Skip to content
Anchored Analytics

Closing the month and building the board package takes too long.

Accounting closes the books, then someone spends two or three more days exporting from the ERP and fixing a workbook full of formulas. The weekly sales report gets built the same way, every Monday.

I automate the pulls from your accounting system, even one running on a server in your office, and move the math into one shared model. The package and the Monday report both become reports that fill in on their own. If a pull fails, someone gets an alert.

If the package took three days a month, that's more than seven weeks of someone's year back.

How it's built in Fabric

Data Factory pipelines, the on-premises data gateway, a lakehouse, a semantic model, paginated reports, and scheduled subscriptions.

The board package goes out on day 5 instead of day 8.
123456789Working dayBy handClosing the booksRebuilding by handReviewAutomatedClosing the booksReviewPackage refreshes overnight3 working days sooner
Sample data

Sales and finance show up to the same meeting with different revenue numbers.

Sales counts revenue when the deal is signed. Finance waits for the invoice and backs out credits, and operations keeps its own count off shipments. The meeting loses twenty minutes to it.

Building the model is the easy part. Getting sales and finance to agree on the official number takes a couple of conversations. Then I build that definition once, give the others honest names like bookings, and every report reads from it.

How it's built in Fabric

A shared semantic model with documented measures, endorsed as certified so other reports reuse it.

Why sales, operations, and finance each report a different revenue number
October, from booked orders to net revenue
$0$1M$2M$3M$4M$5M$4.82MSalesBookedorders−$0.21MCancelled$4.61MOperationsShipped−$0.34MShipped, notinvoiced−$0.09MCredits andreturns$4.18MFinance · officialNetrevenueShipped, waiting on invoices
Sample data

You're paying for Power BI, and everyone still lives in Excel.

A few dashboards went up fast, didn't match finance, and people went back to Excel. Now there's a monthly Microsoft bill nobody is sure about, and a payroll export sitting where half the company can open it.

I right-size the licensing first, then fix what's underneath: a test area, access that matches people's jobs, and labels on confidential data. And I connect Excel to the official numbers instead of fighting it.

How it's built in Fabric

Capacity sizing and the Capacity Metrics app, workspace roles, deployment pipelines, sensitivity labels, certified semantic models, and Analyze in Excel.

47 of 120 Power BI licenses haven't been opened in 90 days.
One square per seat
Finance1 of 18 unusedSales23 of 40 unusedOperations16 of 35 unusedLeadership2 of 15 unusedHR5 of 12 unused
Used weekly (43)Monthly (30)Not opened in 90 days (47)Sample data

You find out a location is slipping after the quarter is over.

The quarterly review shows one branch's margin has been sliding for ten weeks. By then the quarter's gone.

I build a weekly view of every location against its targets, and each regional manager sees only their own. When a location stays under target two weeks running, its manager gets an email or a Teams message.

How it's built in Fabric

Power BI on a semantic model, row-level security by region, and Activator alerts to email and Teams.

Harbor slipped below target in week 3. The alert in week 4 gave its manager time to fix it.
Fixed before the quarterly review28%30%32%34%36%Target 32%Other 5locationsHarbor 33.3%Alert sentQuarterly review1471013Week
Sample data

By the time you see budget vs. actual, it's too late to do anything about it.

Budget vs. actual arrives three weeks after month-end, and the explanations trickle in by email over the following week.

Department heads get a page that updates daily. Next to any big variance there's a box to explain it, and the note is saved with the numbers and who wrote it. A few people may need an extra Microsoft license for this; I'll price that first.

How it's built in Fabric

Power Apps or translytical task flows writing to a SQL database in Fabric, read by the same semantic model as the reports.

Facilities' overspend showed up on Oct 9 with the reason attached, not in a report on Nov 21.
$0K$10K$20K$30K$40K$50K$60K$70KSpent $64.5KBudget $52K"Roof repair after the storm.One-time, not in this year's budget."M. Ortiz · entered Oct 9, 2:14 PMOct 1Oct 8Oct 15Oct 22Oct 29
Sample data

Too much cash is sitting in inventory and unpaid invoices.

The aging report and the stock list live in different systems, and nobody looks at them together. Purchasing reorders off last week's export.

I pull receivables, inventory, and open orders into one view that's current every morning. Collections gets a list of who's late. Purchasing sees what isn't selling before reordering, and gets an alert when stock runs low.

How it's built in Fabric

Mirroring for supported databases, a lakehouse, a semantic model, and Activator alerts on stock levels.

$612K is stuck in late invoices and stock that isn't selling.
Receivables and inventory this morning, sized by dollars
Receivables $3.04MInventory $2.60MCurrent$1.92M1–30 days late$640K31–60$280K61–90$115K90+$87KSold this month$1.64MSold in the last 90 days$550KNo sales in 90 days $410K
Sample data

Leadership wants to try Copilot, and you're not sure the data is ready.

Copilot and Fabric data agents answer everyday questions about company data. On messy data they answer anyway, and one wrong margin number in front of the CFO ends the experiment.

Most of the work happens before anyone asks a question: one definition per number, readable names, and instructions on what not to answer. Then we test against questions with known answers, and again after every model change, because it's still software and it can be wrong. Copilot needs a paid Fabric capacity, and you'll know the cost up front.

Answers you can put in front of the CFO, and a test set you rerun after every change.

How it's built in Fabric

Copilot in Power BI, Fabric data agents, and the Prep data for AI settings: descriptions, synonyms, verified answers, and AI instructions.

Copilot's answer matches the report, and it says when it can't answer.
Which locations were under margin target last month?

Only Harbor, at 30.7%. Every other location was at least a point above the 32% target.

Target 32%Westside36.0%North35.2%Online34.5%Central34.0%Valley33.2%Harbor30.7%

Based on Gross margin %, Finance definition. Same number as the Monthly Results report.

What's our margin by customer?
I can't answer that from this data. Costs are tracked by location, not by customer. Revenue by customer is in the Sales report.
Test questions with known answersLast run Nov 4, after a model change
39 of 40 match. One sent back to add "store" as another word for location.
Sample data

Your reporting runs on aging tools, and only one person knows how they work.

Years of logic live in an old SQL Server with SSIS jobs, an Access database, or a workbook full of macros. The server is near the end of support, and the one person who understands it is retiring, or already gone.

I go through it with them if I can, or work it out from the files. Then I move it to Fabric one piece at a time and run old and new side by side until every number matches or we know why it doesn't. Nothing gets switched off until it does.

Whoever takes over gets a guide and two trained people to ask.

How it's built in Fabric

SSIS packages rebuilt as Data Factory pipelines, SQL Server logic moved to a Fabric warehouse or lakehouse, SSRS reports rebuilt as paginated reports, and comparison queries between old and new.

Every difference between old and new was fixed or explained by the seventh run.
Differencesper run5432110000RevenueCost of goods soldFreightRebatesPayrollReceivablesInventory valueOpen orders12345678910Run
MatchesDiffersOld version was wrong (confirmed)Sample data

Recognize one of these?

Tell me which one, or what's going on if it's none of them.