Our Client Was Spending 20 Hours a Week on Manual Data Entry. They Are Not Anymore.

A client came to us last year with a problem that sounded almost embarrassingly simple: they were copying numbers from supplier invoices into their inventory spreadsheet. Every week. For 20 hours.

They had two full-time operations staff. One of them was spending half her work week doing this. The other picked it up when she was sick or on vacation. It was boring, it caused errors, and it was the first thing that fell apart whenever the business got busy.

They had looked at buying a new ERP system. The cheapest option was $18,000 a year. It would have solved the data entry problem, but it would have also forced them to change how they ran basically everything else. They were not ready for that.

What Was Actually Happening

Before building anything, we spent a few days just watching how the work got done.

Suppliers sent invoices in three different formats: some by email as PDFs, some through a supplier portal as CSV exports, some as scanned paper documents sent by fax (yes, still). The operations team would open each one, find the relevant fields, and type them into a master spreadsheet. Then they would cross-check quantities against purchase orders in a second spreadsheet. Then flag discrepancies in a third document.

That sounds like three steps. It was actually closer to 15 steps once you counted all the tab-switching, file-renaming, and email-archiving involved. Each one small. Together, crushing.

The error rate was around 3%. That might sound low. On 200 line items per week, it meant about 6 mistakes. Each one took 20 to 40 minutes to investigate and correct. So errors were adding another 2 to 4 hours on top of the 20.

What We Built

We did not build an ERP. We built a small, focused agent that did exactly the work the team was doing, and nothing else.

The agent watches a dedicated email inbox. When an invoice arrives, it reads the attachment, whether that is a PDF, a CSV, or a scanned image. It extracts the relevant fields: supplier name, invoice number, line items, quantities, unit prices, totals. It matches those against the open purchase orders in their system. Then it writes the results directly into the spreadsheet in the exact format they were already using.

If everything matches, it archives the email and moves on. If something does not match, it flags the line item with a note explaining what it found and leaves it for the team to review.

We also built a simple web interface so the team could see what the agent had processed each day, reprocess anything it had missed, and add new suppliers to its recognition list without needing to call us.

Total build time: 5 weeks. Total cost: well under what a single year of that ERP subscription would have cost.

What Happened When It Went Live

The first two weeks were messy, which we expected. The agent had to learn the quirks of each supplier's format. One supplier used a non-standard date format. Another buried the invoice number in the footer. We made adjustments as these came up.

By week three, the agent was processing about 85% of invoices without any human review needed. The team was spending roughly 2 hours a week instead of 20. Most of that time was reviewing the flagged items, which was actually useful work they had not been doing before because they were too buried in the routine entries.

The error rate on agent-processed invoices dropped to under 0.5%. That was not because the agent was perfect. It was because it applied the same rules every single time, without fatigue, without misreading a number when it was tired.

What This Kind of Project Actually Requires

The technology is not the hard part.

The hard part is spending enough time understanding the actual workflow before you write a line of code. Every business has accumulated small conventions over years, and if you miss them, the automation breaks on day one and the team stops trusting it.

It also requires being honest about what the agent cannot do well. In this case, some suppliers sent invoices in formats that changed every few months. For those, we kept the manual process. Trying to automate the outliers would have added weeks of work for marginal gain. Knowing where to stop is most of the job.

You also need to build something the team can actually see and trust. A black box that processes invoices and "hopefully" does the right thing is not a tool your operations staff will rely on. The review interface we built was not technically complicated, but it mattered enormously for adoption.

The Broader Point

This client's problem was not unique. Most businesses that have been running for more than a few years have at least one process like this: repetitive, time-consuming, error-prone, and not interesting enough to fix until someone finally does the math.

The math here was straightforward. 20 hours a week at their loaded labor cost was over $60,000 a year. The automation paid for itself in the first few months.

The operations manager who used to do the data entry is now doing supplier relationship work that she is much better suited for. She said that to us directly, and it is the thing we remember most from this project.

If you have a process like this sitting in your business, we would be glad to take a look. It usually takes one conversation to figure out whether automation makes sense, and what it would realistically cost to build.

SimplyShip App builds AI agents and custom software for businesses that want things done properly. Get in touch.