Numbers your CFO trusts, then AI on top of them.
We clean and connect the data in your ERP, CRM and bank, build the Power BI reports finance actually opens, and only then add AI that answers questions from that data. Clean data first is the whole trick: Microsoft says Copilot on a poorly prepared model can return generic or inaccurate answers.
Microsoft Fabric and Power BI, in your own tenant. Your ERP stays the system of record.
AI-ready data is business data that is clean, connected and described well enough that a report or an AI agent gives the same answer a careful analyst would. For a finance team that means one customer record instead of three, ERP, CRM and bank data joined on real keys, a semantic model with clear names and measures, and permissions that follow the person asking. Without it, dashboards disagree and AI guesses.
MIT Sloan Management Review (Redman, 2017) estimates bad data costs most companies 15% to 25% of revenue, and that two-thirds of that cost can be removed.Most failed BI projects start at the top. We start with the systems you already run and work up, so every number on a dashboard can be traced back to a ledger entry.
Nothing replaced. We read from them, and the ERP stays the system of record.
Fabric mirroring copies SQL-based ERP data in near real time, with replication compute included.
Every report and every AI question uses the same definitions. No more two DSO numbers.
The Fabric data agent turns plain-English questions into read-only queries, using the asker's own permissions.
Start with the smallest that does the job. Most mid-sized finance teams need the first or second column, rarely the third.
| Power BI on your ERP | Microsoft Fabric | Separate data lake | |
|---|---|---|---|
| Fits when | One or two systems, under about 50 report users, finance reporting and KPIs | ERP plus CRM, bank and spreadsheets, a governed warehouse, many viewers | A large data team, heavy data science or streaming, several clouds |
| What you pay | Power BI Pro seats | A Fabric capacity plus Pro for report builders | Storage, compute and the engineers to run it |
| AI and Copilot | Not included. Copilot needs a paid capacity | Copilot in Power BI and Fabric data agents from F2 up | Build your own |
| Watch for | Logic copied report by report, and 8 refreshes a day on Pro | Sizing the capacity right; if overage billing is on, extra use costs 3 times the normal rate | The highest build and running effort |
From Microsoft Fabric licensing and Copilot documentation, October 2026.
US list prices, billed yearly. Below an F64 capacity, everyone who views a report needs a Pro or Premium Per User licence.
| Licence | List price | Suits | What it adds |
|---|---|---|---|
| Power BI Pro | $14 a user a month | Small teams sharing reports | Publish and share. Needed by every viewer below F64 |
| Premium Per User | $24 a user a month | Under about 250 users | Bigger models and 48 refreshes a day. No Fabric items |
| Fabric F2 | About $263 a month pay-as-you-go, about $156 reserved | A first Fabric project | OneLake, mirroring, Copilot and data agents |
| Fabric F64 | About $8,410 a month pay-as-you-go, about $5,003 reserved | Roughly 360 or more report viewers | Viewers with free licences can read Power BI reports |
Power BI prices from Microsoft's pricing page; Fabric prices worked out from the Azure Retail Prices API for East US ($0.18 per capacity-unit hour, or $938 per unit a year reserved). Microsoft says actual prices show at checkout.
Built once on the shared model, so every number agrees. Business Central's own Power BI finance app ships many of these, and we build the same set for any ERP.
By month, by entity and by dimension, drilling down to the ledger entry.
Month by month, with liquidity ratios alongside.
Variance by department, with the lines that moved most.
What's due, what's late and what discount you're about to miss.
Inventory days plus sales days minus payables days, trended.
Forecast against actual, rolled every week, variance by line.
Gross margin by customer and product, so pricing talks start with facts.
The direct method, built from the data you already have. Weeks one and two are nearly exact; weeks eleven to thirteen are estimates, and the report says so.
Week zero starts from actual balances in your bank feeds, not the ledger.
Actual cashEach invoice lands on the date that customer really pays, based on their days-late history, plus unbilled orders.
Customer habitsSupplier invoices, POs, the payroll and tax calendars, and debt service.
CommittedReplace the closed week with actuals and show the variance line by line, so the forecast earns trust.
Variance trackedThe most common reason two reports disagree. Tick off what you've done; the rest is where we start.
Dataverse allows up to five published duplicate rules per table and merges accounts, contacts and leads two at a time.
Fixed price per phase. You see real numbers from your own ERP in the first weeks.
We profile your ERP and CRM data, list the reports finance relies on and the numbers that disagree, and size the right platform.
Your controller signs off how DSO, margin and cash are calculated, once, for every report.
Duplicates merged, keys shared across systems, ERP data mirrored or loaded into one model.
The core set first, each one reconciled to the ledger before anyone sees it.
A Fabric data agent for plain-English questions, and AI agents that act on the same clean data.
Refresh monitoring, data quality scores and new reports under one SLA. See managed services.
Licence costs are separate and listed above. You get dated phases and a fixed price per phase after the assessment.
Ask to see a report reconciled to a ledger, not just a pretty dashboard. A good consultant agrees definitions with your controller before building, builds one shared model instead of logic copied into each report, and sizes licences so you don't pay for a capacity you don't need.
Sizes the right Fabric capacity, sets up OneLake and mirroring from your ERP, builds the warehouse or lakehouse and the semantic model, and switches on Copilot and data agents safely. A first project usually covers one or two source systems and the core finance reports.
Licences are the predictable part: Power BI Pro lists at $14 a user a month and Premium Per User at $24. A Fabric F2 capacity works out at about $263 a month pay-as-you-go. Build cost depends on how many systems and reports, so we price each phase after a two-week assessment.
Yes, with care. The Fabric data agent turns plain-English questions into read-only queries, using the asker's own permissions, and needs at least an F2 capacity. It works well on a clean, well-described model and badly on a messy one, so the model comes first.
Use the direct method over 13 weeks: start from actual bank balances, add receipts from open AR on the dates each customer really pays, subtract payables, payroll, tax and debt, then roll it every week and track the variance line by line.
Profile both systems, pick which system owns each field, set rules for which record survives, and merge in the ERP only after open invoices and orders are mapped. Then store the CRM ID on the ERP customer and turn on duplicate detection so they don't come back.
Start with the data you already have. We'll show you where it disagrees and what it would take to fix.
Facts on this page checked on 22 October 2026.
Founder and CEO
Chief Sales Officer
Join our live webinar on Sept 24 and see five
ways to automate meetings, approvals, and
document search with tools already in your
tenant.
