The 2026 Bookkeeping Playbook

The close moved

Bookkeeping used to be a monthly event, a week of catch-up work that started on the first business day and ended when the controller ran out of patience. The team pulled bank statements, matched them by hand, chased people for receipts, coded a pile of card charges, and then booked the accruals. The books were correct for about one day a month.

That rhythm came from a constraint: a person had to touch each transaction, and people work in batches. Agents remove the constraint. An agent can code a card charge four seconds after the swipe, match a bank line the morning it posts, and email a vendor for a missing invoice before anyone at your company knows it is missing. The work does not disappear. It moves from the end of the month into each day.

This changes the job of the person who runs your books. You stop asking “how fast can we close?” and start asking “what is still open today, and why?” The answer is almost always a short list of judgment calls. This playbook is about how to build that daily rhythm, how to keep the judgment calls with people, and when to hand part of the work to an outside firm.

Marisol Achterberg, the controller at Tidewater Labs, wrote this with us. She runs a three-person accounting team for a 140-person software company with entities in the United States and the Netherlands. Three other finance leaders added their methods, their numbers, and the places where they disagree.

Code at the swipe

Start with card spend, because it is the highest volume and the lowest risk. At Tidewater, 1,900 card transactions a month across 112 cardholders used to take one accountant about four days to code. Most of those charges repeat: the same software vendors, the same airlines, the same coworking day passes. A person coding them is doing pattern matching with a long memory, which is exactly what an agent does well.

The method is simple. When a charge posts, the agent reads the merchant, the amount, the cardholder, the cardholder’s department, and any memo or receipt. It proposes a GL account, a department, a class, and a vendor record. If it has seen the same pattern at least five times with no correction, it posts the code. If not, it asks the cardholder one question in Slack, such as “Is this $640 at Delta for the Rotterdam customer visit or for the team offsite?” The answer becomes part of the record.

1. 01Export twelve months of coded card history and give it to the agent as its starting memory.
2. 02Set a confidence rule: auto-post a code only after five matching, uncorrected examples.
3. 03Route each uncertain charge to the cardholder, not to accounting, with one specific question.
4. 04Have the agent log every correction an accountant makes, so the rule set improves each week.
5. 05Review a random sample of 25 auto-posted charges each Friday and record the error rate.

Do not start with bills or payroll. Those carry larger amounts, approval chains, and tax treatment, and an error there costs more than a misfiled lunch. Earn trust on card spend first, publish your sample error rate to the team, and expand only when the number holds for two consecutive months.

Reconcile every day

Daily reconciliation is the habit that makes a short close possible, because it removes discovery from month-end. If every bank, card, and payment processor account matches the ledger each morning, month-end has nothing to discover. You have only the work that depends on the calendar: accruals, deferrals, and the review. Teams that reconcile monthly spend most of the close finding problems that are three weeks old.

An agent runs the match overnight. It pulls the bank feeds, the Stripe payouts, and the card settlements, matches each line to a ledger entry, and writes a short morning report: what matched, what it matched with a rule it has not used before, and what it could not match. The last list is the only part a person reads. At Northpine Health, that list averages nine lines a day across 44 bank accounts.

• Reconcile every cash account daily, including payment processors and payroll clearing accounts.
• Keep one exceptions list, sorted by dollar amount, with the age of each item.
• Set an age limit. Northpine escalates any exception older than five business days to the VP.
• Never let the agent force a match with a plug entry. An unmatched line stays unmatched until a person decides.

Chase receipts automatically

Missing documents are the slowest part of most closes, and the reason is social, not technical. An accountant does not want to email the CEO a third time about a hotel receipt. An agent has no such problem. It asks politely, at the right time, in the right channel, and it keeps asking until the document arrives or the deadline passes and a person steps in.

Cobalt Courier pays 260 independent contractors and about 90 vendors, many of them small fuel stations and repair shops that send invoices by text message photo. Their agent checks every bill and every card charge over $75 for a document. When one is missing, it messages the person who made the purchase within an hour. For a vendor, it replies to the original email thread and asks for the invoice by number. On the third business day of the close, it sends one summary to the department head with each item still open.

Two rules keep this from becoming spam. First, batch requests to one message per person per day, with every open item in it. Second, give the agent a stop condition. After three requests, it stops asking and escalates to a person, who decides whether to accept a memo, reclassify the charge, or treat it as personal spend to recover.

The five-day close

The close becomes short once the daily work runs by itself, and what remains is calendar work. With coding, reconciliation, and chasing in place, the close becomes a checklist of work that only the calendar can start. Most AI-native teams we spoke with close in three to five business days. A faster close is possible, but past day three the gain is small and the risk of a rushed accrual grows. Pick a target your team can hit every month, not once.

Here is the checklist Tidewater runs. The agent does the preparation for each step and attaches its work. A person signs off on each step in the close tool, and no step closes without a name on it.

1. 01Day 1: The agent confirms all cash accounts reconciled through the last day of the month and posts the final card and bill batch.
2. 02Day 1: The agent drafts recurring entries: prepaid amortization, depreciation, and the payroll accrual from the last pay run.
3. 03Day 2: The agent drafts revenue entries from the billing system and builds the deferred revenue rollforward.
4. 04Day 2: The controller reviews accruals for open purchase orders and services received but not billed.
5. 05Day 3: The agent runs the flux analysis and writes a first explanation for every account that moved more than 10% and $5,000.
6. 06Day 3: Owners of each variance confirm or correct the explanation in one shared thread.
7. 07Day 4: The controller reviews the balance sheet reconciliations and signs the close.
8. 08Day 5: Reserved for the intercompany true-up and the Dutch entity, which still needs a local review.

Notice what is not on the list: matching, coding, and chasing. Those are done before day one. If your close checklist still starts with “download bank statements,” the agents are not yet doing the daily work, and a faster close will not come from a better checklist.

Keep judgment human

Some decisions look like data entry but are really estimates, and an estimate needs an owner who can defend it to an auditor. Agents are good at preparing these decisions: they collect the contracts, compute three options, and show what each one does to the income statement. A person must still make the call and write down why.

Every contributor drew the line in a slightly different place, but the core list was the same. Keep these with a person, even when the agent drafts them:

• Accruals for services received but not billed, where the amount is an estimate.
• Revenue recognition for a contract with non-standard terms, credits, or a change order.
• Bad debt and allowance reserves.
• Capitalization decisions for software development and large equipment.
• Any entry to equity, intercompany balances, or a prior period.
• Reclassifications that change a reported metric, such as gross margin.

A useful test: if an auditor would ask “who decided this, and on what basis?”, the answer must be a name and a memo, not an agent run. Have the agent write the memo draft and attach the support. Have the person edit it and sign it. That record satisfies your auditor and teaches the agent how you think for next month.

Pick your ledger

Your general ledger decides how much of this an agent can do, more than any other tool you buy. A ledger with a clean API, a real-time sync to your bank and billing systems, and a clear audit log lets an agent post, match, and explain without a person in the middle. A ledger that needs a desktop session or a nightly file import forces a person back into the loop for every step.

Do not change your ledger to get agents. A migration costs a quarter of attention and resets your history. Change it when you add an entity, a revenue model, or an audit requirement the current ledger cannot support, and choose the new one with agent access as a main requirement.

• PilotBookkeeping, tax, and advisory for startups and small businessesAI Accountant that runs bookkeeping end to end
• DigitsAI-native accounting software for businesses and accounting firmsAutomated bookkeeping with real-time financials
• RilletAI-native ERP and general ledger for finance teamsZero-day close for finance teams
• NumericClose management, reporting, and cash software for accounting teamsThe data platform for finance teams

When to outsource

An outsourced bookkeeping firm is still the right answer for many companies, and agents change which part you hand off. Firms like Pilot now run their own automation under the hood, so you are buying a result, not hours. The question is not “firm or in-house?” but “which decisions need someone who knows our business every day?”

Fernhill Goods is the clearest example. With 85 people, two sales channels, and inventory in three warehouses, they keep one finance operations lead in-house and send the monthly close and tax work to an outside firm. Their agent does the daily coding, the Shopify and retail payout matching, and the receipt chasing, so the firm receives clean books on day one and charges them for review, not cleanup.

Use these rules to decide:

• Under about $5 million in revenue with one entity: outsource the close and let the firm run its own automation. Keep one internal owner for approvals.
• Between $5 million and $50 million: keep daily work and judgment in-house with agents, and outsource tax, audit preparation, and any foreign entity.
• Above $50 million or preparing for an audit or IPO: build an internal controller function. Use firms for specific projects, not the core close.
• At any size: never outsource the person who approves accruals and revenue judgment calls. That person must know your contracts.

So what do you do?

Start small, measure the error rate, and move the work into the day one step at a time. The teams in this playbook did not switch on an agent and close in three days the next month. Tidewater took seven months to go from eleven days to three and a half, and most of that time went into clean history and clear rules, not software.

1. 01This week: write down your current close length, your miscoding rate, and your receipt completion rate. These are your baselines.
2. 02Month one: put an agent on card coding with a five-example confidence rule and a weekly sample review.
3. 03Month two: add daily reconciliation for every cash account and one shared exceptions list with an age limit.
4. 04Month three: turn on automatic document chasing with a three-request stop condition.
5. 05Month four: rewrite your close checklist so that day one starts with accruals, not matching.
6. 06Every month: list the judgment calls by name and owner, and require a signed memo for each one.

Then look at what your accountants do with the time they get back. At every company we spoke with, the answer was the same: better analysis, faster answers to the CEO, and fewer surprises in the audit. The books are correct every morning, and the people who keep them spend their days on the decisions that need them.