The invoice problem costing finance teams 4 years of work - and how Agent Fetch fixes it

More than one in five online card payments (21.4%) still have no invoice attached, even six weeks after month-end. For in-store payments, that figure is 6.5%. Pay online, and you’re three times more likely to be missing paperwork.

This isn’t spread evenly. Five categories drive roughly 75% of all invoice-fetch demand: AI tools, travel, advertising, productivity software, and dev tools.

The reason is structural: each of these purchases sits behind a supplier login that nobody has any reason to check. Finance doesn’t own the login, and employees don’t think to log back in once the transaction is done. Instead, the invoice waits there until someone goes looking for it during close.

What four years of manual work costs

Manually finding, downloading, and uploading a single invoice takes around three minutes. Payhawk customers are on track to generate more than 500,000 online invoices in 2026. So, at three a minute, that’s roughly four years of collective manual work sitting inside the chasing part of finance’s month.

Finance teams currently wait an average of 12.6 days for an online invoice to appear after the transaction happens. At around £15 per invoice to process manually, that’s not a rounding error. It’s a recurring cost that scales with the business, and it compounds every month the gap stays open.

What Agent Fetch does

Agent Fetch is the newest capability inside Payhawk’s Financial Controller AI Agent. It’s a Cloudflare-verified AI agent that authenticates with a supplier’s website once, on an employee’s behalf, through an encrypted session. From that point on:

• It retrieves every receipt and invoice tied to that vendor automatically.
• Matches each one to the right card expense by amount and date.
• And codes it with the correct GL code, tax rate, and category.

One login, no portal-hopping, no inbox-forwarding rules, no end-of-the-month scramble.

Agent Fetch doesn’t add another reminder system on top of the problem. Instead, it removes the login barrier that caused the problem in the first place.

A few specifics for anyone weighing this up:

• It never stores portal passwords. The employee logs in themselves, once, through an encrypted session.
• It can’t change account settings or make purchases on the supplier site.
• It runs under the same role-based access and per-customer data segregation as the rest of the Payhawk platform.
• It doesn’t train on customer data.

What this looks like in practice

Take Anne. She’s a marketing manager, and her month touches three of the five categories causing the most trouble: ad spend, AI tools, and travel.

AI tool subscriptions. Every month, her ChatGPT Enterprise and Claude subscriptions renew on her Payhawk card. But nobody ever logs into the billing portal to check on them, because there’s no reason to. Instead, the invoice just sits there behind the login until finance goes looking for it at close.

Agent Fetch connects to that supplier account once. Then every renewal invoice arrives, matched and coded, without Anne or finance doing anything.

Advertising spend. Anne’s team runs paid campaigns across Meta, Google Ads, and LinkedIn Ads. Each platform has its own billing dashboard and login, so reconciling spend means checking three separate portals a month, plus finance is chasing Anne for invoices she hasn't had time to track down.

With Agent Fetch, that chase disappears on both sides. The agent authenticates with each platform once and keeps every invoice current from then on.

Travel spend. A few times a quarter, Anne attends an industry event. And when she’s there, she takes an Uber to almost every meeting, dinner, and airport shuttle. Each ride generates its own receipt, which will sit in an app she’ll never think to open again. Do the maths across a full events calendar, and finance is having to chase down dozens of missing receipts they have no realistic way to track.

Payhawk’s own CEO, Hristo Borisov, describes living with this exact scenario: he took more than 30 Ubers in a single month while travelling in the US, and Agent Fetch downloaded, attached, and submitted every one of those expenses without him touching them.

What ‘fixed’ looks like in practice

Before Payhawk, France-based EdTech and media company Paradox spent large chunks of every month chasing receipts by hand, sharing that the process used to include two days of repetitive messages reminding people to send them in.

With the Financial Controller Agent now handling reminders and fetching invoices directly from supplier sites, Paradox VP Finance and Strategic Partnerships, Farah Rouassi, says there’s no stress:

“My financial controller agent sends all the reminders during the month... meaning, at the end of the month, I don’t have to send any messages. No stress, no patience needed!”

And it’s not just Paradox seeing results with automation; other numbers tell the same story:

• Automation brings the cost of processing an invoice from roughly £15 down to about £5. That’s a £10 saving per invoice across any high-volume AP operation.
• 68% of AP teams still key invoice data into their ERP by hand, and manual entry is implicated in close to 40% of invoice errors, from duplicate payments to wrong coding and rework. Removing the manual step removes the source of the error, not just the time spent.
• Automated retrieval creates a fully digital, auditable trail for every invoice, a stronger position in an audit or HMRC review than reconstructing the story after the fact.
• Documents land within hours rather than weeks, so finance works from a real-time view of spend instead of rebuilding it at month-end.

For any CFO weighing this up, a £10 saving per invoice sounds insignificant, but if you multiply it by half a million invoices a year, it adds up fast. A faster audit trail might also just sound like a nice-to-have until it’s the difference between a clean VAT recovery and a queried one. Together, they change what month-end actually costs.

Why this tool won’t stall at rollout

Most finance tools fail to get adopted because they ask employees to build a new habit: a new dashboard to check, a new app to open, a new process to remember. That’s a fair concern to raise about any automation rollout.

But Agent Fetch sidesteps this problem, because it’s not a separate tool sitting alongside your existing stack. It’s one agent, doing one more thing.

All your employees need to do is, when a receipt or invoice flags as missing, click ‘authorise agent’ (or turn it on directly in a chat with the agent), log into that one supplier portal, and that’s it. Every future transaction with that vendor, and any past pending expenses, get fetched, matched, and coded from that point on, as part of the chasing schedule Payhawk already runs.

There’s no new habit to build and nothing to train people on beyond clicking authorise once.

The rollout order follows the demand data. Connect the five categories responsible for 75% of fetch demand first: AI tools, travel, advertising, productivity, and dev tools, and most of the value lands without anyone needing to think about it again.

If you’ve watched an AI rollout stall before, know that the agent doesn’t ask anyone to change how they buy, book, or expense anything. It works quietly behind spend that has already happened, which means adoption is not something you need to chase either.

Where this fits in the bigger picture

Half of finance organisations are still experimenting with AI without it touching a single core workflow. Agent Fetch is built to be the opposite. It’s a feature that removes real work on day one, with no change management required.

It sits inside a broader shift Payhawk is building toward: an AI Office of the CFO, where the Financial Controller Agent chases and codes invoices, the Travel Agent books and reconciles trips in minutes, and the Procurement Agent handles purchase requests before spend happens.

More than 6,000 finance teams worldwide already run their spend through Payhawk. If you want to see what Agent Fetch can take off your team’s plate, schedule a meeting here.