Invoice automation for everything your AP tool hands back to a person
dotfun builds custom invoice automation aimed at that pile, around the payables and accounting tools you already run. We build what your software leaves behind.
.png)
Your software reports on what it processed. Your team lives with what it returned.
The pile nobody put in the business case
Ask a controller where the month goes and you will not hear "data entry." You will hear about the supplier who redesigned their invoice in March. The freight charge that broke a three-way match. The approval sitting in a director's inbox since Tuesday, on a bill whose early-payment terms expire Friday. None of those invoices failed. They were handed back.
That difference changes what you buy. Your software processed what it recognized and set the remainder in front of a human. That human is now the process. The pile is quiet. It triggers no alert and reaches no budget conversation. It is also the only part of payables that does not shrink as you grow, because more vendors means more formats and more people to chase for a signature.
So count what a person touches after your software is finished. That number, not your invoice volume, is the thing worth automating.
Invoice automation at dotfun is one process inside our workflow automation practice, part of the AI Solutions wing that also builds AI agents and custom software. IBM calls the pairing of AI judgment with process automation intelligent automation. On a payables process, that judgment is the whole job.
Capture and extraction from whatever the vendor sends
PDFs, scans, an image pasted into an email, and the spreadsheet one supplier refuses to abandon. We build extraction that reads a document the way your AP clerk reads it, pulls the fields your books need, and flags a low-confidence value instead of guessing it into your ledger.
Matching, and a real path for the ones that do not match
Two-way and three-way matching is the easy half. The half that costs you is the mismatch: quantity off by one, freight added after the order. We build logic that works out which kind it is, settles anything inside your tolerances, and puts the ambiguous ones in front of the person who can answer, reason already attached.
Approval chases that do not wait on anyone remembering
Most late invoices are not disputed. They are unanswered. We build follow-up that reaches the approver where they already work, escalates on the schedule you define, and gives them something they can approve in one tap. No portal login to postpone until Friday. Our design team shapes that moment, because an approval that looks like work gets treated like work.
The reconciliation your team rebuilds by hand every month
Your payables tool holds one version of the truth and your ledger holds another. Somebody stitches them together in a spreadsheet before close. We automate the keying and the checking, and surface only the differences that need a judgment call. Close stops depending on who has the file open.
Reporting your finance team will actually trust
An automation nobody trusts gets checked by hand, which means you paid for it twice. We build reporting that shows what was processed, what came back, and why. Your exception rate becomes a number you manage instead of a feeling you get on the twenty-eighth.
How the build actually goes
Nothing for your team to administer. No spec handed to a vendor you never meet. Three phases, and the first one is arithmetic.
We take a real sample of your last ninety days and sort it: what your software finished alone, what came back to a human, and why. Sorted by cause, the returns often cluster. Where they do, a few patterns account for most of the hours. One supplier. One approval bottleneck. One mismatch type. That ranked list decides what gets built, and just as often what is not worth building.
.png)
Anything works on a tidy PDF. We develop against the real ones: the vendor with no purchase order, the credit memo, the statement that arrives as three attachments. How much authority the build gets is your call. A clean match under the dollar figure you choose can post by itself. Anything touching a vendor's bank details waits for a signature regardless of amount. You move both numbers once you have watched it run.
.png)
Suppliers change formats. Systems get upgraded. Approvers leave. We document the build, hand over the system along with everything it has recorded, and stay on to adjust it as your process shifts. The pile does not get to quietly refill a year from now.
Where AP software fits, and where it stops
We do not sell accounts payable software, which is exactly why we can be blunt about it.
If you are still approving bills over email and paying from a spreadsheet, do not hire us yet. Buy an AP automation product first. Category tools such as Bill.com, Tipalti, and Ramp exist to handle capture, routing, and payment for standard invoices, and starting there is usually the right move. We can say that plainly, because that category is not ours.
Here is the boundary. Accounts payable automation is built for the standard case, because a product must behave the same way for everyone who signs up. Your exceptions are not standard. They come from your vendor mix, your approval politics, your chart of accounts, and the systems your books close against. No product roadmap covers that. It is the part we build, and it sits on top of the AP tool you already pay for.
Why a build, and why this team
Senior operators, not a handoff.
The people who sit with your AP clerk are the people who write the code. That math only works with a few clients at a time, so that is how dotfun runs. Nobody inherits your build from a junior account manager in month two.
No product allegiance.
We do not resell payables platforms and we take no vendor margin. The recommendation is allowed to be "your current tool already does this, switch it on."
Built around your stack, not against it.
The people doing the work spent fifteen years inside B2B tech, leading tiger teams and software builds. It shows up as a bias for the smallest system that ends the manual work.
Sizing the pile is the first job, and you can start it without us.
Take the free Iris by dotfun AI Maturity Assessment: 15 minutes, conversational, with a written action plan in your inbox within the hour and an optional 30-minute walkthrough. No credit card.
TAKE THE ASSESSMENT
The five things finance leaders raise first
Finance buyers ask harder questions than most, and they should. Money moves at the end of this process.
The build runs on your systems under your permissions. You decide which roles see vendor banking detail and which see nothing. We do not train models on your data, and it does not become ours because it passed through something we built.
Extraction, coding, approvals, overrides, and exceptions are logged with a timestamp and an actor, so any invoice can be reconstructed months later. The IRS expects you to keep the records that support what you filed. An automation should make that easier, not murkier.
Nothing enters your ledger outside the rules you set. Low-confidence extractions and out-of-tolerance matches stop and wait for a person. You choose the confidence threshold, and you can tighten it after the first month.
We build to whatever your systems expose, and we confirm the path during scoping rather than printing an integration badge on a web page. Where no clean connection exists, we build the bridge and maintain it.
Formats drift constantly, and a build that cannot absorb that has a short life. New layouts are handled as part of keeping the system current, so one redesign does not put the pile back on a desk.
Who builds it.
It is the discipline we already apply on the revenue side, where a lead that waits is a lead that cools and follow-up cannot depend on someone remembering. Payables runs on the same physics. The clock is just called terms.
Bring last quarter's exception list and we will tell you what is worth automating.
Straight answers on invoice automation
Invoice automation, also called invoice processing automation, is software that carries an invoice from arrival to payment with as little human handling as possible. That means capturing the document, extracting the fields, matching it against the purchase order and receipt, routing it for approval, and posting it to your books. Products do this well for standard invoices. Custom invoice automation handles the ones a product hands back to a person.
Start with the exception test. Count how many invoices a month a human touches after your software is finished with them. If that number is small, buy an accounts payable automation product and stop there. If it is large and growing with your vendor list, your cost lives in the exceptions, and no product covers those. Yours come from your vendors and your books. That is a build.
No, and keeping it is usually the point. Replacement is right in one situation: you have no AP tool at all, in which case buy one before you commission custom work. If you already run one and the pile is still there, the tool is not what is broken.
PDFs, scans, emailed images, portal downloads, and structured feeds. Extraction quality depends on your actual documents, so we test against your real mix during scoping rather than quoting an accuracy number we have not earned on your data. Uncertain fields get flagged for a person.
This is the question the whole build answers. First it classifies the mismatch: quantity, price, freight, tax, partial receipt, or duplicate. Anything inside the tolerances you set is settled and logged without a human. Anything outside them goes to the person who can decide. The invoice, the order, the receipt, and the difference arrive assembled, so nobody goes hunting through four systems.
Through whatever your systems expose: an API, a supported connector, a scheduled file exchange, or a controlled export and import when that is all there is. We do not publish an integration list, because the honest answer depends on your edition, your version, and your permissions. We confirm the path during scoping, before you commit.
The data is yours, and we do not train models on it. Access follows the roles you define, so vendor banking detail can be visible to two people and invisible to everyone else. Where the data lives and how it is protected depends on your environment, and we build to your requirements. We will not claim a certification we do not hold.
Invoices are one process. AI workflow automation is the parent practice, the same build discipline applied wherever manual work collects. If you need to know which process to take on first, AI consulting sets the roadmap and the fixed-scope AI readiness assessment answers it. Iris, our productized platform, is front-office work and does not touch payables.
Who this is
built for
Automated invoice processing is worth building when the returned invoices, not the raw volume, are what hurt.
Controllers and VPs of Finance at growth-stage companies
You are measured on close speed and headcount. Both get spent on invoices your software already looked at once. We take the recurring exception patterns off your team, so the close stops running on overtime.
Ops and RevOps leads who hit the ceiling of their AP tool
You have switched on every native rule your platform offers, and the pile is still there. What remains is specific to your business, which is another way of saying it has to be built.
Finance teams closing across more systems than people
Your numbers live in a payables tool, a ledger, and a spreadsheet somebody rebuilds monthly. We automate the reconciliation between them, so accuracy stops depending on one person's attention.
