Automating an Excel-and-email process without replacing it

Every few weeks someone tells us, a little apologetically, that their business "still runs on Excel". It is usually said as if it were a problem to be fixed before AI could be considered.

It is not. A spreadsheet that the whole team understands, that has survived five years of real use, is a strength. The problem is rarely the sheet. It is the re-typing around it: the email that has to be read and copied into a row, the invoice that has to be keyed into the accounting tool, the follow-up that depends on someone remembering to look.

This article describes how we scope an automation around a spreadsheet-driven process so that the sheet stays and the re-typing goes.

Start by watching one process, not by listing all of them

Ask the person who does the work to walk you through one real item from arrival to completion. Not the process as it is supposed to work; the process as it worked last Tuesday.

You are looking for four things:

Where the item arrives. Email, WhatsApp, a web form, a scanned PDF, a phone call written on a sticky note.
What gets extracted. Which fields end up in the sheet, and which of them are copied exactly versus interpreted.
What decisions are made. Which rows go to the accounting tool, who approves a discount, when a supplier is chased.
Where it gets stuck. The items that wait, the rows that are wrong, the follow-ups that never happen.

One process, followed honestly, gives you a better scope than a workshop about all of them.

Decide what a person must still see

The first instinct is to automate everything. The better question is: which entries would you be uncomfortable making without a person looking at them?

Agree which actions require a person before the pilot starts. If automatic writes are later included in scope, define the validation rules, logging and exception handling explicitly.

Three controls are involved, and they are not the same thing:

Manual review. During a first pilot, writing to a business record is a consequential action, so the entry waits for a person. The model's confidence is not, on its own, permission to write.
Automatic validation. Rules you set, such as totals that must add up and supplier codes that must exist, run on every item. Failing a rule sends the item to the review tab with the reason shown. Passing a rule is not the same as being approved.
Agreed automatic writes. After the evaluation, you may decide that typed PDF invoices from known suppliers that pass every rule can be written without a person. That decision is written into the scope, every such write is logged, and exceptions still go to a person.

In a supplier-invoice process, that usually means scans, credit notes and anything whose total does not add up wait for a person from day one, and the question of whether anything flows through automatically is answered by the evaluation, not assumed at the start. The review queue is not a failure of the automation. It is the automation working as designed.

Keep the tools that already work

The tooling is chosen per workflow. A workflow platform such as n8n suits a sequence of well-defined steps between known systems: it connects to your email, your spreadsheets, your accounting tool, your CRM and WhatsApp, and the model does the reading while the workflow moves the result to where it belongs. When the logic, volume or security needs outgrow a workflow platform, we write a small custom service instead. When a vendor tool already does the job well, we say so.

Each connection depends on what the tool and your account allow. WhatsApp needs a Business account with API access; some accounting tools permit reading but not writing without a paid plan. We confirm this during the assessment rather than assuming it. Replacing the spreadsheet is considered only if the scope and constraints genuinely justify it, which is rarer than people expect.

Measure before and after

Before building anything, take three numbers from the current process:

Handling time per item, measured on a normal day.
Correction rate: how often an entry has to be fixed later.
Unresolved items: how many are waiting at the end of the week.

After about four weeks of the pilot, take the same three numbers. That is the evaluation. It is more persuasive than any demo, and it tells you whether to roll out, adjust or stop.

What this looks like as an engagement

A free fit call to hear the process. An assessment of 1 to 2 weeks to map it, gather real examples through an agreed secure route and agree the evaluation and the approval rules. A pilot of about 4 weeks, evaluation included. Then a written result and a decision, with rollout scoped separately.

If your business runs on Excel, email and WhatsApp, that is not the obstacle. It is the starting point. The service behind this article is workflow automation; the situation it describes has its own page for spreadsheet-driven businesses.