To forecast support ticket volume, take the trailing average of your weekly tickets per channel over the last 12 to 13 weeks, multiply it by your growth rate, then multiply each future week by a seasonal or event factor. To turn that into staffing, multiply forecast volume by average handle time, divide by one minus shrinkage, and divide by the hours one agent works per week.
That is the whole method. The rest of this guide shows it step by step, with a worked example you can copy into a spreadsheet, and explains how support software forecasts conversation volume when an AI agent handles part of the load.
What is support volume forecasting?
Support volume forecasting predicts how many conversations will arrive in a future period, split by channel and ideally by topic. The output is a number per week (or per day) that drives three decisions:
- How many people to schedule.
- How much AI capacity to budget for.
- When to schedule risky work, like a migration or a price change, so it does not land on a week the queue cannot absorb.
You do not need a data team. You need one spreadsheet, a weekly habit of checking the forecast against reality, and honesty about the seasons your business actually has.
The forecasting formula
The method most small teams should use comes down to one line per week:
Forecast (week) = Baseline × Growth factor × Seasonal or event factor
- Baseline is the average weekly ticket count over a trailing window. Supportbench's forecasting guide recommends a trailing 13-week baseline per channel, multiplied by account growth and adjusted for known events like launches, renewals, and seasonal peaks. We use 12 weeks in the example below because it maps cleanly onto a quarter.
- Growth factor comes from the metric that drives your tickets: orders, active accounts, or signups. If orders grow 4% a month, the factor for next month is 1.04.
- Seasonal or event factor is 1.0 for a normal week. A launch week might be 1.25. A peak retail week might be 2.0 or more. You get these factors from your own history, not from optimism.
The formula is deliberately plain. It works because it starts from a pattern you already have and adjusts only for the things you already know.
Worked example: from 12 weeks of history to a forecast
The numbers below are illustrative, for a small ecommerce team handling chat and email. Swap in your own.
Step 1: collect 12 weeks of history
| Week | Tickets (chat + email) | |---|---| | 1 | 410 | | 2 | 395 | | 3 | 430 | | 4 | 420 | | 5 | 445 | | 6 | 438 | | 7 | 452 | | 8 | 460 | | 9 | 441 | | 10 | 470 | | 11 | 465 | | 12 | 478 | | Total | 5,304 |
Step 2: calculate the baseline
Baseline = 5,304 / 12 = 442 tickets per week.
If one of those weeks contained a one-off spike, such as an outage, remove it before averaging or it will inflate every week you forecast.
Step 3: apply growth and event factors
Orders are growing about 4% a month, so the growth factor is 1.04. In week 14 there is a product launch. The last launch produced a week about 25% above normal and a smaller tail the week after, so we use 1.25 and 1.10.
| Week | Baseline | Growth | Event factor | Forecast | |---|---|---|---|---| | 13 | 442 | 1.04 | 1.00 | 460 | | 14 (launch) | 442 | 1.04 | 1.25 | 575 | | 15 (tail) | 442 | 1.04 | 1.10 | 506 | | 16 | 442 | 1.04 | 1.00 | 460 | | Total | | | | 2,001 |
That is your volume forecast for the next four weeks. It will not be exact. It will be much closer than staffing from memory.
From forecast to staffing
Volume alone does not tell you how many people to schedule. You need two more inputs: average handle time (AHT) and shrinkage, the share of paid time agents are not available for tickets (breaks, meetings, training, leave).
Workload hours = Forecast volume × AHT (in hours)
Scheduled hours = Workload hours / (1 − shrinkage)
Agents needed (FTE) = Scheduled hours / hours per agent per week
Continuing the illustrative example, with an AHT of 9 minutes (0.15 hours), shrinkage of 30%, and a 40-hour week:
| Week | Forecast | Workload hours | Scheduled hours | FTE | |---|---|---|---|---| | 13 | 460 | 69.0 | 98.6 | 2.5 | | 14 (launch) | 575 | 86.3 | 123.2 | 3.1 | | 15 (tail) | 506 | 75.9 | 108.4 | 2.7 | | 16 | 460 | 69.0 | 98.6 | 2.5 |
The launch week needs about 0.6 FTE more than a normal week. That is a decision you can make a month ahead: move someone's training, borrow a colleague for inbox duty, or plan the AI line (below) to absorb it.
Measure your own shrinkage from real schedules rather than borrowing a figure. We go deeper on this in our guide to support staffing math.
A note on queueing. For live chat, where customers wait in real time, the classic model is Erlang C. Verint's overview of call-centre forecasting traces the Erlang C and Erlang A models back to A.K. Erlang in 1917. They convert contact volume, handle time, and a service-level target into the number of agents required. For a small team on email and chat, the linear workload math above is enough for weekly planning. Erlang matters once you promise tight live-chat wait times.
How to find your seasonal factors
Every business has a seasonality shape. Ecommerce peaks in Q4. B2B SaaS often peaks when new budgets land in Q1 and again in September. Travel peaks around booking windows, not travel dates.
Cobbai's guide to support forecasting describes seasonal decomposition, which breaks ticket history into three layers: trend, seasonal, and residual. You do not need a statistics library for this. A pivot table of tickets by week-of-year across two years gives you the seasonal shape. If a week is consistently 30% above the annual average, its factor is 1.3.
History length matters. Supportbench's guidance on forecasting without fancy tools suggests two to three years of history for accurate seasonality, and documenting every manual adjustment. With less, use your parent metric (orders, active accounts) as a proxy and keep a wider buffer.
How big do peaks get?
Planning for the average of the last quarter and being surprised by a routine peak is the most common mistake. Some sourced anchors:
- Cobbai notes that seasonal peaks from holidays and product launches can multiply volume 2 to 5 times.
- LiveAgent's holiday support guide puts Black Friday and Cyber Monday week at 80% to 200% above a normal week for many stores.
- Salesforce's 2024 holiday data reported that shoppers used AI and agent-powered chat for customer service 42% more than in the 2023 holiday season. AI capacity needs planning too, not only human capacity.
Treat these as ranges to sanity-check your own factors, not as numbers to paste in. Your history beats anyone's average. For the peak-season playbook itself, see peak season support without hiring.
Check the forecast against actuals every week
A forecast nobody checks gets worse over time. Every Monday, log last week's forecast and last week's actual per channel, and the error as a percentage.
Two patterns usually appear within a month:
- The baseline drifts. Growth, churn, and product changes move the underlying number. A trailing window corrects itself, but only if you refresh it.
- Event factors are too small on the way up. We anchor on the average, so peaks get under-forecast and quiet weeks over-forecast. Log the actual shape of every launch or campaign so the next factor is based on evidence.
How support software forecasts conversation volume with AI in the loop
Most helpdesk tools forecast the same way this spreadsheet does: a historical baseline, a trend, and seasonality, sometimes with a statistical model doing the decomposition. What changes when an AI agent answers part of the volume is that one forecast becomes two.
- The AI line scales with total incoming volume. It is a cost line: with per-reply pricing, a week at 3x volume costs 3x in AI capacity, but does not need 3x people.
- The human line scales with what the AI cannot or should not handle: exceptions, disputes, anything that needs judgement.
In the worked example, suppose an AI agent fully resolves 45% of launch-week volume. The human line drops from 575 to about 316 tickets. But the remaining tickets are the harder ones, so AHT rises. At 12 minutes (0.2 hours): 316 × 0.2 / 0.7 / 40 = 2.3 FTE, instead of 3.1. Plan with the higher AHT, or your human line will be too optimistic.
How Keloa approaches capacity planning
Keloa's customer service setup is built around the two-line model. The AI agent absorbs routine volume at a per-reply cost you can plan against, so a peak week does not turn into a rushed hire. The unified inbox shows where the AI handed off and why, which is exactly where your human capacity goes.
Our pricing is per AI reply, so your forecast and your bill move in the same units. For teams burned by the per-resolution pricing trap, that is the point: no surprises when volume spikes.
Frequently asked questions
What is the simplest way to forecast support volume? Average your weekly tickets over the last 12 to 13 weeks per channel, multiply by your growth rate, and apply a factor for known events or seasons. It fits on one spreadsheet tab and is good enough for most teams under 20 agents.
How much history do I need? Twelve weeks gives a usable baseline. Supportbench suggests two to three years for reliable seasonality. With less, forecast against orders or active accounts and add a wider buffer while you collect data.
How accurate should a small-team forecast be? There is no reliable public benchmark for small teams. Our rule of thumb: aim to land within about 10% at the monthly level and accept wider misses on individual peak weeks. Track the error weekly and adjust your event factors when a pattern shows up.
Should I forecast in tickets or in hours? Both. Forecast tickets by channel, then convert to hours with AHT and shrinkage for staffing. Volume tells you what is coming. Hours tell you how many people to schedule.
How do I include a launch or campaign? Add it as an event factor on top of the baseline, based on the shape of the last comparable event. Record the actual shape afterwards so the next factor is better.
Does AI capacity need its own forecast? Yes. AI absorbs volume linearly and cheaply; people absorb complexity. Forecast an AI line from total volume and a human line from what remains, using a higher AHT for the human line.
Sources
- Supportbench: How to Forecast Support Ticket Volume and Staffing Needs
- Verint: Call Center Forecasting: Methods, Techniques & Best Practices
- Cobbai: Forecasting in Customer Support: Volume, AHT, and Seasonality Models
- Supportbench: Forecast ticket volume and staffing without fancy tools
- LiveAgent: Ecommerce Holiday Support Guide
- Salesforce: 2024 Holiday Shopping Data