SynthesisLogic

AI agent for Excel and spreadsheets

An agent that fills your spreadsheets from emails and reports, checks them for errors, combines files from branches into one master workbook and sends the daily or weekly summary, while you keep working in Excel or Google Sheets.

Book an intro call How it works
A morning with this agent
  1. Sales sheets from 5 branches combined into oneDone
  2. 2 rows with a missing price foundAsks a person
  3. Weekly stock report built from the merged sheetDone
  4. Report sent to the manager for reviewAwaiting approval

An example. Your agent is built around your own work.

How it works

  1. Pick up the files

    Source files arrive by email, in a shared folder or from a system export. The agent checks that each file has the layout you agreed: column names, date format, units. A file that does not match is rejected before any calculation runs.

  2. Fill and combine

    It reads the data with code, not copy and paste, and maps the columns into your master workbook. Files from several branches or people end up in one clean table with the formulas left intact.

  3. Check the numbers

    Row counts, totals against the source, blank keys, duplicate periods and broken formulas are tested. Anything that does not add up is listed for a person, and the run stops there.

  4. Report and send

    The summary sheet, the chart pages or the client file are produced from the same data, with a short written comment on what moved. After the checks pass and a person releases it, the file goes to the people who need it.

New to AI agents? How they work, in plain words

Real examples, setup steps and tool connections for this agent

An example conversation

An example of how a conversation runs. The grey lines show what the agent does in your systems meanwhile.

  1. CustomerMorning. The three branch stock files should be in the shared folder. Can you build the weekly sales workbook?
  2. Found 3 files in /Weekly/Branches: north.xlsx, east.xlsx, central.csv. Checked headers against the agreed layout: east.xlsx has the date column in a different format, central.csv uses commas as decimal marks.
  3. Converted both formats, logged the change. Loaded 2,184 rows into the master workbook. Row counts match the source files. No blank product codes, no repeated weeks.
  4. Totals check: north and east match their own totals. Central differs by 412.50 from the total in its last row.
  5. AgentThe master workbook is updated with all three branches and the summary sheet is rebuilt. One thing needs a person: the central file adds up to 412.50 less than its own total row. I have not hidden the difference.
  6. CustomerWhich rows?
  7. AgentRow 311 has a negative quantity, minus 5 of product 4471. It looks like a return that was typed as a sale. I left it as it was in the source.
  8. CustomerThat is a return, keep it. Send the summary to the three branch managers, with a note about the difference.
  9. Added the note to the summary. Sent the PDF and workbook link to 3 managers. Saved the run log with the input files and the checks.
  10. AgentSent. Sales this week are 6.2% up on last week, driven by the east branch. The run log is in the folder next to the workbook.

How it is built

The workbook stays yours
We do not replace your spreadsheets. The agent reads and writes your existing Excel or Google Sheets files, so your team keeps working in the same files. Formula cells can be locked so the agent writes only into the data areas.
Code does the arithmetic, the model reads and explains
Calculations run in code (for example Python reading and writing the workbook), so the numbers are exact. The language model is used where judgement is needed: reading a file whose layout varies, suggesting which column is which, and writing the short comment on the results. It never does arithmetic on raw cells.
Column mapping is confirmed once
When a new source file has different column names, the agent proposes a mapping and a person confirms it once. The confirmed mapping is saved, and the next file with that layout needs no question.
A check for every step
Each step has a test: the rows loaded equal the rows in the source, totals match, no blank keys, no duplicate periods, formulas still present. If a test fails, the run stops and the master file is not touched.
Files come in by folder, email or export
A folder watch or a naming rule picks up new files from OneDrive, SharePoint, Google Drive or a network share. Attachments can be taken from a mailbox. For a system with no API, a scheduled export is used.
One writer, read-only copies for everyone else
To avoid two people overwriting each other, the agent is the only process that writes to the master file, and others work on copies or read-only views. Every run is logged with the files used, the check results and who approved the release.
Where it runs
In the cloud for Excel Online and Google Sheets, or on your own server or a machine in your office if the files must not leave your network. A local language model is possible for private data, but it is weaker than a hosted one, so the tasks are kept narrow and the output is checked.

What you can connect to it

Microsoft Excel (desktop files and Excel Online)
Reads and writes tables and ranges, keeps formulas, refreshes the summary sheets and builds the reporting file.
Google Sheets
Reads and writes values and formulas through the Sheets API, and can use a sheet as the place where people review doubtful rows.
SharePoint and OneDrive
Watches folders for new files, saves the finished workbook and keeps the previous version.
Google Drive and shared network folders
Picks up exports from branches or people and files the outputs.
Power BI
Refreshes the dataset after the workbook has been updated, so dashboards show the new figures.
ERP and accounting exports (SAP, Odoo, QuickBooks, Xero)
Reads the CSV or Excel exports they produce, or their APIs, and can post workbook figures back as entries when you allow it.
Outlook and Gmail
Takes spreadsheets from email attachments and sends the finished report to the right people.
Word and PDF templates
Turns the figures into a written report for managers or clients.
Book an intro call

Connects to your tools. You stay in control.

Works with what you have
Microsoft Excel, Google Sheets, SharePoint, OneDrive, Google Drive, Power BI, Outlook, Gmail, your ERP or accounting exports (SAP, Odoo, QuickBooks, Xero), Word templates.
A person approves what matters
Payments, orders, discounts or anything you choose wait for approval. Every action is logged.
Can run on your own servers
The agent, its data, its logs and even the language model (open models such as Llama, Qwen or Mistral) can run on your servers, so documents and customer data stay inside your network. How that works

What it does not do

  • It does not replace your accountant's judgement or sign off figures. It prepares and checks; a person releases.
  • It does not do arithmetic with the language model. Totals come from code and are compared with the source.
  • It does not overwrite your master file when a check fails, and it does not delete data without approval.
  • It does not read a file whose layout is not agreed and unchecked. It asks first.
  • It cannot fix numbers that are wrong at the source. It points to the row where the figures disagree.

Questions

Will it break our spreadsheets?

No, if it is set up as we describe. It works on copies until you trust it, formula cells can be locked, it logs every change, and it never deletes data without approval. A failed check stops the run and leaves the master file as it was.

Can it move us off Excel later?

Yes, if you want that. The agent can keep your sheets in step with a database or an ERP when you are ready, so the move can happen in stages.

What can an AI agent do in Excel that macros and formulas cannot?

It can read files whose layout varies, such as emails, PDFs and exports from different branches, and decide where each value goes. It also writes the explanation next to the figures. Fixed tasks that a macro or Power Query already handles well can stay as they are.

Can it analyse our data and answer questions about it?

Yes. The model writes the analysis, code does the calculation, and the model then explains the result. Each figure that reaches a report is checked against a total computed directly from the data.

Can it produce an automated weekly or monthly report in Excel?

Yes. We build a template workbook with the fixed tables and charts, the agent fills it with new data each period and refreshes the pivot tables, and writes the comment cells. A person approves it before it is sent.

Can it run on our own computer so the data stays inside the company?

Yes. The agent can run on your server or an office machine, reading and writing the files there. A model hosted outside is not required, though local models are weaker, so we keep the tasks narrow.

Do we need Excel or Google Sheets specifically?

No. It works with either, and with CSV exports from systems that have no spreadsheet at all. If your team has a mix of both, it can read from one and write to the other.

How much does an agent like this cost?

It depends on the task and the systems it connects to. After a first call we give you a clear estimate of cost and time.

Want this agent built around your work?

Tell us in a few lines what the job looks like today and we will set up an online intro call.

Type of organisation