How to Automate Invoicing With AI: From Finished Job to Paid

Make "job finished" a status that triggers billing. An automation pulls the recorded time, fixed fee or quote into a draft invoice, AI writes clear line descriptions and checks the draft against what was agreed, a person approves it, and it goes out with a payment link. Reminders and payment matching then run inside your accounting software.

Typing the invoice is rarely the slow part. The days disappear between finishing the work and anyone noticing it's ready to bill, and again when a client queries a vague line and puts the whole invoice on hold. Measure those two gaps before automating anything; they tell you which part to fix first.

Follow me on Instagram@sagnikteaches

Where the days go between finished job and paid

The running example is an illustrative 12-person law firm handling fixed-fee work (wills, lease renewals, company formations) alongside hourly commercial work. It issues about 140 bills a month. A partner pulled three months of dates from the practice management system and the accounting software, and found this:

Connect on LinkedInSagnik Bhattacharya

From the end of the work to money in the bank took over seven weeks on average, and very little of that was anyone actually writing an invoice. Your own figures will differ, but it's worth an afternoon to get them, because a firm losing 11 days before billing needs a different fix from one whose invoices sit unpaid.

Subscribe on YouTube@codingliquids

Stage 1: one "ready to bill" trigger

Automation needs a clear signal that the work is finished. Pick one and make it the only route to an invoice:

• A matter or job status such as "Complete: ready to bill" in your practice management or job system.
• A project stage marked done in your project tool, for milestone billing.
• An approved timesheet, for work billed weekly by the hour or shift.
• A signed delivery note or completion form, for jobs on site.

In the law firm, closing a fixed-fee matter or marking an hourly matter "bill now" moves it into a billing queue. Time entries are locked at that moment, so nothing changes under the draft while it's being reviewed. The rule that made the biggest single difference: a matter can't be marked complete without the billing step being triggered, so there's no separate thing to remember.

Stage 2: draft the invoice from what's already recorded

The draft should build itself from records that already exist: time entries, the fixed fee in the engagement letter, the accepted quote, or approved timesheets. How that happens depends on your software.

• Practice management systems increasingly do this themselves. Clio's help centre describes its Manage AI assistant (formerly Clio Duo) generating draft bills from time and expense entries on a schedule, optionally refining entry descriptions (the refined ones are highlighted for review), and routing drafts to named approvers. Check which plan includes it before relying on it.
• Xero: its JAX assistant can create draft invoices from a typed request or a forwarded email, for example an invoice based on the last quote for a contact, and can approve and send invoices once they have a due date. Xero says there's currently no extra charge for JAX chat.
• QuickBooks Online has its own AI features for invoices, and its Payments AI suggests reminder schedules and drafts reminder emails that you review before approving.
• Anything else: Zapier or Make can watch for the ready-to-bill status, gather the time entries or quote, pass them to an AI step to draft lines and descriptions, and create a draft invoice in your accounting system. Drafts only, never sent from this step.

For fixed-fee work the draft is simple: the agreed fee plus any disbursements (costs paid out on the client's behalf, such as registration or search fees) with their receipts attached. In the law firm, closing a will matter at signing produces a draft for the agreed $650 fee plus the $45 electronic identity check paid out at the start, receipt attached. The only human check needed is that the client's name and address match the engagement letter, and those bills now go out the same afternoon. Hourly work is where descriptions matter.

Stage 3: descriptions that clients don't query

Raw time entries are written for the fee earner, not the client. Here are five entries from an illustrative commercial lease matter:

12/03  0.3  call w/ client re break clause
12/03  1.8  rev LL draft lease
13/03  0.4  email LL sols re rent review wording
14/03  1.2  amend lease draft + notes for client
15/03  0.5  att. call w/ LL sols

A bill that lists these as "Professional services: 4.2 hours" invites a query. A bill that lists them verbatim invites a different one ("what's att.?"). The drafting prompt:

Write the narrative for this bill from the time entries below.
- Plain English a business client understands; no abbreviations.
- Group related entries into 2-4 short sentences, in date order.
- Describe ONLY work that appears in the entries. Do not add
  advice, meetings or documents that aren't listed.
- Keep the total hours and amounts exactly as given.
- Flag any entry you can't interpret instead of guessing.

An illustrative result:

Reviewing the landlord's draft lease and discussing the break clause with you on 12 March. Corresponding with the landlord's solicitors on the rent review wording, amending the draft lease and preparing notes for you on the changes, and a call with the landlord's solicitors on 15 March. Also advising on the repairing obligations.

The first two sentences are good. The last one is invented: nothing in the entries mentions repairing obligations. It's the kind of plausible addition AI makes when it pattern-matches a lease review, and it's exactly what the fee earner must delete before approval. After the "describe ONLY" line was strengthened with "if in doubt, leave it out", this stopped happening in the firm's tests, but the review still checks every narrative against its entries.

Stage 4: a pre-send check against what was agreed

Before a person approves, a second AI step compares the draft with the engagement letter, quote or purchase order and lists anything that doesn't match. This is the check that stops the invoices that cause queries:

Compare this draft bill with the engagement letter and client
record attached. List any of the following, or "no issues":
1. Total (excluding disbursements) above 90% of the estimate.
2. Client name or billing entity different from the engagement
   letter.
3. Disbursements without a receipt attached.
4. Client requires a purchase order or reference number and none
   is on the bill.
5. Rates different from those in the engagement letter.
6. Work in the narrative that falls outside the agreed scope.
Quote the relevant line from each document for every issue.

An illustrative output for a company client:

Issue 1: Total $4,380 is 97% of the $4,500 estimate. Engagement
letter says: "We will tell you before costs exceed our
estimate." Confirm the client was told, or note the remaining
work.
Issue 2: Billing entity on draft is "[Client] Holdings Co.";
engagement letter is addressed to "[Client] Trading Co.".
Issue 4: Client record says "PO number required on all
invoices". None on draft.

Any one of those three would have held up payment for weeks: a bill to the wrong company goes back unpaid, and a corporate client's accounts team often rejects an invoice without its PO number without telling anyone. With the checks done, the partner's review takes a minute rather than ten, and approval becomes one click in the queue.

Stage 5: send it with an easy way to pay

Once approved, the invoice goes out from the accounting system with the payment options printed on it and, where it suits your clients, an online payment link. Xero and QuickBooks Online both let clients pay invoices online through connected payment providers; payment fees apply, so decide whether that cost is worth the faster payment for each type of client. Card payment suits smaller private-client bills. Larger corporate bills are usually paid by bank transfer in a payment run, and for those the PO number and correct entity matter more than a link.

The covering email doesn't need AI at all. A short fixed template does the job, with the invoice number, amount, due date and the one sentence that reduces queries: "If anything on this bill isn't clear, reply to this email and [fee earner] will explain." A named person to ask is more reassuring than a generic accounts mailbox.

Stage 6: reminders and payment matching

Automatic reminders are built into both main systems. Xero lets you set up to five reminders on one schedule for the whole organisation, before or after the due date; its defaults are 7, 14 and 21 days overdue. QuickBooks Online allows up to three automatic reminders, each up to 90 days before or after the due date, set under Account and settings, Sales, Reminders. The default wording in both is stiff, and editing it is worth half an hour. Writing late payment reminders that sound human covers the wording, the sequence and when a person should take over.

Payment matching closes the loop. Bank feeds suggest matches between incoming payments and open invoices; the ones that need a person are part-payments, one payment covering several invoices, and payments with no reference. If matching is where your month end stalls, see automating bank reconciliation and checking the matches.

Three months after the changes, the law firm's figures (illustrative) read: matter finished to bill sent, 2 days; bills queried, 1 in 20; bill sent to paid, 27 days. Most of the gain came from stages 1 and 4, the trigger and the pre-send check, not from anything clever in the drafting.

Catch the work that never gets billed at all

Slow billing is visible. Work that is never billed isn't, because nothing prompts anyone to notice it. Once a month, give an AI step the unbilled time and disbursements report from your practice or job system and ask for three lists:

From the attached unbilled time and disbursements report and the
matter list, give me:
1. Matters with unbilled time older than 60 days, with value
   and the fee earner.
2. Closed or completed matters that still have unbilled time
   or disbursements.
3. Fixed-fee matters marked complete with no bill raised.
Sort each list by value, highest first. Do not estimate values;
use the report figures.

The law firm's first run was uncomfortable reading. It found 23 matters with unbilled time older than 60 days, recorded at $31,400. Some of that was never going to be billed: time on matters with a capped fee, and hours a trainee spent learning. After the partners went through the list, $9,800 was billable and went out that week, $6,200 of disbursements had been paid out and never recharged, and the rest was written off properly instead of sitting on the report. The monthly list now takes the billing partner about 15 minutes, and the second run found under $2,000.

What it costs and how long it takes to set up

If your practice or accounting software already includes the AI drafting, the direct cost can be close to nothing: Xero says JAX chat currently carries no extra charge, and Clio and QuickBooks include their AI features on some plans, so check yours. Building the flow in Zapier instead, each invoice uses a handful of tasks (gathering entries, an AI step that uses 1, 3 or 5 tasks depending on the model tier, creating the draft, running the check), so 140 bills a month comes to roughly 600 to 1,700 tasks depending on the tier. At anything above the cheapest tier, that's beyond Zapier's Professional plan at 750 tasks and within the Team plan's 2,000, which lists from $103.50 a month billed monthly, or $69 billed annually.

A realistic set-up plan for a firm this size:

1. Week 1: measure the four gaps from the first table, and agree the ready-to-bill trigger.
2. Week 2: switch on draft generation for one team, drafts only, and compare them with manual bills.
3. Week 3: add the pre-send check and the narrative prompt; review every draft in full.
4. Week 4: edit reminder wording, turn on reminders, and roll out to the other teams.

After that, the ongoing time is approval (a minute or two a bill), the monthly sample check and the unbilled work list.

Two other billing patterns the same approach handles

Milestone billing in an engineering consultancy. An illustrative consultancy bills design fees in stages: 20% at concept, 40% at detailed design, 30% at tender issue, 10% at completion. The trigger is a stage marked complete in the project tool. The AI step reads the fee schedule, calculates the stage amount, and checks for variations agreed by email since the appointment that should be added. A realistic catch: a variation for an extra plant room, agreed in an email thread, that the project lead had forgotten to add to the fee schedule. The pre-send check found the email because the prompt asked for "any agreed additions to scope or fee in the project correspondence".

Weekly invoices in a recruitment agency. An illustrative agency bills clients weekly for temps' hours. The trigger is the approved timesheet; the invoice is hours multiplied by the agreed charge rate for each role, with overtime and night rates applied. Here the AI's job is mostly the check: a timesheet showing 52 hours for a role capped at 45 in the client's contract, or a charge rate that doesn't match the latest rate card. Because the same timesheets drive pay, the checks overlap heavily with catching timesheet errors before payday, and it makes sense to run them once for both.

Billing decisions that stay with a person

• Approving every variable invoice. One click, but a real look at the narrative and the check results.
• Sensitive or disputed matters. A bill to a client who has complained, or a matter that ended badly, needs a call before an invoice, not an automated email.
• Write-offs and discounts. Deciding to bill less than the time recorded is a judgement about the relationship. AI can show you the recorded value against the estimate; it shouldn't decide.
• Credit notes. Always raised by a person, with a reason.
• Client money rules. If you hold money on account for clients, the rules that apply to taking payment from it are yours to follow, and the automation should stop at the draft.

A monthly check on the billing automation

Once a month, pull ten invoices at random and check three things against the source records: every narrative line matches recorded work, the amount matches time, fee or quote, and the pre-send check caught what it should have. Then look at a simple report of drafts older than three days, which is where billing quietly stalls when an approver is on holiday. Keep tracking the four gaps from the first table. If "finished to drafted" stays near zero but "sent to paid" doesn't move, the problem has moved downstream to reminders and client payment runs, which is a different fix and a different tutorial.

Invoicing automation questions

Should invoices be sent without anyone approving them?

Only for simple, repeat invoices where the amount is fixed and agreed, such as a monthly retainer or a subscription. Anything built from time, variable quantities or a quote should get a person's approval, because a wrong invoice costs more in credibility and credit notes than the minute it takes to check. Approval can be one click if the draft arrives with its checks already done.

Will clients mind invoices written by AI?

Clients mind vague invoices, not who typed them. A description generated from your time records, checked by the person who did the work, is usually clearer than a rushed manual one. Where AI goes wrong is adding work that wasn't recorded, so the reviewer's main job is to confirm every line matches what actually happened.

What if my practice or job software doesn't connect to my accounting system?

Most job, project and practice management tools either sync with the main accounting systems or can be connected through Zapier or Make. If neither works, a weekly export of completed jobs into a spreadsheet can still feed an AI drafting step. It's slower, but the ready-to-bill trigger and the pre-send check still deliver most of the benefit.

Further reads

• Quote to Cash: Connect Quotes, Invoices and Payments With AI — The wider chain from accepted quote through to payment.
• AI Accounts Receivable: Predict Late Payers and Act Early — Spot which clients will pay late before they do.
• How Freelancers Use AI for Invoices, Proposals and Chasers — A lighter version of the same process for sole traders.
• Best AI Accounting Software for Small Businesses in 2026 — Compare accounting systems by the AI invoicing features they include.
• How to Use Xero's JAX for Invoices and Cash-Flow Questions — More on what Xero's JAX can do with invoices.
• How to Track Profit per Project in a Service Business With AI — See which jobs were worth billing for in the first place.
• Can a Print Shop Use AI to Price Jobs and Check Artwork? — How a print shop splits the work between AI, its price matrix and preflight software, with prompts, a proofing example and a customer report.
• How to Check a New Customer's Credit Before Offering Terms — Check a new business customer before giving 30-day terms: the application, identity checks, credit report and references, with AI summarising the evidence.
• What Finance Tasks Can AI Automate in a Small Business? — Fifteen finance jobs AI can take over in a small business, each with an example, a first step and the check that stops it going wrong.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Xero JAX product page and Xero Central (draft invoices via JAX, invoice reminders: up to five, defaults at 7, 14 and 21 days overdue); QuickBooks Online help (automatic invoice reminders, Payments AI suggestions needing approval); Clio help centre (Manage AI billing automation).