Anthropic publishes a remarkable thing: a ranked list of the 50 work tasks people most often hand to AI, worldwide, matched against the official task catalogue of 718 occupations.
We read all fifty with one question: how many of these are, functionally, the daily job of an accounts payable department?
The answer: at least six of the top thirty-one. Including the #1 task on Earth.
And one more number before the list. Globally, tasks that a classifier estimates would take a person about 5 hours working alone are running as roughly 40-minute AI conversations (automated estimates, order-of-magnitude buckets โ but the ratio is the story). The work your invoice depends on didn't just get delegated. It got 7ร faster.
Source and full methodology: Anthropic Economic Index, May 2026.
The AP tasks hiding in the global top 50
The Task Ledger
Humanity's most-delegated work tasks โ AP-relevant entries stamped
Selected from the global top-50 work-task ranking. Shares describe conversations matched to tasks, not people or jobs. Source: Anthropic Economic Index, May 2026.
Look at what's stamped. Looking up records. Answering financial inquiries. Explaining policies. Composing business correspondence. Reviewing documents to respond to requests.
That's not "AI is coming for the back office." That's a functional description of what happens between your invoice falling due and you getting paid โ already sitting in the global top thirty-one.
The 5-hours-to-40-minutes problem
Now combine the list with the speed number.
The New Physics of the Back Office
Global figures ยท May 2026
Source: Anthropic Economic Index, May 2026. Time figures are automated order-of-magnitude estimates.
What used to protect debtors was friction. Finding the contract took time. Drafting a defensible reply took time. Checking your country's late-payment rules took an afternoon.
What used to protect creditors was also friction โ a debtor could only produce so many polished stalling letters per week.
Friction is gone, in both directions. Which means the outcome of a collection case now depends less on who can produce correspondence โ anyone can, instantly โ and entirely on who controls the calendar and the consequences.
Three moves for the frictionless era
1. Stop pricing effort into replies. A detailed, well-researched dispute response is now a 40-minute artefact, not a week of legal preparation. Judge its evidence, not its polish.
2. Assume perfect information on their side. Task #1 on Earth is searching records. Your debtor can check your contract, your jurisdiction's interest rules and your own past leniency instantly. Never bluff โ every empty "final notice" is now verifiable as empty.
3. Win on the two things the list can't do. Scroll the top 50 again: nothing on it pays an invoice, and nothing on it makes a decision under pressure. Payment and commitment remain stubbornly human. Aim every action at forcing a named human to do one of those two things by a fixed date.
What the Machine Does vs. What Collects Money
The top-50 list, reduced to its collection consequence
Frequently asked questions
What tasks do people use AI for most at work?
The global #1 is searching databases and records for information (4.95% of all matched usage), followed by reference research, product and service advice, software troubleshooting, and programming. Responding to financial inquiries ranks #7 worldwide.
How does AI affect accounts payable departments?
At least six of the world's top-31 AI work tasks map directly onto AP work: record lookup, answering financial inquiries, explaining policies, composing business correspondence, and reviewing documents to respond to requests. The correspondence side of AP is heavily AI-touched; the payment decision is not.
How much faster does AI make office tasks?
Globally, tasks a classifier estimates at roughly 5 hours of solo human work are running as roughly 40-minute AI conversations โ an order-of-magnitude estimate, but a consistent one across the dataset.
Does AP automation make it harder to collect unpaid invoices?
It changes the game rather than hardening it. Producing correspondence is now free on both sides, so letter volume and polish no longer determine outcomes. Fixed escalation timelines, verified consequences and forcing human commitment matter more than ever.
The bottom line
The task list is public. The machine reads records, writes letters, answers inquiries and explains policies โ for you and against you, at 7ร speed.
It still can't pay an invoice. Build your entire process around the two things left on the human side of the ledger: the decision, and the payment.
