PMsquareAI data readiness
For the people who have to trust the answer

Your data has the answers.
Your AI keeps guessing.

Ask your AI the same business question five times and count the number of different answers. More than one answer (or five wrong ones!) means it’s guessing.

Watch it guess, liveGet my readiness report
Try it

Five people ask the same question.

A small, inexpensive AI model answers five wordings of one question against a fictional retailer’s tables, then answers them again against the same tables with descriptions added. Nothing renamed, nothing migrated.

Modelgpt-oss-20bopen-weight, very inexpensive

Pick a question. Each has a certified answer, and each is asked five different ways, the same five on both rows. The one marked control is an easy question both rows should get right, so you can see the test isn’t rigged.

Scorecard

All 4 questions, five times per side, on gpt-oss-20b.

Synthetic data for a fictional outdoor retailer. Answers vary from run to run, and that variation is part of what you are seeing. Scoring is deterministic: each answer’s SQL is run and compared to the correct result, and no AI grades another AI.
The real problem

Your AI can see your data. It can’t understand it.

What the AI sees
T_FCT_ORD_LN_X2
AMT_CINTEGER
DSC_BPINTEGER
RTN_FLGTEXT
DT_1TEXT
DT_2TEXT
What it needs
Order lines
AMT_CExtended line amount before discount, in US CENTS (quantity x list price). Divide by 100 for dollars.
DSC_BPDiscount in BASIS POINTS (1500 = 15%). Net amount = AMT_C * (10000 - DSC_BP) / 10000.
RTN_FLGY = line was returned; exclude from sales. N = kept.
DT_1Order date (YYYY-MM-DD). Use for all sales periods.
DT_2Ship date (YYYY-MM-DD). NULL for cancelled orders. Do not use for sales periods.

These are the demo’s real definitions. Without them, the AI reads cents as dollars, counts returns as sales and picks a retired copy of the table.

The approach

Don’t rebuild. Describe.

Rebuild

  • Re-platform the warehouse
  • Migrate, rename, rewrite the reports
  • Risk breaking what already works
  • The AI waits until it’s all done

Describe

  • Keep every system, model and report, and the reports keep working
  • Add the meaning on top of what you have
  • Answers improve in weeks

Describing pays off fastest and shows where the real gaps are. Anything it won’t fix, like a table that needs restructuring or a missing index, goes into your plan.

Why PMsquare

We already know where your meaning lives.

Our consultants work alongside software we built for this job.

What our software does

Platform describe buttons cover only their own data. Ours works across all your sources.

  1. Reads your systems: the warehouse, your Cognos and Power BI models, dbt projects and roughly forty other sources.
  2. Drafts a plain-language definition for every table and column, with a confidence score.
  3. Lines up the versions of a metric that sources calculate differently, and sends those conflicts (along with draft definitions) to a named owner on your side. Your people end up reviewing a few hundred items instead of tens of thousands.
  4. Publishes approved definitions to your AI tools over MCP, exportable as JSON or YAML.
  5. Tests them against your questions, in the loop below.

What our consultants do

  • Up frontPick the domain and the test questions with your sponsor, from your most-used reports.
  • EarlyRead your Cognos and Power BI models, planning cubes and warehouse logic for rules like a filter set once in a Cognos package that every report inherits.
  • ThroughoutRun the testing loop: find why answers missed and rewrite the definitions responsible.
  • Week 3Run the sessions where your owners settle each contested definition.
  • Week 4Test which models answer reliably, including ones you can run in-house, and write the plan for what comes next.

Definitions that get tested, fixed and tested again

Every answer is checked against your own numbers, on every model and setup you care about, with its steps, tokens, time and cost recorded.

  1. BaselineQuestions come from your most-used Cognos reports. Each report’s output is the answer key.
  2. TraceEach wrong answer is traced to the definitions behind it.
  3. ReviseConsultants rewrite those definitions.
  4. Re-testSame questions, same models. The scorecard shows what changed.
  • Each question asked five times
  • Scored by running the query, never by another AI
  • Can report no gain
  • Shows whether a smaller, cheaper model is enough
The offer

One domain. Four weeks. A fixed price.

Week 1

Baseline

Read-only access. We ask your AI your questions today and score the answers.

Week 2

Describe

AI drafts definitions across your sources and flags every conflict.

Week 3

Decide

Your owners settle the contested definitions. We publish them to your AI tools.

Week 4

Prove

Answers live in the AI you already own, with a before-and-after scorecard.

You keep

  • A described domain, exportable any time as JSON or YAML
  • Correct answers in the AI you already own
  • The before-and-after scorecard
  • A plan for what comes next, including anything that needs more than describing

We need

  • Read-only access to one domain
  • One data owner, a few hours a week
  • Thirty minutes with your sponsor

Start with the domain that has questions you most need answered—often finance or sales. Data quality and permission clean-up sit outside the sprint; we flag what we find, with owners. Afterward, stop and keep everything, add a domain at its own fixed price, or have us re-run the scorecard every quarter.

Free, no obligation

Get a readiness report for your data.

Tell us how your data and AI are set up. A PMsquare consultant reviews and approves every report, so expect yours within one business day: where your AI is likely to guess wrong, which domain to start with, and the first questions to test.

01Which AI tools are people using, or piloting, against your data?

Check everything that applies, including pilots.

02Have you asked your AI the same business question five times?

Choose one. Optional. The test on this page: same question, same data, five tries. If the answers differ, the AI is guessing.

03Where are your metric calculations and business rules built today?

The systems that actually compute numbers like net revenue or on-time delivery. Check everything that applies.

04Are the meanings of your key business terms written down where people can read them?

Choose one. Plain-language definitions, such as what counts as an active customer or which date on-time delivery is measured from. Not the code that computes them.

05Do any key metrics mean different things in different systems?

Choose one. For example, net revenue calculated one way in the warehouse and another way in a Power BI model.

06Which data platforms hold the data for that logic?

Check everything that applies.

07If you described one domain first, which would it be?

Choose one. The sprint covers one domain in four weeks. Pick the one where a wrong answer hurts most.

08What limits apply to the data in that domain?

Optional. Check everything that applies.

09Do documents explain the numbers, such as SharePoint glossaries, policies or report specs?

Choose one. Optional. Documents can be brought in alongside the data.

10What's the question you've never gotten a straight answer to?

Optional, and the most useful thing you can tell us. We'll put it first on your list of questions to test.

Where should we send it?
Who would approve a sprint? Optional

The report opens with a short summary written for them, so you can forward it as is. If that’s you, put your own name.