One Set of Numbers Your Whole Team Can Trust

Stop the three-numbers meeting by fixing three things: resolve every customer to one entity across systems, write a named definition for each metric, and require every figure to cite its source rows. Then all three teams read the same traced answer instead of three extracts.

TL;DR

Why does everyone bring a different number to the same meeting?

The Monday leadership meeting starts with revenue. Sales quotes the CRM. Finance quotes the ledger. Operations quotes the spreadsheet that drives the fulfilment plan. Three numbers, and twenty minutes spent on which one to use.

Each team is quoting a system built for their job. The CRM counts contract value on the close date because sales is measured on closing. The ledger recognizes revenue as it is delivered because that is what an auditor tests. The spreadsheet holds the corrections and allocations that neither system reflects. All three are internally correct.

Why your business systems do not agree on the numbers sets out the five recurring causes: conflicting definitions, cut-off timing, duplicate customer records, currency and rounding, and manual spreadsheet edits. Every version of this meeting traces back to one of them.

Why does declaring one system official not work?

Most companies try this first. Finance is declared authoritative, and for a quarter it holds.

It slips for three reasons. The official system does not contain some of the data, so pipeline and operational detail still come from elsewhere. The definition was never written down, so the next report built on the official system applies a different date field and produces a different number. And the corrections still live in a spreadsheet, so the official figure is quietly overridden before anyone presents it.

Authority without a written definition is a preference, and preferences do not reconcile systems.

What has to be true for numbers to agree?

Three conditions, and all three are needed.

One resolved list of entities. Each customer, product, vendor and location is recognized as the same thing across every system, matched once and reused.
One named definition per metric. Booked, invoiced, collected and recognized revenue each get a name and a rule, so a request for revenue is answered with the measure it asked for.
One traced path back to the rows. Every figure names the system, table and records behind it, so a challenge is settled by opening evidence.

Notice what is absent from that list. No system has to be replaced, and no data has to be moved into a warehouse first. What has to change is where the matching and the definitions live: in a shared place rather than in a person's spreadsheet and memory.

Each condition maps to a familiar failure: why your dashboards do not match covers conflicting definitions, and where did this number come from covers figures nobody can trace.

How does a Knowledge Graph deliver that?

A Knowledge Graph stores your business as entities and the relationships between them, with each source record attached to the entity it describes. The customer in the CRM, the payer in the billing system, the account in the ledger and the requester in the support desk become one customer entity with four sources behind it. Concepts defines the graph, entity resolution and lineage precisely enough to quote in an internal document.

SIGNLD, a decision intelligence platform by Inzata Analytics, connects read-only to 800+ business systems, including spreadsheets, and builds that graph automatically on connect rather than waiting for someone to author a schema. Ask for last month's revenue and the answer states which definition it used and cites the rows it came from.

Three consequences matter. Definitions live in one place, so teams stop arguing over a word. Answers carry evidence, so disagreements resolve by inspection instead of seniority. And the matching work compounds, so once the graph knows three records are one customer, every future question inherits it.

Warehouses, semantic layers and BI tools reach a similar destination through modelling maintained by data engineers. Compare sets out how those paths differ, which matters if you are choosing between them rather than adding to them.

Inference runs on a single-tenant AWS Bedrock instance, a private LLM powered by AWS Bedrock. Connections are read-only, your data stays yours, and we never train on it.

What changes in the weekly meeting?

The first twenty minutes come back. Instead of three extracts, the meeting opens one traced answer that names its definition and its rows.

Disagreement does not disappear, and it should not. What changes is what disagreement produces. Today it produces a debate. With citations it produces a specific list: 40 duplicate accounts, 6 internal orders included in one view, one contract signed in March and recognized across twelve months. That list is actionable in a way an argument is not.

How do you get there in a month?

Four steps, one per week.

Week one: connect the three systems the meeting quotes from, plus the spreadsheet operations reports from. Connecting the first system takes about 15 minutes.

Week two: review the entity matches, especially parent and subsidiary groupings, since how to group a corporate family is a business decision.

Week three: name the metric definitions in use, decide which one the meeting means by revenue, and write it down where everyone can read it.

Week four: run the meeting from traced answers and keep the old spreadsheet open beside them. Where they differ, open the citation and fix the cause rather than the number.

Key takeaways

The three-numbers meeting is caused by three different extracts, three filter sets and three customer lists, not by anyone being careless.
Resolving each customer to one entity across systems fixes every per-customer metric at once.
Each metric needs a written name and rule, so booked, invoiced, collected and recognized revenue stop competing for one word.
Numbers become settled when any figure can be opened into the specific rows behind it during the meeting.
Declaring one system official without writing down what it counts does not survive the next new report.

FAQ

How does a small business get every team to agree on the numbers?

By fixing three things rather than choosing a winner: resolve each customer to one entity across systems, write a named rule for each metric, and require every figure to cite the rows behind it. Agreement then comes from being able to inspect a number, not from a policy that one department's report is official.

Do we need a data warehouse to make our numbers consistent?

No. A warehouse plus a modelled semantic layer achieves it and assumes you have engineers to build and maintain the model. Resolving entities across the live systems reaches consistent answers without that project, which is why companies of 10 to 500 employees with no data team usually get there faster that way.

Why do our revenue numbers differ between departments?

Because each department's system counts a different event at a different moment. A CRM books contract value at signature, a ledger recognizes revenue as it is delivered, and a payment processor records cash when it clears. Duplicate customer records and different date fields widen the gap further, especially for per-customer metrics.

Should we still keep our spreadsheets?

Keep them and connect them. Spreadsheets usually hold corrections and allocations that never made it back into a source system, which makes them a real system of record. Connecting one puts the manual figure next to the system figure so the difference is visible, rather than leaving it on a single laptop.

How long until the meeting stops arguing about numbers?

Most of the change happens in the first month: connect the systems the meeting quotes, review the entity matches, and write down which revenue definition the meeting means. The durable part is the citation habit. Once any figure can be opened into its rows during the meeting, disputes turn into a short list of specific records.

Try SIGNLD free

Connect the three systems your Monday meeting quotes from and open one traced answer instead. Try SIGNLD free or browse all articles.

How to get one set of numbers your team can trust