AI use case · Is your data ready for natural-language analytics?Ask your data a question. Get three different answers.
Talk-to-your-data tools promise any question answered in plain English. The demo is easy. The dependable version needs one definition of each metric, a documented data model and source systems that agree. Without them, the AI answers confidently from whichever table it finds first.
Starts with a 30-minute call. Commits you to nothing. This use case is assessed with your team, not from exports alone.
Assessed with your team, not from exports alone
Whether people can safely ask questions of your data depends on the data model, metric definitions and access rules, not just the records. We assess natural-language analytics with your team: a working session on your warehouse or reporting layer and the questions people actually ask, alongside exports from the source systems to test the answers against.
What it needs
What natural-language analytics needs from your data.
Each requirement, the systems it lives in, and what we check. The AI is the easy part; this list is the hard part. Each check feeds the Truth Score™, a score out of 100 for how far your data can be trusted.
Truth Score pillars that gate it mostConsistencyGovernanceConnectivityThe six pillars
Where it breaks
Every system looks fine. The gaps between them don’t.
Where natural-language analytics breaks: not inside one system, but where two of them disagree about the same customer, deal or invoice.
-
Warehouse ↔ Finance
Warehouse revenue that doesn’t match the ledger
What it does to the AIThe AI answers from the wrong table, with total confidence.
-
Spreadsheets ↔ Warehouse
Board metrics calculated in spreadsheets
What it does to the AIThe AI can’t reproduce the number the board already knows.
-
CRM ↔ Warehouse
Customer counts that differ by system
What it does to the AIThe same question gets a different answer depending on the table.
Gaps to test for this use case, not client findings.
Before you sign
Questions to ask the vendor selling it.
Four questions that separate a demo from a deployment. Bring the answers, or the proposal, to the Truth Score Read and we test them against your records.
- Q1
When two tables could answer a question, how does the tool choose?
If the answer is “the model decides”, the same question can return different numbers on different days.
- Q2
Can it show the query and the source behind every answer?
A number without its working is an opinion.
- Q3
Does it respect our existing access rules, row by row?
Plain English makes it easy to ask for salaries. The tool has to know who is asking.
- Q4
How do we tell it which definition of revenue is the right one?
Someone has to own the semantic layer. Ask who, and how changes are approved.
How a Truth Score Read tests it
Natural-language analytics, tested against your own records.
Our engine matches the customers, people, deals and invoices across the exports from the systems behind the decision. Every red flag is checked by hand against sample records before sign-off, and the verdict comes in writing. Here is what that means for this use case.
- 01
Collect the questions
A working session with your team on the questions people actually ask, and where today’s answers come from.
- 02
Compare the definitions
Revenue, customers and other headline metrics compared across reports, the warehouse and the source exports.
- 03
Reconcile to source
Warehouse figures tested against exports from the CRM, billing and the ledger.
- 04
Give the verdict
Which questions can be answered safely now, and which need an agreed definition first.
This use case includes a working session with your team; the memo’s scope line says exactly what was reviewed.
How the Truth Score Read worksWhat you get back
See what comes back: a real sample memo.
One fictional B2B software company, Example Co: 72,837 records in 7 exports from its CRM, finance and HR systems, run through the same engine and the same checks as a paid Truth Score Read. Each system passes its own checks. Matched across systems, only 13% of deals tie to an invoice.
Questions
Natural-language analytics, answered plainly.
Anything else, ask on the 30-minute call. It commits you to nothing.
Book the callCan you assess natural-language analytics from exports alone?
No. It depends on your data model, definitions and access rules, so we assess it with your team in a working session, alongside exports from the source systems to check the answers. The memo’s scope line says exactly what was reviewed.
Why does our AI give different answers to the same question?
Usually because the same metric is defined in several places, and the tool picks a different table each time. The fix is an agreed definition and a documented model, not a better AI.
Do we need a semantic layer first?
You need agreed definitions first; a semantic layer is one way to hold them. We test whether the definitions exist and agree, whichever tool you use.
Which warehouses does this apply to?
Any: Snowflake, BigQuery or a reporting database. We compare it with the source systems’ own exports, because the warehouse is a copy, and copies drift.
Is your data ready for natural-language analytics? Find out before you sign.
A working session with your team, exports from the systems behind it, and a signed verdict. Starts with a 30-minute call. Commits you to nothing.
Starts with a 30-minute call. Commits you to nothing.