HomeTech Talk

Two Dashboards, One Number, Zero Trust

Sales says $12.4M, Finance says $9.8M. Why two dashboards disagree, and four plain-language fixes that make company numbers trustworthy again.

· 7 min read

  • data governance
  • single source of truth
  • certified data
  • Data Vault 2.0
  • 3NF
On this page 5 sections

TL;DR

  • When Sales and Finance both report "Bookings," both numbers are usually right. They answer different questions under one name.
  • Every new data tool made data easier to reach. None of them made people agree on what it means.
  • The root cause is no shared rulebook for how company data is built, or a rulebook nobody maintains.
  • The fix: one shared foundation, one owner per word, one trust badge, and a published bridge that explains every gap.
  • For data teams: model the foundation as Data Vault 2.0 underneath with a 3NF-shaped face on top, so it absorbs change and stays readable.

Picture an executive meeting. Sales puts up its dashboard: bookings this quarter are $12.4M. Ten minutes later, Finance puts up theirs: $9.8M. Same quarter, same word, two answers.

Nobody talks strategy after that. They argue about whose number is right. The meeting ends with an action item to “reconcile the numbers,” and everyone walks out trusting data a little less.

WHEN SALES AND FINANCEBOTH BRING “BOOKINGS”SALES$12.4Msigned, full valueFINANCE$9.8Mapproved, netVS??…which one goesto the board?
The numbers are made up. The meeting is very real.

I was talking with a few people in the data industry today, and every one of us had a version of this story. Different companies, different tools, same ending.

My take: this isn’t a tool problem or a people problem. It’s what happens when a company has no shared standard for how its data gets built, or has one on paper that nobody keeps up.

This Problem Is Older Than Your Data Stack

Every generation of data tools promised to fix this. Each one made data easier to reach. None of them made people agree on what it means.

  • The spreadsheet years. Every department had its own Excel file, and the big question was “which version is final?”
  • The dashboard boom. BI tools let anyone build a dashboard. Companies ended up with hundreds of them and three definitions of revenue.
  • The data lake era. All the data finally sat in one place. Sitting in one place is not the same as agreeing on what it means.
  • Today. Self-service, “data products,” and AI assistants answering questions from the same data. An AI that finds two conflicting numbers won’t stop the meeting. It will confidently pick one.
New tools, same fightSpreadsheetsWhose file isthe final one?DashboardsHundreds of charts,three revenuesData lakesSame place,different meaningsAI assistantsPicks one numberand sounds sureEvery era solved access. None solved agreement.New tools, same fightSpreadsheetsWhose file is the final one?DashboardsHundreds of charts, three revenuesData lakesSame place, different meaningsAI assistantsPicks one number and sounds sureEvery era solved access.None solved agreement.
Four eras of data tools, one recurring argument.

The tools got faster. So did the disagreements.

Why Do Two Dashboards Show Different Numbers?

Strip away the tools, and every number on a dashboard is made of three things: ingredients (the raw records), a recipe (the rules for counting), and a chef (the person who answers for it). Two dashboards disagree when any one of those is different.

  1. Different ingredients. Sales counts from the CRM, like Salesforce. Finance counts from the ERP, like Oracle. The same customer lives in both systems under two different IDs. If nobody matches “Account 123” to “Customer 4567,” you can’t even line the two numbers up. (I went deep on this in One Customer, Five Systems.)
  2. Different recipes. Sales counts a deal the day the contract is signed, often at its full multi-year value. Finance counts it after approval, minus discounts, maybe one year at a time. Both are right. They just shouldn’t share a name. (One deal can have three right answers.)
  3. No chef. Nobody owns the word “Bookings.” Whoever builds the first dashboard defines it, and whoever builds the second defines it again.
Every disagreement traces back to a missing rulebookTwo numbers for one wordDifferent ingredientsCRM and ERP use differentcustomer IDsDifferent recipesSigned vs. approved,gross vs. netNo chefNobody ownsthe word “Bookings”Root cause: no shared rulebook, or one nobody keeps upEvery disagreement traces backto a missing rulebookTwo numbers for one wordDifferent ingredientsCRM and ERP use different customer IDsDifferent recipesSigned vs. approved, gross vs. netNo chefNobody owns the word “Bookings”Root cause: no shared rulebook,or one nobody keeps up
Three causes, one root.

Underneath all three is one missing piece: a shared rulebook. That’s all an enterprise data architecture standard really is: the company’s agreement on where data comes from, how it gets combined, what each word means, and who owns it.

Without one, every team builds its own kitchen. With one that nobody maintains, it’s worse: people assume a standard exists, so they stop questioning the numbers.

How to Fix It: One Foundation, One Owner, One Badge

The fix is not a new tool. It’s four agreements, and someone who keeps them.

1. Build One Shared Foundation

Before any team builds reports, data from every system lands in one shared layer. That’s where the Salesforce customer and the ERP customer get matched, once, for everyone. Teams still build whatever they like on top: one team likes classic star schemas, another builds big flat tables for Tableau. That’s fine. They just start from the same ingredients.

For data folks. This is the conformed core, or silver layer. Whether to model it in 3NF or Data Vault 2.0 is its own debate, and I compare the two further down.

2. Give Every Important Word One Owner

The simple rule: whoever answers for the number outside the company owns the plain word. If “Bookings” goes to the board, Finance owns “Bookings.” Sales keeps its version and calls it “Sales Bookings.” Two numbers, two names, no fight.

3. Hand Out One Trust Badge, and Only One

Think of the blue check on social media. Certified means checked, owned, tested and safe for an exec meeting. Teams can mark their own trusted work as “Endorsed,” but “Certified” means one thing across the company. Badges expire when tests fail, and only certified numbers go in the exec deck. (Here’s how I automate most of the certification checks.)

WHAT EVERY TEAM ASKS FORBookingsBookings v2Bookings_finalReal BookingsWHAT THE CEO SEESSo which one iscertified-certified?A BADGE EVERYONE HAS MEANS NOTHING
Two kinds of “Certified” add up to zero kinds.

4. Explain the Gap Instead of Hiding It

Publish a simple bridge from Sales Bookings to Finance Bookings: multi-year deals, discounts, deals waiting on approval, cancellations. When executives see both numbers and exactly why they differ, the argument ends.

Different teams, same ingredients, one badgeSalesforce (CRM)Sales’ view of dealsOracle (ERP)Finance’s view of dealsShared foundationEach customer matched once, for everyoneTeam AStar schemas (Kimball)Team BFlat tables for TableauCertified badgeOne owner, one definition, tests, a bridgeExec meetingTwo named numbers, one explanationSourcesSilver layerGold layerTrust gateDecisionsDifferent teams, same ingredients,one badgeSOURCESSalesforce (CRM)Sales’ dealsOracle (ERP)Finance’s dealsSILVER LAYERShared foundationEach customer matched once, for everyoneGOLD LAYERTeam AStar schemasTeam BFlat tablesTRUST GATECertified badgeOne owner, one definition, tests, a bridgeDECISIONSExec meetingTwo named numbers, one explanation
The fix, end to end.

Trust doesn’t come from every number matching. It comes from every difference having an explanation.

Under the Hood: Data Vault 2.0 vs. 3NF

Skip this section if you’re not on the data team.

The shared foundation needs a modeling style, and the choice usually comes down to two: third normal form (3NF), or Data Vault 2.0, the best-known member of the ensemble modeling family. Both work. They fail in different ways, and which failure you can live with depends on who reads the layer.

3NF: Easy to Read, Hard to Change

3NF stores each business entity once: customer, contract and invoice each get their own table, joined by keys. It’s the Inmon tradition, and its biggest strength is that people can read it. An analyst who opens customer and contract knows what they’re looking at.

Its weakness is change. Relationships are baked into the tables: a contract has one customer because it carries a customer_id column. When a contract starts spanning several legal entities, or a third system brings its own customer ID, you migrate tables and break whatever sits on top. History is opt-in too. Unless every table gets slowly-changing-dimension handling, an update overwrites the past.

Data Vault 2.0: Easy to Change, Hard to Read

Data Vault splits every entity into three kinds of tables:

  • Hubs hold business keys only: one row per customer, keyed on a hash of the identifier the business actually uses.
  • Links hold relationships between hubs, and they’re always many-to-many.
  • Satellites hold descriptive attributes and their history, split by source system or by how often they change.

Every table is insert-only, and every row carries a load timestamp and its source. A new source system becomes new satellites. A relationship that turns many-to-many is already a link. Nothing downstream has to change, and you can rebuild the data as of any date for an auditor.

Same customer, two ways to store it3NFDATA VAULT 2.0customercustomer_id PKnamesegmentupdated_atcontractcontract_id PKcustomer_id FKamount_usdsigned_onEasy to read. One row per customer;history is overwritten unless you add it.sat_customer_crmsat_customer_erphub_customerlink_customer_contracthub_contractInsert-only. New sources add satellites;nothing downstream has to change.Same customer, two ways to store it3NFcustomercustomer_id PKnamesegmentupdated_atcontractcontract_id PKcustomer_id FKamount_usdsigned_onEasy to read. One row per customer;history is overwritten unless you add it.DATA VAULT 2.0sat_customer_crmsat_customer_erphub_customerlink_customer_contracthub_contractInsert-only. New sources add satellites;nothing downstream has to change.
One customer and one contract, stored both ways. Hubs carry keys, links carry relationships, satellites carry everything else.

The weakness is the mirror image of 3NF’s. One customer lives across a hub, several satellites and a handful of links. Point self-service analysts at the raw vault and they’ll write ten-way joins with slightly different date logic, which brings back the exact two-numbers problem this post is about.

Question3NFData Vault 2.0
Readable by analystsHighLow, unless you publish views on top
Adding a source systemRemodel existing tablesAdd satellites; existing tables untouched
One-to-many becomes many-to-manyTable migrationAlready many-to-many through links
History and auditOpt-in, table by tableBuilt in; every version is kept
LoadingUpdates and mergesInsert-only; easy to parallelize and generate
Number of tablesFewerMany more
Best fitOne main source, stable schema, small teamMany sources, frequent change, regulated industries

What I’d Pick

For a foundation fed by a CRM and an ERP, with one team building star schemas and another building flat tables for Tableau, I’d use Data Vault underneath and a 3NF-shaped face on top:

  1. Raw vault. Hubs, links and satellites, loaded straight from each source. The platform team owns it, and nobody else queries it.
  2. Business vault. Shared rules stored as computed satellites. This is where “Sales Bookings” and “Finance Bookings” get defined once, along with the bridge between them.
  3. Published entities. Point-in-time and bridge tables exposed as clean current and history views: customer, contract, booking. To the domain teams, it reads like a tidy 3NF model.

The Kimball team builds its slowly changing dimensions from the history views. The Tableau team flattens the current views. Both start from the same keys and the same rules.

Two things decide whether this works:

  • Business keys are the whole game. A hub built on a source-specific ID, like a Salesforce account ID, just copies the CRM-versus-ERP mismatch into the vault. Agree on the business key and give the customer mapping an owner before the first hub ships.
  • Generate the vault; don’t hand-write it. The loading patterns are repetitive by design. Drive them from metadata with a package like datavault4dbt or AutomateDV, and spend people’s time on keys and business rules instead.

3NF still wins in one setup: one dominant system of record, a stable schema, a small team, and analysts who need to read the foundation directly. There, being readable is worth more than absorbing change.

Where to Start on Monday

Don’t start with a 40-page governance framework. Nobody reads it. Start with the one number that causes the most arguments. It’s usually Bookings or Revenue.

  1. Get the two teams in a room for an hour. Agree on names and owners.
  2. Build the bridge between the two numbers.
  3. Certify the official one and put it in the next exec deck.
  4. Name a keeper for the standard, someone whose actual job includes keeping it current.

That last step is the one most companies skip. A standard nobody maintains is how you got two Bookings numbers in the first place.

Your company doesn’t need fewer dashboards. It needs fewer arguments about them.