Follow one expense from the card swipe to the month-end close. Turn the dial to see what AI takes off your plate at each step, what it hands back to you, and which tools do it today.
Some tools let an assistant like Claude read your expense data directly. Brex hosts its own connector in beta, and Expensify has a read-only one any member can add.
Free email course
One lesson every few days, each a four-minute read with one small thing to try that week. It starts with the jobs almost every business has: paying bills, payroll and the month-end close.
The last lesson links back to this walkthrough and the other seven.
Today. Employees guess what the policy allows, put it on a personal card or the one shared company card, and finance finds out at month end.
With AI. Answers what the policy allows from your written policy, and checks merchant, amount and budget at the swipe. In: a question, or a card swipe. Out: an in-policy purchase on the right budget, or a decline with the reason.
Who decides. AI-assisted: Card rules decline, finance reviews the flags. AI-run: Rules and limits decide at the swipe, and finance sets them.
Control. Each person gets their own card with limits by merchant, category and amount, so nobody shares a card.
Tools that do this. Ramp (score 85): card controls block spend by merchant, category and amount before it happens. BILL Spend & Expense (score 74): budget-linked cards by team, project or vendor, with charges hitting the budget in real time. Brex (score 83): Brex Assistant answers policy questions for employees. Zoho Expense (score 73): answers policy questions from your uploaded travel policy.
Today. Chase every receipt by email and Slack, then match faded paper receipts to card lines by hand.
With AI. Pulls receipts from email and partner apps, reads them and matches each one to its card charge. In: the card charge, the employee's inbox, a phone photo. Out: a card line with its receipt attached, or a reminder to the person who spent.
Who decides. AI-assisted: AI finds and matches, and the employee confirms the gaps. AI-run: Matched receipts close, and gaps get an automatic reminder.
Control. Cards can freeze until the receipt arrives. Over $75, the IRS still wants the merchant's receipt.
Tools that do this. BILL Spend & Expense (score 74): Transaction Agent pulls receipts from Gmail, Outlook and partners such as Lyft. Expensify (score 75): Concierge reads and matches receipts, and flags ones that may be AI-generated. Zoho Expense (score 73): reads receipts in 38+ languages and auto-matches card feeds. Ramp (score 85): can auto-lock cards when receipts are late, on Plus.
Today. Pick the category, GL account, class and project for every line, and split out tax by hand.
With AI. Suggests the category, GL account and description from the merchant and past coding. In: the receipt, the merchant and your coding history. Out: a coded expense ready for review.
Who decides. AI-assisted: A person accepts or fixes each suggestion. AI-run: Confident codes post, the rest wait.
Control. Corrections are logged and teach the next suggestion. Some tools sync nothing until a person accepts it.
Tools that do this. Navan (score 82): Expense Agent fills in descriptions, categories and GL codes. Brex (score 83): suggests GL codes and mapping rules, and imports NetSuite classes, departments and locations. Expensify (score 75): Concierge codes expenses and auto-corrects incomplete entries. Rho (score 73): Rho Close suggests GL codes, and nothing syncs until a person accepts it.
Today. The manager glances at the report, and finance spot-checks a sample after the money has gone out.
With AI. Checks every line against the policy and flags duplicates, altered or AI-generated receipts, ranked by risk. In: the coded expense, its receipt and your written policy. Out: a clean expense, or a flag with its reason and risk level.
Who decides. AI-assisted: Every flag goes to finance. AI-run: Low-risk lines pass, high-risk ones go to a person.
Control. Every line is checked against the same written policy, instead of a sample after the fact.
Tools that do this. Ramp (score 85): policy agent reviews expenses against your written policy. Brex (score 83): audit agent checks every expense against the policy you upload and ranks violations by risk. Emburse (score 76): Assurance scores each expense for duplicates and altered or AI-generated receipts (paid add-on). AppZen (score 71): scores every line low, medium or high risk, instead of sampling 10 to 20 percent of reports.
Today. The manager approves a stack of reports on Friday afternoon, mostly without reading them.
With AI. Auto-approves low-risk, in-policy items and sends the rest to the manager with the reason. In: the checked expense and its risk flags. Out: a short queue of real decisions.
Who decides. AI-assisted: The manager approves with the AI's risk notes. AI-run: Low-risk items auto-approve, and the manager decides the rest.
Control. Nobody approves their own expense, and every approval, by a person or by a rule, is logged.
Tools that do this. Brex (score 83): review agent auto-approves low-risk expenses and escalates exceptions. Ramp (score 85): approvers can approve or reject from Claude through Ramp's own connector, with their own permissions. Zoho Expense (score 73): with Zoho MCP, Claude can list and approve reports, with every call logged.
Today. Add out-of-pocket claims to payroll or key bank transfers one by one, and work out mileage from a map.
With AI. Works out mileage from the trip and batches approved claims into one payout. In: approved out-of-pocket claims and trips. Out: a reimbursement batch waiting for release.
Who decides. AI-assisted: Finance releases the batch the tool builds. AI-run: A person releases payouts, always.
Control. Only approved amounts can be paid, and releasing the payout is its own step.
Tools that do this. Expensify (score 75): Collect plan includes ACH reimbursement from $5 per member a month. Navan (score 82): approved amounts go back through payroll or straight to a bank account within days. Zoho Expense (score 73): GPS mileage tracking for trips.
Today. Export a CSV, re-key it into QuickBooks, and guess at accruals for reports nobody has submitted yet.
With AI. Links receipts, card payments and bookings, and syncs coded spend to the ledger. In: coded expenses, card feeds and bookings. Out: a reconciled card account, or a list of what is still missing.
Who decides. AI-assisted: Finance reviews every sync. AI-run: Clean spend syncs, gaps go to a person.
Control. Every expense carries its receipt, coding and approval into the ledger, so the audit trail is complete.
Tools that do this. Navan (score 82): Reconciliation Agent links receipts, payments and bookings. Brex (score 83): exports journal entries with accruals and amortization. Ramp (score 85): ask Claude for last month's transactions missing GL codes through Ramp's own connector. Rho (score 73): codes card and bank activity into NetSuite, and nothing syncs until someone accepts it.
“Can I expense a $65 team lunch?” Reads the question, finds the meals rule in your written policy and answers with it. AI-run sends the answer. AI-assisted drafts it for finance.
“Where is my reimbursement for the Denver trip?” Finds the report, reads its approval status and the next payout date, and writes the reply. AI-run sends it. AI-assisted drafts it for finance.
“I lost the hotel receipt. Can you just approve it?” Never approves a claim without a receipt. Drafts a request for a copy and routes the claim to a person. A person, always. Over $75, the IRS wants the merchant's receipt.
Tools that do this. Brex (score 83): Brex Assistant answers policy questions and files reimbursements for employees. Zoho Expense (score 73): answers policy questions from your uploaded travel policy. Expensify (score 75): Concierge answers questions about spend inside the app. Ramp (score 85): policy agent reviews expenses against your written policy.
Put in your own figures. The result is labour only, and you choose how much of the work you expect AI to take on.
We don't show an industry benchmark here. The best-known figure, GBTA's $58 per expense report, dates from 2015.