FOR MORE INFORMATION
847-766-0990

Your AI is only as good as your data foundation

Your AI is only as good as your data foundation

Ask three people in most companies what “gross margin” means and you will often get three answers. Net of returns or not. Standard cost or actual. Freight in or out.

None of them is wrong. Each is working from a different definition, and many monthly reviews begin with agreeing on the numbers before anyone gets to what to do about them.

AI doesn’t fix that on its own. It just gets you to the number faster.

A confident number on the wrong basis

Ask a general AI assistant for last quarter’s margin and it will pick a definition from a column name and hand you a figure, with no sign that it had to choose. It doesn’t know that your business reports margin on standard cost, that freight is excluded, or that your fiscal year starts in April. The answer arrives quickly and reads well, on whatever basis the AI guessed.

The model isn’t the problem. Revan Analytics runs on frontier AI too. What makes an answer trustworthy is what sits underneath the model: data that has been brought together and checked, and definitions that say what each number means in your business. That is the data foundation, and it matters more than the AI on top of it.

What a strong data foundation is

Some companies already have a data foundation worth building on. Others have data spread across an ERP, a CRM and a collection of spreadsheets that has never been pulled together. Either way the first job is the same: establish what you have, and get it into a state that is fit for analysis.

In the DataMaaS Platform, your data sits in a governed data layer together with three things written down:

  • Your schema: your tables, how they connect and what each field actually means.
  • Your KPIs: margin, churn and on-time delivery, defined the way you define them.
  • Your business rules: what counts, what is excluded and which basis a figure is reported on.

These are your Business Definitions. They are written for each source, each dataset and each field, so margin means what your business means by margin, for everyone who asks.

Why Revan reads your definitions first

When someone asks Revan a question, it reads your Business Definitions before anything else. That is how it establishes which basis the question needs. Then it asks the DataMaaS Secure Query Service for the data it needs, with a request, never a query.

The definitions also tell Revan where the data stops. If margin is reported on standard cost and the standards were updated late in the quarter, the true cost variance may be larger than the figures show. Revan says so in the caveat, alongside the call, the evidence and the next step. An answer that states its limits is one you can take into a leadership meeting.

Built one source at a time

Setting this up is real work, and DataMaaS does it with you. Most companies start with one source, usually the one where getting a straight number takes the longest, and add others as they go. Every source goes through the same nine stages:

  1. Prepare: fix the source in scope and capture any existing definitions.
  2. Profile: examine the data as it stands, with evidence.
  3. Decide: agree what each record and measure means.
  4. Design: agree and build the final datasets.
  5. Inspect: test each dataset the way a user would.
  6. Write the definitions: write Business Definitions for the source, each dataset and every field.
  7. Publish: check everything, then put it live.
  8. Verify: confirm that realistic questions are answered correctly.
  9. Hand over: give users the finished datasets.

Every decision is recorded, and done means tested: a source is finished when a new reader can answer realistic questions correctly from the definitions alone. After that, DataMaaS keeps the setup current as your business changes and refreshes your data on the schedule you set.

A foundation that also decides who sees what

A strong data foundation is about more than meaning. It also decides who can see what, and it keeps the AI away from your systems.

  • The AI never touches your source systems. Revan holds no credentials and has no connection to them.
  • Read-only. Revan can’t change, delete or add anything.
  • Your permissions, enforced at the database. Access is set by table, row and column. A sales rep who can’t see margin in your systems can’t get it out of Revan either, and data a person isn’t entitled to see is never retrieved.
  • Never used to train AI. Your questions and answers are contractually excluded from AI training.
  • SOC 2 Type II attested. The whole DataMaaS Platform is in scope, and your data is stored in the US.

Start the habit with your own files

You don’t have to wait for a governed data layer to start working this way. In a Revan project, add your files, then write instructions the way you would brief a new colleague:

  • “Revenue is net of returns and rebates, by invoice date, not order date.”
  • “Gross margin is revenue minus standard cost. Freight is excluded.”
  • “Our fiscal year starts in April. Q1 means April to June.”

Every conversation in the project follows them, for everyone you share it with. When an answer comes back on the wrong basis, you fix the instruction, not the question, and the next answer is right for everyone. It is the same job we do when we build Business Definitions for a company, done in your own words, on the files you already have.

Where to start

DataMaaS has worked with manufacturing and operational data since 2002, on the systems companies already run: SAP, Microsoft Dynamics 365, M1 and NetSuite for ERP; Autotask, ConnectWise and Halo for PSA; and CRM, data warehouses, data lakes, reporting databases and spreadsheets.

Governed data is scoped with each company, because every company’s systems are different. It starts with an assessment of your systems, your data and the questions you need answered. From there we propose the solution, timeline and costs, build your governed data and Business Definitions with you one source at a time, and hand over datasets that are tested and ready to use.

See how Revan the Analyst works with your data

If every number in your business meant the same thing to everyone, what would you decide faster?

Learn more about Revan Analytics

Share on LinkedIn

Related Posts