Case study · June 4, 2026

How we automated our own accounts payable

Andy Baumgartner

CEO, SparkDynamic GmbH

We build SparkOffice AP — and we run our own company on it. This is a transparent look at what changed in our finance workflow, with real numbers from our own production tenant. No customer anonymization, no cherry-picking: this is the same system we sell, doing our own accounts payable since February 2025.

Where we started

Our setup before SparkOffice AP looked like most Swiss SMEs: suppliers emailed PDF invoices to various inboxes, employees paid with company cards, and our finance team collected receipts from emails by hand. The painful part was always the bank and card statements: closing a statement meant days of matching receipts to transactions — finding the document for each line, chasing the ones that were missing, and retyping supplier, amount, and date into the accounting system.

How did we set it up?

The honest answer: there was no migration project. We did not import historical documents, clean up a supplier master list, or run the old and new process in parallel. We started our own tenant in February 2025 and changed one thing at a time, in the order that hurt the most.

Document intake came first. We set up email workflows so that supplier invoices could be forwarded straight into SparkOffice AP, and asked employees to forward card receipts the same way. That single change is why 74% of our documents now arrive without anyone collecting them — the inbox-hunting stopped within the first weeks, because the forwarding habit is easy to adopt: people already had the document in their email anyway.

Suppliers built themselves. We deliberately skipped the step most software rollouts start with — importing master data. AI recognition creates and assigns suppliers from the documents themselves, which is how we ended up with 1,145 suppliers in the system without a single manual master-data entry session. The supplier list grew as fast as the documents arrived.

Statements came next. Once documents were flowing in, we started importing bank and card statements and let the system match statement lines to the expenses it had already captured. This is where the compounding effect kicked in: the more complete the document intake, the higher the automatic match rate on each new statement.

Approvals and ERP export came last. Only after capture and matching were running did we formalize the approval workflow and the export to our accounting system. In hindsight that order mattered: approvals on top of clean, matched data are a quick review; approvals on top of incomplete data would have just moved the chaos to a different screen.

What results did 16 months of AP automation deliver?

All figures below are from our own production tenant, measured on June 4, 2026, covering usage since February 2025.

What actually changed

Receipts stopped being collected — they collect themselves

Three quarters of our documents now arrive through email workflows: suppliers and employees forward PDFs, and the documents land in SparkOffice AP already captured. Nobody on the finance team goes hunting through inboxes anymore.

Statement closing went from days of matching to review-and-approve

The system matches receipts to card transactions and expenses to statement lines automatically. What used to be days of manual matching per statement is now a review task: the finance team looks at what the system has matched, handles the exceptions, and approves. Across 42 imported statement files, every one of the 9,327 lines ended up backed by an expense record with its supporting document — a 100% documentation rate.

The gap between documents and account balance became visible

Before, a missing receipt surfaced weeks later — usually during closing, at the worst possible moment. Now the system shows exactly which statement lines have no matched expense yet, so the gap between paperwork and account balance is visible continuously, not discovered at month-end. Alerts flagged 723 statement lines for attention along the way — duplicates and inconsistencies we would previously have found late or not at all.

Spam, duplicates, and one fake invoice never reached the books

Opening an email channel for documents means junk arrives too — and the raw intake numbers show it. Of the 23,959 files our tenant received in 16 months, about 1,000 turned out to be spam or duplicates and were removed; 98% of that junk came in by email, exactly where you would expect it. The content-hash check flagged 1,200 arrivals as copies of documents already in the system — the same invoice forwarded twice, or sent by both the supplier and the employee. And once, a fraudulent invoice with no business behind it arrived by email; it was blacklisted by content hash, so any re-send is now blocked automatically before anyone sees it. Before SparkOffice AP, all of this filtering was a human squinting at an inbox — and a convincing fake invoice in a busy month had real odds of getting paid.

Data entry became an exception, not a routine

AI recognition extracts supplier, dates, and amounts from each document — and gets the supplier right on 97% of our expenses, across 1,145 suppliers. The finance team corrects the occasional miss instead of typing everything.

What do we still handle manually?

Automation did not make the finance team disappear — it changed what they spend time on. Here is what still involves a human, by design or by necessity:

• The documents that don't arrive by email. About a quarter of our documents (5,530 of 21,428) are still uploaded manually — paper receipts that get photographed, portal downloads, and documents from suppliers who only offer them behind a login.
• The supplier assignments the AI gets wrong. 97% auto-assignment means 3% correction work: 325 of 11,953 expenses needed a human to fix or pick the supplier. That is minutes per week, not days per month — but it is not zero.
• Approvals. Every approval is a deliberate human decision, and we want to keep it that way. Roughly 60% of our expense records have completed the full approval cycle — the rest are in progress or did not need the full chain.
• The flagged exceptions. The system raised 723 alerts on statement lines along the way — potential duplicates and inconsistencies. Someone looks at each one. Before, these would have surfaced late or not at all; now they surface early, but they still need a decision.
• Expenses without a statement line. 24% of expense records are not matched to a bank or card statement line — some legitimately (not paid by card or bank transfer yet), some because the match needs a human call.

The pattern across all five: the routine disappeared, the judgment stayed. We consider that the correct outcome, not a limitation to apologize for.

Month-end close: before and after

The clearest way to show the difference is to walk through what closing a bank or card statement actually looked like, then and now.

Before SparkOffice AP:

• Export or print the statement and go line by line
• Search inboxes and folders for the document behind each line
• Email employees about missing receipts — then wait, then remind
• Retype supplier, amount, and date into the accounting system
• Discover duplicates and inconsistencies only when the numbers refused to reconcile

Total: days of work per statement, compressed into the worst week of the month, with the missing-receipt chase stretching it further.

With SparkOffice AP:

• Import the statement — the system matches lines to already-captured expenses (all 9,327 of our imported lines ended up backed by a document)
• Review the exception list: unmatched lines and flagged alerts, visible continuously during the month rather than discovered at close
• Approve and export to the ERP — no retyping, because the data was extracted at capture time

Total: a review-and-approve task instead of a matching marathon. The month-end work that remains is the part that should be human — judgment on exceptions — not the part that was always mechanical. And because the gap between documents and account balance is visible all month, closing stopped being the moment of unpleasant surprises.

How do we meet the Swiss 10-year archiving requirement now?

Swiss law requires business records — including invoices and receipts — to be retained for ten years (Code of Obligations, OR Art. 958f). Before SparkOffice AP, we met that requirement the way many Swiss SMEs still do: we printed everything. Invoices and receipts were printed, filed into binders, and stored in the office — including documents that had arrived digitally as PDFs and were printed purely for the archive. Shelf space grew every year, and finding a specific receipt from two years ago meant going through binders by hand.

What does that cost? Here is a transparent approximation based on our own volume. At today’s rate of roughly 2,400 documents a month (~29,000 a year), printing every document for the archive — at a conservative average of two pages per document — means nearly 60,000 printed pages a year. At CHF 0.05–0.10 per page for paper, toner, and printer wear, that is CHF 3,000–6,000 a year in printing alone. Add roughly 150 binders a year (about CHF 450) and the office space to hold them: with a 10-year retention obligation the paper never leaves, so the archive compounds to well over a million pages — several square meters of dedicated storage, which at Zurich-area office rents is another CHF 500–1,000 a year and growing. All in, paper archiving at our volume would run on the order of CHF 4,000–7,500 a year, before counting the filing time — at even 30 seconds per document, another ~240 hours a year of someone standing at a printer and a shelf.

Now the archive is the system itself. Every captured document is retained in SparkOffice AP under the 10-year, law-compliant retention policy — stored on Microsoft Azure in the EU, encrypted, and searchable by supplier, date, or amount in seconds instead of binder-hours. The printing stopped, the binders stopped growing, and an auditor asking for a document gets it on screen instead of from a shelf. For us this was one of the quietest but most permanent wins: compliance went from a physical storage problem to a property of the workflow we were already running.

The honest caveats

We are the vendor, so read this as a first-party account — that is exactly why we publish the raw numbers rather than adjectives. Automation did not remove review: roughly 60% of expense records have completed the full approval cycle, because approval is a deliberate human step we want to keep. And the 76% expense-to-statement match rate reflects reality — some expenses are simply not card or bank transactions yet, and some need a human decision.

Is our experience representative for other Swiss SMEs?

Partly — and it is worth being precise about which part. Our document mix (supplier PDF invoices, employee card receipts, bank and card statements) is typical for a Swiss SME, and nothing about our setup relied on us being the vendor: we use the same tenant, the same AI recognition, and the same matching that every customer gets. The 97% supplier auto-assignment rate, in particular, comes from the breadth of ordinary Swiss suppliers in our data — 1,145 of them — not from a curated test set.

What may differ for you: our 74% email-in rate took a few weeks of habit-building, and companies with more paper receipts or supplier portals will start lower. And our volume — roughly 2,400 documents a month today — sits in the range the Basic and Professional plans are built for; much larger organizations will care more about the Enterprise-tier rollout support than about our self-serve setup story.

Try it on your own AP

If your month-end still involves collecting receipts from inboxes and matching statements by hand, SparkOffice AP plans start at CHF 89/month.

Create account Compare plans