Mohammad Al Abdullah

Portfolio

AI integration · 2026

AI integration with Microsoft Dynamics 365 Business Central

Analysis, design and build

A full integration between email, the documents on the finance team's desktops and Microsoft Dynamics 365 Business Central for an industrial manufacturer. It reads and writes data, builds budget and forecast analysis, and follows the company's own month-end workflow.

finance workbooks reverse-engineered into rules

wrong cells when rebuilding a month blind, out of 8,111 key cells

workbooks built for a month in their real dependency order

postings made without a person approving them

The brief

The finance team ran its month-end close through a chain of Excel workbooks that fed each other, with figures arriving by email and moved into Business Central by hand. The knowledge of how it all fitted together sat with one or two people.

Before automating anything, I reverse-engineered the workbooks into written rules: what each cell does, where its inputs come from, and which workbook feeds which. Only then did I build the AI skills that do the work, in the same order the team does it.

Deliverables

What I delivered.

Business Central bridge

Reads lists and ledger data out of Business Central, writes the month's accrual vouchers into an empty journal in the exact column order the system expects, and reconciles what was posted against the vouchers.

Email and document intake

Invoices, meter readings and payroll exports that arrive by email or sit in desktop folders are picked up and put into the right workbook, instead of being copied by hand.

Month-end build and checks

One command builds the month's workbooks in the order they depend on each other, then runs the controls and produces a close report that names any missing input.

Budget, forecast and management reporting

A live-formula reporting pack with budget against actual, trend, deviations and a monthly management report, tied to the trial balance line by line.

Actions

What I did.

  • Mapped every workbook, formula and external link before writing any automation, and wrote the result down as rules the finance team can read.
  • Tested the rebuild blind: the system rebuilt a month without seeing the team's answers, and the result was compared cell by cell.
  • Kept the human in charge of the books: the system prepares and checks the journal, and a person posts it.
  • Worked on pseudonymised copies of the payroll data, so no employee names left the client's own environment.

Problems and solutions

What went wrong, and how I handled it.

Every project hits problems along the way. These are the ones that mattered here.

Problem

Some workbook figures were wrong in the team's own files, so matching them exactly would have meant copying their mistakes.

Solution

The system reproduces the existing process and records each defect it finds as a separate finding, so the team can decide what to fix. Several were real: a missing day of hours, a lookup reading the wrong month, refunds that never reached the ledger.

Problem

A missing input could make a month look finished when it was not.

Solution

Every written cell must be derived, supplied or carried forward. Anything else is emptied and listed. A missing file shows up as not run, with the exact file named, and no totals are released until the checks pass.

Outcome

The month-end close now runs as one process from the incoming email to the journal in Business Central, with checks at every step and a person approving what is posted.

Skills and tools used

Microsoft Dynamics 365 Business CentralExcel automationPythonLLM workflowsEmail automationBudget and forecast analysis