadership (the Data Council) for direction and decisions, and delivered by the people who actually own the data — the domain teams. You'll meet that structure in the next section.
A domain is a business area of data with one named owner — for example, Benchmarks, or AUM & Flows. It's a business concept, not a system or a database. The same domain might live across several systems over time, but it stays the same domain because the same person and team are accountable for it.
Each domain in Atlas has a small team around it:
Data Owner — a senior person accountable for the domain. They don't need to be technical. Their job is to know what the data is used for and sign off that it's good enough.
Data Steward — the person who looks after the data day-to-day. They're the one you'd contact if something looked wrong.
Data Engineer — builds and maintains the technical pipeline that moves the data into the EDP.
Product Delivery Manager — keeps the team's work on track and escalates if something's blocking progress.
Between them: the Owner decides if it's right, the Steward checks if it's healthy, the Engineer makes sure it's running, and the PDM makes sure it's on schedule.
That's most people. If you use a report, dashboard, or AI-generated output, your role is simple: if something looks wrong, say something — to the Data Steward for that domain — rather than quietly working around it. Every domain has a named Steward whose job is to deal with exactly that.
Every domain moves through six states. Think of them as checkpoints — each one describes what's true about the domain, not how much technical work has happened behind the scenes.
| State | Name | What it means in plain English |
|---|---|---|
| 1 | Activate | An owner is confirmed and the domain is registered. Starting point. |
| 2 | Discover and Define | The team has agreed what "good data" looks like and written it down. |
| 3 | Integrate into EDP | The data now flows automatically into the trusted platform. |
| 4 | Produce and Publish | People can find the data, see what it contains, and trace where it came from. |
| 5 | Measure and Trust | The data is checked every day and proven reliable. This is the goal. |
| 6 | AI Ready | An AI tool is actually using the data live, and the old way of getting it has been switched off. |
Because State 6 isn't something you can plan your way into — it only happens once a real AI use case is live and using the data. State 5 is the thing we control: proving the data is trustworthy. Once it's trustworthy, AI use naturally follows. You can't skip ahead to State 6 by trying harder; it has to be earned through actual use.
Every domain is given a tier based on how critical it is: M1 (used directly in investment decisions or client reporting), M2 (important to operations but not investment-critical), or M3 (internal, supporting work). The tier affects how strict the quality bar is — M1 domains are held to the highest standard.
In the EDP, which runs on Databricks. If you're not an engineer or steward, you won't need to log into this directly — you'll keep using the reports, dashboards, and tools you already use, which will increasingly read from the EDP behind the scenes.
For almost everyone — no. The tools and dashboards you already use stay the same on the surface. What changes is what's powering them underneath: trusted, checked data instead of someone's manually maintained spreadsheet.
A short intake form to register your domain, and a simple dashboard that shows your domain's trust score and whether anything needs attention. Engineers work in Databricks directly; Owners and Stewards generally don't need to.
Every employee sits within at least one data domain. If you're not sure which one, ask your manager or reach out to the Atlas team — they can point you to the right person.
Reach out to the Atlas team directly, or contact Lorian Barrett if it's something specific. There's no wrong door — if you're not sure who to ask, ask anyway and you'll get pointed the right way.