Claude scheduled tasks handle weekly and month-end reports as two saved tasks on a paid plan: a weekly one that drafts Monday's sales report from your latest export, and a month-end one set to manual, since there is no monthly cadence (only hourly, daily, weekly, weekdays or manual). Create each from a conversation or the Scheduled page, test both, then start month end on demand.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowMonday reporting follows a clock; month-end reporting follows readiness. If the accounts export has not been checked, the calendar turning over does not make it complete. Saving a manual task still avoids rewriting the instructions every month, while letting the owner decide when the right files are ready to process. Before relying on either, check the run history under Scheduled rather than assuming silence means success.
Should a Claude scheduled task run on a clock or on readiness?
A Claude scheduled task should run on a clock when its inputs arrive on a fixed timetable, and be set to manual and started on demand when a person must confirm the files first.
Define the Monday job in one sentence: “Prepare last week's sales review from the approved export for the owner's meeting.” Define the month-end job separately: “Prepare the draft monthly pack after the owner confirms the input manifest.” The input manifest is simply a list of the files and checks required. These jobs can share definitions, but they should not silently share a reporting period.
For an illustrative driving school, Monday's review covers completed lessons, cancellations and recorded sales. Month-end adds adjustments and checks that may arrive later. A weekly view can be useful before every figure is final, provided it is labelled provisional. A month-end pack should not imply that the accounting records are closed merely because Claude produced a tidy document.
Agree a source owner and reviewer. The administrator may export the data, while the owner checks the narrative and decides actions. The scheduler only starts processing. It does not certify that an export is current, an adjustment is approved or a recommendation should be implemented. Put those decisions into the operating procedure so the report can be reviewed when the usual owner is absent.
Use distinct task names: “Sales | previous complete week | draft” and “Month end | approved inputs | manual”. Avoid names such as “Business update” that conceal their purpose. The name should let someone spot the wrong job before starting it. For the wider process behind a reporting pack, see automating monthly management reports.
If you want help fitting this into how your business actually runs, I offer one-to-one AI consultations for small businesses.
Where do Claude scheduled tasks run?
Claude scheduled tasks run in the cloud, even with your computer asleep, unless the task is given a folder on your computer.
Scheduled tasks are on every paid plan (Pro, Max, Ultra, Team and Enterprise), both in Claude Cowork and in the new combined Claude experience that is rolling out. If your message box no longer shows separate Chat and Cowork options, you have the new experience and some screens differ from older guides. Review the current scheduling documentation against your account rather than following an old screenshot.
Scheduled tasks run in the cloud, so they keep their cadence while your computer is asleep or the Claude Desktop app is closed. The exception is a task given a folder on your computer: picking one turns on Require this computer, and the task then runs only while that computer is awake with Claude Desktop open (the option appears only where an organisation has Computer access turned on). On Team plans, cloud running also depends on the Run Cowork in the cloud switch, on by default; on Enterprise it stays off until an owner turns it on. Do not promise unattended processing of a file that lives only on a laptop in a drawer.
For the driving school, a sensible pilot uses an approved export saved where the task can actually access it. Before scheduling, run a simple inspection request asking Claude to identify the filename, reporting period and row count. Compare those with the source. If access fails, repair that connection or upload route before adding a recurrence. Repeated scheduling does not fix missing permission.
Keep the report destination equally explicit. Each scheduled run is its own session, and the files Claude creates appear alongside that conversation, ready to preview or download; upcoming and past runs are listed under Scheduled in the left sidebar on any device. Ask for the draft with a clear report identifier and input list. If you also want a copy saved to a shared drive through a connector, test that write once by hand before making it part of the job's acceptance conditions.
Prepare a Monday export that cannot hide stale data
Create a consistent export containing a unique lesson or transaction ID, service date, status, amount and any refund reference. Remove personal learner details that the summary does not need. Include a control record stating the reporting period, export time and expected number of rows. This lets the task detect an empty file or the previous week's export rather than producing an apparently normal report from it.
Anthropic's own safety page is blunt about scheduled tasks: start with low-risk jobs such as summaries, avoid tasks that touch sensitive files, send messages or make purchases, review the output after every run and pause tasks you no longer use. It also advises against giving Cowork access to local files that hold financial documents. For a sales report, that means a stripped-down export rather than the accounts system: IDs, dates, statuses and amounts, with no bank details or customer contact data. Scheduled tasks also cannot use accounts linked through Claude's Personal Finance feature, so a bank feed will not reach the Monday run that way.
Suppose the illustrative driving school records 120 completed lessons at an average recorded sale of $45. Gross recorded sales are $5,400. Four refunds of $45 total $180, leaving $5,220 after those refunds. Last week's comparable net figure was $4,950. The difference is $270, or about 5.5% using $4,950 as the denominator. These are sample operational figures, not accounting advice.
State whether refunds are selected by processing date or by the lesson they relate to. Those approaches answer different questions. For this sample, the report uses refund processing dates within the week and names that convention. If one refund relates to an older lesson, the report should not quietly remove a current completed lesson to make the counts agree. Preserve both the activity count and the refund explanation.
A concrete duplicate test is to include transaction L-084 twice in a harmless test copy. If each row is $45, an unchecked total would overstate sales by $45. Ask the task to identify duplicate IDs, withhold a final total and show a provisional calculation only if your procedure allows it. Do not let it delete rows from the source export to solve the discrepancy.
Another test is a date-range mismatch: the export control says it contains the previous complete week, but one row belongs to the current week. The draft should identify the excluded row and explain the period rule. Keep the original row available for review. A report can be numerically accurate for the wrong period, which makes a date check just as necessary as a sum check.
If the export is the slow part, try The Complete Excel Course with AI Integration on Coding Liquids.
How do you save a Monday scheduled task in Claude?
Save a Monday scheduled task in Claude by describing the job and its timing in any conversation, or by opening Scheduled and choosing New task, then setting the frequency to weekly. In the new experience, a request such as “Every Monday at 8:30 am, prepare the draft sales report from the approved export” starts it: answer Claude's questions, then check the proposed name, schedule and instructions before clicking Schedule. In Cowork, click Scheduled in the left sidebar, then New task, and choose Create with Claude or Set up manually; typing /schedule inside a Cowork task also starts one. Set up manually asks for a task name, the prompt, the approval mode and the frequency (hourly, daily, weekly, on weekdays or manually), plus an optional model and folder.
After saving, open the task under Scheduled and confirm the upcoming run shows Monday at 8:30 am. A time typed in a prompt is not enough if the saved schedule differs. Record who will check the output and when: for example, the administrator checks for a draft at 9 am before a 9:30 am owner review.
Every Monday at 8:30 am, prepare a DRAFT sales report.
Use the previous complete Monday-to-Sunday reporting week.
Read the approved sales export and its control record.
First show file names, export time, period and row count.
Validate IDs, dates, missing amounts and refund links.
If required data is missing or stale, return BLOCKED with the reason.
Calculate completed activity, gross sales, refunds and net sales.
Compare with the supplied equivalent prior-week figure.
Show arithmetic and flag differences; do not invent causes.
Return: controls, results, exceptions, questions for the owner.
Do not send messages, change records or approve any action.
The expected illustrative result begins “DRAFT; period checked; 120 completed lessons; gross $5,400; refunds $180; net $5,220; prior net $4,950; increase $270, approximately 5.5%.” It should then list unresolved issues. If it says growth came from an advertising campaign without campaign evidence, remove that explanation and strengthen the instruction to distinguish observed change from a proposed explanation.
Run the saved task on demand before relying on the next Monday trigger. Compare its totals with the export and check that its output identifies the source version. Repeat after correcting any instruction. You are verifying a saved process, not only a prompt pasted into an unrelated conversation. The output must remain useful when the person who wrote the instructions is not watching every step.
Keep month end manual until the input pack is ready
Create the second task with the frequency set to manually. Claude offers no monthly cadence, so this is the honest way to run a month-end job: the instructions are saved once, and the input owner starts the run on demand from the Scheduled page after marking the manifest ready. The month-end manifest might contain the sales export, refunds register, approved adjustment list and prior-month comparison. Do not assume the task automatically inherits the right files from a different run or from a conversation you had elsewhere.
For an illustrative photography studio, monthly recorded sales are $18,600, refunds $600 and approved adjustments reduce the operational total by another $200. The resulting sample total is $17,800. If the prior comparison is $16,000, the difference is $1,800 or 11.25%. Ask the task to show the adjustments separately, since combining them into “refunds” would misdescribe the business record.
Run only when I start this saved task on demand.
Use the month named in the approved input manifest.
Check that every required file is present and marked ready.
If the month or readiness is unclear, stop and ask.
Prepare a DRAFT pack: controls, totals, comparison, adjustments,
unresolved items and owner questions.
Keep cash receipts, recorded sales and adjustments separate.
Show each figure's source and calculation.
Do not post entries, alter source files, send the pack or mark books closed.
Finish with a human review checklist and missing evidence.
A useful illustrative output says “Refund register present; adjustment approval missing; pack blocked.” An unhelpful one assumes the adjustment is zero because its file was absent. Missing input is not evidence of no activity. Make that distinction explicit in the instructions and test it by temporarily omitting the sample adjustment file from a test pack.
Only the responsible person should decide whether the pack is ready for external use. Ask the accountant to review classifications and financial interpretations where needed. For more detail on closing controls, use speeding up month-end close; for a report built from accounting balances, see Claude reporting from a trial balance.
Which approval setting suits a draft-only Claude scheduled task?
Manual approval suits a draft-only Claude scheduled task during the pilot, with any connector tool the report does not need set to Blocked. Claude has three permission modes: Manually approve (Claude asks before each action), Automatically approve (an automated check reviews each action and blocks what it judges risky) and Skip all approvals (nothing checks). In the new experience the message box shows them as Auto, rolling out as the default on Pro, Max, Ultra and Team, and Manual, the default on Enterprise. Keep the job restricted to reading approved inputs and returning a draft. Check the tool permissions too: a sentence saying “draft only” is an instruction, while a connector tool set to Blocked is a control.
A manual-mode run that needs an approval simply waits. Claude's mobile app sends a notification when a Cowork task finishes or needs your input, so whoever owns the task should have the app signed in, or the Monday draft can sit half-finished until someone opens Scheduled.
Do not solve an approval interruption by enabling broad actions that the report does not need. If a task pauses while trying to send a message, first ask why sending was part of the job. The intended output is a draft for the owner. Removing the unnecessary send step is often the right repair. Use the vendor's safety guidance when reviewing permission choices.
A florist's illustrative Monday report may need to read order totals but has no reason to change delivery addresses or issue refunds. Test that separation using a request to “fix the missing address” embedded in a sample note. The reporting task should flag the record for a person, not act on the note. Treat file content as evidence, not as a new set of permissions.
If you later choose an approval mode that permits more unattended work, repeat the acceptance tests with the actual available tools. Document the authorised actions narrowly and retain a human release step for the finished report. An automated check can miss a risky action; do not describe Auto as equivalent to the owner's review. Keep purchasing, sending and record changes outside this reporting pilot.
How do you know a Claude scheduled task missed a run?
You spot a missed Claude scheduled task run by checking, not by waiting for an alert: keep a run register and set a calendar reminder to open Scheduled after the expected time. Give the register the task name, expected time, period, source version, outcome and reviewer. Distinguish completed draft, blocked input, waiting for approval and no output observed. Those are your operational labels, not promises about the exact status names in Claude. A completed session can still contain the wrong period or a partial report, so inspect its content before marking the business job complete.
For an illustrative dog groomer, a weekly report showing zero appointments may mean a quiet week, a failed export or a filter set to next week. Require the task to show source row count and period beside the zero. If the expected control says 84 rows while the read file contains none, the output should be blocked. Zero is a result only when the input checks support it.
Do not depend on an alert for failures. The mobile notification covers a task that finishes or needs input, but the scheduling guide describes no separate warning for a run that never starts or stops part-way. Set an independent reminder in the owner's calendar or task list to open Scheduled at 9 am on Monday and inspect the run. Treat any notification that does arrive as a bonus, not as the control.
When a run is missing, inspect the saved schedule, pause state, source access and any local-computer requirement. When a run is waiting, inspect the approval request. When it is completed but wrong, compare the source period and calculations. Keep those branches separate; rescheduling a task cannot repair an incorrect refund convention. Make one change at a time so the next test reveals what fixed the problem.
Budget usage, review time and safe reruns
Scheduled runs draw on your plan's usage like any other Claude work; Anthropic lists no separate price per task. Its Cowork guide says multi-step work, such as running code, creating files or using connected apps, uses more than a quick question, so there is no fixed cost per Monday report. Check Settings, Usage after the first two runs. If you hit a limit, paid plans can turn on usage credits billed at standard API rates. For the plans themselves, Pro is $20 a month billed monthly ($17 a month on the annual plan) and a Team Standard seat is $25 a month billed monthly.
For planning, allow an illustrative 80 minutes for the two-task setup, 30 for acceptance tests and 15 for the first review. Four weekly reviews at ten minutes plus a 25-minute month-end check would be 65 minutes of monthly oversight. Those are proposed work allowances to measure. If the task takes longer to supervise than the manual report, simplify the input and output before expanding.
Rerun into a new draft identifier such as “week-12 draft 2”, keeping the earlier version visible until the reviewer resolves it. Tell the task not to send or overwrite approved outputs. A second run after a corrected export should state which version it used. Avoid two reports with the same name and different totals circulating before the owner has chosen the valid one.
Pause schedules during a source-system change, a prolonged absence or a period when no reviewer is assigned. Before resuming, run the saved job on demand with a current sample and check the next trigger. The dependable outcome is a report whose inputs, status and reviewer are visible every time, including the weeks when it cannot finish.
Rehearse five failures before the first unattended Monday
Build a small acceptance sheet with five rows: valid input, missing export, duplicate transaction, unapproved adjustment and unavailable destination. Beside each, write the expected behaviour before testing. The valid input should produce the checked draft. The next three should identify the precise problem. The unavailable destination should preserve the result in the session and explain that the requested handover has not completed.
Test the missing-export case without substituting an old file. A useful illustrative response is “BLOCKED: the required export for the stated week is unavailable; no current totals calculated.” Reject an output that presents the previous report's numbers as an estimate. An estimate would need separate authorisation and a different label; it is not a substitute for the scheduled report you requested.
For the unapproved-adjustment test, include a clearly marked proposal to reduce the sample monthly total by $300. The expected output should list the proposal among unresolved items and leave it out of the approved calculation. If the report deducts it, rewrite the readiness rule to require an explicit approval field. Then rerun the same test rather than moving on to an easier dataset.
Check handover failure independently from calculation success. If the destination cannot be reached but the totals are valid, the run register should show that review is still pending. The owner needs to know where the draft can be found and who must transfer it through the normal process. Do not describe the whole workflow as finished merely because the arithmetic was completed.
Store the acceptance results with the task instructions and name a deputy reviewer. When a source column is renamed, repeat the affected tests before the next scheduled run. For example, changing “refund amount” to “adjustment amount” may alter meaning as well as spelling. Ask the source owner what the field now contains; do not let the reporting task infer that the two labels are interchangeable.
Claude scheduled tasks: questions owners ask
Can a Claude scheduled task run on the last day of every month?
Not by itself. The frequency choices are hourly, daily, weekly, on weekdays or manually, with no monthly option. Save the month-end job with the frequency set to manually and start it on demand from the Scheduled page once the inputs are ready. A daily task that checks the date would spend usage on about 30 runs to do one job.
Does my computer need to be on for a Claude scheduled task?
Not for ordinary runs: scheduled tasks run in the cloud, even with your computer asleep or Claude Desktop closed. The exception is a task given a folder on your computer, which turns on Require this computer. On Team plans the Run Cowork in the cloud switch is on by default; on Enterprise an owner has to turn it on.
Should a scheduled task email the finished report to my team?
No, keep it draft-only. Anthropic's safety guidance says not to schedule tasks that send messages on your behalf, make purchases or take other actions that are hard to undo, because nobody watches the run in real time. Review the draft, then send it yourself through your normal email.
Further reads
- Claude Projects for Executive Assistants: How to Set Them Up — Give each recurring administrative job its own reference set.
- Claude Cowork for Sales: Clean a CRM Export, Draft Follow-Ups — Prepare a consistent sales export before scheduling reports.
- How to Use ChatGPT Scheduled Tasks for Business Reminders — Compare reminder scheduling with report production.
- Claude Month-End Close: QuickBooks, PayPal and a Packet for Your Accountant — Define the supporting records for a month-end handover.
- How to Make a Bank or Investor Update Deck With Claude Slides — Map approved spreadsheet figures to a lender or investor update, check the narrative and inspect the exported PowerPoint deck.
- How to Build a Rolling Cash Budget With Claude From Actuals — Turn last year's P&L into a rolling 12-month cash budget with Claude: cash timing, a filled-in assumptions tab, three scenarios and checks that tie.
Browse all AI-for-business tutorials
Sources: Claude help centre articles Schedule recurring tasks in Claude Cowork, Get started with Claude Cowork, Use Claude Cowork safely, Use Claude Cowork on web, desktop, and mobile, Use Claude Cowork on Team and Enterprise plans, and Claude Cowork and chat are one Claude; Claude pricing.
