Your team is at full capacity when its committed work uses the available time of a required person, skill or machine before the deadline, after allowing for routine duties and disruption. Compare demand with capacity by role and day or week. AI can organise that comparison, but you must verify the hours and constraints.
A team can be full even when its overall workload is below 100%. Spare packing time cannot replace a fully booked furniture finisher. Start by finding the constraint that controls delivery, then test whether a new booking fits around it. A single team-wide percentage can conceal the answer.
Choose the booking decision your plan must support
Write down the question you need to answer repeatedly. It might be “Can we accept another catered event next Friday?” or “What delivery date can we offer for this cabinet?” These questions determine the time window and the level of detail. You do not need a complex forecast for every possible business activity.
For work delivered over several weeks, start with a rolling four-week view. For work concentrated into service windows, add a daily view for the coming week. Choose these as practical starting points, not fixed rules. A month can have plenty of spare hours while one afternoon is impossible.
Separate capacity planning from performance assessment. You are estimating how much work can fit, not ranking people by apparent busyness. Ask staff what preparation, checks, cleaning, communication and handover their work requires. If those activities disappear from the plan, the plan will depend on them happening invisibly.
Record deadlines as either fixed or negotiable, with the person authorised to change them. An AI suggestion to “move the delivery to Monday” is useful only if someone checks the customer agreement and transport arrangements. A schedule is not feasible merely because its hours add up.
Build supply from real availability, not contracted totals
Use one row per person and planning period. Record paid or agreed hours, absence, routine non-project duties, capacity available for planned delivery, and skills. Then reserve time for the interruptions that your own records show happen. Keep each deduction in one place so you do not subtract the same time twice.
The basic calculation is simple: available hours minus absence minus routine duties minus your disruption reserve equals bookable hours. Routine duties include recurring meetings, ordering and general administration. The reserve covers less predictable work such as urgent customer queries or minor rework. Explain what each bucket contains.
Do not apply an unexplained universal utilisation target. Utilisation is the share of a defined capacity figure already assigned to work. A steady production process and a service team handling unpredictable calls need different reserves. Start with a provisional allowance, then compare it with the interruptions actually recorded over several weeks.
Illustrative absence check: a catering team member has 32 available working hours next week, eight hours of approved leave and four hours of ordering duties. Before allowing for disruption, there are 20 delivery hours. An old rota showing 32 would overstate capacity by 12 hours. Updating the absence record fixes the cause; a more elaborate AI prompt does not.
Use a spreadsheet such as Excel for the arithmetic. Its SUMIFS function can total hours for matching roles and weeks. Keep the totals visible and test a small example manually. An assistant can help draft the table structure, but a manager should approve who is qualified and available for each activity.
A furniture team has spare hours and still cannot take a job
Here is an illustrative weekly plan for a four-person furniture maker. The team begins with 140 working hours. Fourteen hours are unavailable because of leave, and 18 hours cover routine duties. That leaves 108 hours. The owner reserves another 12 hours for interruptions, giving 96 hours that can safely be promised in this example.
Those 96 hours are then allocated by skill. Preparation has 36 hours, assembly has 38 and finishing has 22. These allocations already account for people who can perform more than one task; no person's hours appear twice. Existing orders need 24 preparation hours, 32 assembly hours and 32 finishing hours.
| Activity | Bookable hours | Committed hours | Remaining capacity |
|---|---|---|---|
| Preparation | 36 | 24 | 12 hours |
| Assembly | 38 | 32 | 6 hours |
| Finishing | 22 | 32 | 10 hours short |
| Total | 96 | 88 | 8 hours on paper |
The total workload is 88 divided by 96, or about 91.7% of bookable capacity. That percentage looks manageable. Finishing, however, is at 145.5% of its allocation. The team is already overcommitted for the promised finishing dates, despite having eight spare hours in the aggregate calculation.
A new order needs six hours of preparation, six of assembly and eight of finishing. Adding it creates 108 hours of demand against 96 available, and finishing demand becomes 40 hours against 22. The owner must address an 18-hour finishing shortfall, not simply ask everyone to work a little faster.
Consider three options. First, move a confirmed flexible delivery after agreement with the customer. Second, use a qualified subcontractor for a defined finishing batch. Third, offer the new customer a later slot. Test dependencies, equipment, supervision and collection times for each option. A subcontractor's hours are useful only when the work can actually be handed over.
Suppose an already assembled batch needs ten finishing hours and its customer agrees to a later date. Moving that work reduces this week's finishing demand to 22 hours. It resolves the existing shortage, but it still leaves no finishing room for the new order. This is a useful distinction when explaining why solving one delay does not create a new saleable slot.
Record the decision: “New order offered the following week; existing flexible batch moved with approval; finishing capacity to be reviewed on Thursday.” Keep the approval and revised dates beside the relevant orders. The plan should show what changed, not just a refreshed set of green cells.
Count the work customers have already bought
Build a demand register with Job reference, Activity, Remaining hours, Required skill, Earliest start, Deadline, Status and Estimate owner. Use remaining hours rather than the original quotation estimate. A 20-hour task with 12 hours completed contributes eight hours to future demand, provided the person doing it confirms that estimate.
Include internal obligations with deadlines. Stocktaking, equipment maintenance and agreed improvement work also consume time. Either reserve them in the supply calculation or list them as demand, but not both. Write down which approach you chose so next week's update does not quietly change the basis.
Illustrative service-window constraint: a food truck needs two qualified staff for a four-hour event, plus two hours each for preparation and clear-down combined. That is 12 staff-hours (two people for six hours each), not six. Six hours from one person cannot cover the booking. The plan needs both the total effort and the requirement for simultaneous attendance.
Illustrative peak hidden by an average: an e-commerce homeware brand expects 120 returns in a week, at ten minutes each. That is 20 hours. If 80 arrive on Monday, Monday needs about 13.3 hours. A worker with four returns-processing hours that day cannot clear them, even if the weekly total fits. Decide whether the service promise allows a queue.
Use ranges for uncertain work. A bespoke assembly might need six to ten hours depending on a customer's final measurements. Keep the normal and cautious estimates in separate columns. Ask the delivery owner what event would push the task towards the upper end, and record that condition beside the estimate.
For clearer starting data, connect this process to practical time recording. If you have little history, collect estimates and actual times for a few representative jobs. It is better to show an uncertain range than a precise number with no basis.
Keep enquiries separate from confirmed commitments
A possible order is not a confirmed booking, but ignoring the pipeline can leave you unprepared. Use three views: confirmed work only, a likely scenario and a busy scenario. Make the status of each job explicit. Do not blend every enquiry into a single number and treat the result as a promise you can safely make.
Illustrative pipeline calculation: a delicatessen has two possible staffed tastings. One needs eight hours and is assigned a 50% planning probability; the other needs 12 hours at 25%. Their weighted demand is seven hours: four plus three. Yet the actual outcomes are zero, eight, 12 or 20 hours. Seven hours is a forecasting aid, not a bookable event.
If both tastings require the same Saturday slot, test the 20-hour scenario before accepting either customer's confirmation. Decide how long a provisional hold lasts and who can release it. A spreadsheet full of indefinite holds makes the team look permanently busy even when orders never materialise.
Do not let AI invent probabilities from friendly emails. Derive them from your own conversion history where possible, or clearly label them as the owner's estimates. Keep those estimates separate from customer commitments. Review whether they were useful after the enquiry is won, lost or postponed.
A cancellation also needs a deliberate update. If a catering event is cancelled after six preparation hours have already been worked, release only its remaining work. Keep the six completed hours in the historical record, and check whether cancellation administration or stock handling adds new tasks. Removing the entire job would make the past week's workload look artificially light and could distort the estimates you use for similar events. Confirm that any reserved vehicle or subcontractor time is actually released before showing it as available elsewhere.
Give AI constraints and ask for options you can reject
Use an approved assistant, such as Claude, to turn the checked supply and demand tables into a list of clashes and choices. You can paste a small redacted table without connecting scheduling systems. Claude Pro has a list price of $20 a month; an existing suitable assistant may be enough for this exercise.
Compare these capacity and demand tables for the coming week.
Keep each person's total allocation within their available hours.
Respect qualifications, fixed deadlines and task dependencies.
Do not assume overtime, subcontractors or customer agreement.
Return:
1. Overloaded roles and dates, with the arithmetic.
2. Missing information that prevents a booking decision.
3. Three possible changes, each with an approval needed.
4. The earliest feasible slot for the proposed job, if supported.
If no feasible slot exists in the data, say so.Illustrative output and correction: the assistant reports, “Finishing is ten hours over capacity. Reassign ten preparation hours to finishing.” The arithmetic is right, but the recommendation ignores qualifications. Reject it unless a named person can do that work and is actually available. Ask for a revised option that respects the skill column.
A better output would say, “No feasible allocation exists within the supplied finishing capacity. Ask whether the ten-hour flexible batch can move, or whether qualified outside capacity is available.” This gives the owner a decision without presenting an unapproved assumption as a completed solution.
Check that the AI has not used the same person in two roles at the same time. A multi-skilled worker contributes flexibility, not duplicate hours. If they spend six hours assembling, those six hours must leave their availability for preparation. Keep a person-level cross-check even when the main report is grouped by role.
Look for physical constraints after the hours fit
Illustrative equipment bottleneck: a butcher has 12 hours of cutting work and enough qualified staff, but the relevant preparation station is available for only eight hours because of another production run and cleaning. Adding a spare packer does not remove the four-hour shortage. The constraint belongs in the equipment schedule as well as the staffing plan.
Distinguish effort from elapsed time. A finishing process might need one hour of hands-on work followed by an approved waiting period before the next stage. Use the process specification for that waiting time. AI must not compress curing, chilling, inspection or maintenance intervals merely to make the calendar fit.
Check storage, vehicle access and handover requirements where they matter. A catering company may have enough preparation hours for two events but insufficient approved storage space for both orders together. That is a capacity question even though it does not appear in the labour total.
Use the staff skills matrix to identify where the business relies on one person. Broader capability may eventually improve capacity, but someone learning a task can require supervision at first. Do not count a future skill as fully available before the person responsible confirms readiness.
Use a weekly acceptance check and learn from the misses
Before accepting work, check that its effort fits, the required skill is available, the equipment can be used, dependencies can finish in time and the disruption reserve remains intact. If one condition fails, offer a different date or scope, or obtain approval for a specific capacity change. Record the decision with the booking.
Keep commercial judgement alongside the plan. A job that fits is not automatically worth taking. Compare its likely contribution and interruption cost using project profitability tracking. A small urgent order can consume the only slot needed for a larger existing commitment.
As an initial planning allowance, set aside 90 minutes to build the first small-team view and 20 minutes a week to update it. At an illustrative internal time value of $30 an hour, those amounts represent $45 for setup and $10 per review. Cleaning unreliable records or coordinating several teams will take longer.
For four weeks, record promised dates, actual completion, estimated hours, actual hours and the reason for any missed commitment. Review the pattern with staff. Repeated overruns on one task suggest a weak estimate; clashes concentrated on one day suggest scheduling; repeated disruption suggests an inadequate reserve or an underlying process problem.
Also inspect unused capacity. If work is consistently completed well below its estimate, adjust future estimates carefully after enough comparable jobs. Do not remove every spare hour after one unusually easy week. The useful result is a plan that supports credible customer promises and manageable work, with clear reasons whenever the answer to a new booking is “later”.
Questions about sharing your capacity plan
Should customers see our internal capacity percentages?
Usually, give customers the information they can act on: an available start date, a delivery window and any decision deadline. Internal percentages can confuse people because they depend on your planning assumptions. Keep individual availability and staff details private. If a customer needs a formal capacity report, agree its definitions before sending it.
How should I reserve capacity for a priority customer?
Record reserved hours as a separate commitment with a named owner and an expiry date. Explain who can release them if the customer does not confirm. Do not show those hours as freely available while also promising the customer access to them. Review whether the reservation remains commercially worthwhile when the agreement is renewed.
Can I show staff capacity without revealing employment details?
Yes. A shared planning view can show available hours and qualified activities without showing pay, medical information or reasons for absence. Keep sensitive records in the appropriate restricted system. Ask the manager responsible for staffing to provide the capacity figure, and give the wider team only the information needed to allocate work.
Further reads
- AI Shift Scheduling vs a Rota Spreadsheet for Small Teams — Turn an agreed capacity limit into a workable rota.
- How to Automate Holiday Requests and Absence Tracking — Keep approved absences reflected in future availability.
- AI Job Scheduling and Dispatch for Trades and Field Service Teams — Handle travel and appointment windows when allocating mobile work.
- Budget vs Actual: How to Explain Monthly Variances With AI — Compare planned workload with what actually happened.
- How to Get a Weekly Business Summary Emailed to You by AI — Include capacity decisions in a short weekly owner report.
- How to Build a Job Costing Sheet in Excel With AI Help — A four-table Excel job costing workbook, the overhead maths, AI prompts for the formulas, and a bespoke commission costed from quote to finish.
- How to Forecast Next Quarter's Sales With AI Using Your History — Three baselines, a backtest that exposes over-confident models, an adjustments log, and a wine merchant's festive quarter forecast worked end to end.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Microsoft Support, SUMIFS function; Anthropic Claude pricing page (Pro plan), checked September 2026. All team, order and hour figures are illustrative.