Forecasting ticket volume as your store grows

Support volume feels like weather: some Mondays drown you, some weeks go quiet, and nobody can say why. But in e-commerce the feeling is wrong. Support demand is downstream of a number you already forecast — orders — and once you model the connection, next quarter's ticket volume becomes as predictable as next quarter's shipping bill. This guide builds that model in three layers, then shows how to turn the forecast into the right size of AI reply pool.

The engine: tickets follow orders

For a given store, the ratio of support contacts to orders is remarkably stable month to month. Compute yours from three months of history — total inbound conversations divided by orders, expressed per 100 orders — and you have the engine of every forecast:

Forecast tickets = forecast orders × contact rate.

If you ship 8,000 orders in a typical month and your history shows 14 contacts per 100 orders, your baseline is about 1,120 conversations. Your order forecast already exists somewhere — finance, inventory planning, the ad budget model. Support planning starts by borrowing it, not by guessing tickets directly. The contact rate itself moves slowly (product fixes push it down, new complexity pushes it up), which is exactly what makes it useful: the volatility in your ticket volume comes almost entirely from the order side.

Layer 1: the seasonality index

Orders aren't flat across the year, and support is even less flat, because peak-season buyers skew toward first-time customers who ask more. Build a simple index from last year: each month's contact volume divided by the yearly average. A typical retail pattern:

• November–December: 1.6–2.5× the average month, with Black Friday–Cyber Monday week alone running 2–3× normal daily volume.
• January: the returns wave — often 1.3–1.5×, arriving after revenue has already dropped.
• Mid-summer: the trough, 0.7–0.85×.

Multiply the baseline by the month's index and the forecast stops being surprised by things that happen every year. If you don't have last year's support data, borrow the shape of your order curve and add 15–20% amplitude — support seasonality consistently swings harder than order seasonality.

Layer 2: promo waves — plan the wave, not the day

Every promotion produces a support curve with a shape you can sketch in advance:

• Before the sale (2–3 days): pre-purchase questions — stock, sizing, shipping deadlines, "will the discount apply to…".
• Sale days: payment and checkout issues, discount-code mechanics. Noticeable, but not the peak.
• Days 3–7 after: the delivery wave. Every order placed in the spike becomes a potential "where is my order?" as packages move — this is usually the true maximum.
• Weeks 2–4 after: the returns wave, proportional to how impulsive the buying was.

The rule of thumb: a promo that doubles daily orders roughly doubles your contacts over the following ten days — not on the sale day itself. Teams that staff for the sale day and relax afterward get hit precisely when they stopped watching. Put every planned promo into the forecast as a wave spread over two weeks, sized by the expected order uplift.

Layer 3: growth drift

Growth changes the contact rate itself, not just the volume. The predictable drifts:

• New markets add language, customs and delivery-time questions — international orders typically contact 1.5–2× more often than domestic ones in the first months.
• New product categories bring their own question profiles: apparel drags sizing questions in, electronics drags compatibility, anything fragile drags damage claims.
• New sales channels — a marketplace listing, a social shop — add buyers who never saw your FAQ and message through new surfaces.

None of this is a reason to panic; it's a reason to recalibrate. Recompute the contact rate quarterly, segmented by market and category if volume allows. A drifting rate you track is a planning input; a drifting rate you ignore is next quarter's "nobody saw it coming."

The one-tab model

Everything above fits in a twelve-row spreadsheet. For each month ahead: forecast orders × contact rate × seasonality index + promo waves for that month's calendar. Then split the total by the share your AI front line resolves versus what lands on humans — the automatable share you've measured, not aspired to. Update the sheet monthly with actuals: after two or three cycles, forecast error settles under 15%, which is tighter than most stores forecast revenue.

The output isn't just a staffing number. It's the input for the one decision metered billing never lets you make calmly: how much AI capacity to buy, in advance, at a price you know.

Keep the model honest by writing down each month's miss and its reason — a delayed shipment batch, a promo that outperformed, a carrier strike. Most "forecast errors" turn out to be events you'll recognize next time, and after a year the notes column is a better planning document than the numbers.

From forecast to pool size — and the peak-month move

With a fixed-pool model, the forecast maps directly onto a plan. Say your sheet shows AI-resolved volume of roughly 1,900–2,200 answers in a typical month, spiking to about 3,800 in November. The naive read is "size for November" — buy the big plan year-round. The better move uses the calendar:

• Pick the plan that covers your typical month — here, a 2,500-answer pool ($219/month on ReplyPool) holds ten months comfortably.
• Top up deliberately for the named peaks. November's extra ~1,300 answers cost two one-click top-ups of 1,000 at $49 each — $98, decided when the forecast says so.
• Compare: a 7,500-answer plan year-round is $499 × 12 = $5,988. The 2,500 plan plus November top-ups is $219 × 12 + $98 = $2,726. Same coverage where it's needed, less than half the spend — because you bought the peak for the month that has one.

And if the forecast misses low? A hard-capped pool fails gracefully: the AI pauses, conversations route to your team, and the miss becomes a calibration data point instead of a surprise invoice. That's the quiet advantage of forecasting into a capped system — being wrong costs you a busy week, not a budget.

Where ReplyPool fits

ReplyPool gives the forecast something honest to land on: fixed monthly pools of 1,000, 2,500 or 7,500 AI answers at $99, $219 and $499, a usage dashboard that shows the burn against your forecast in real time, and top-ups of 1,000 answers for a flat $49 when the calendar calls for them. Your spreadsheet predicts the demand; the pricing page tells you — in advance, to the dollar — what meeting it will cost.

Start with the three-month lookback this week: one contact rate, one seasonality index, one promo calendar. That's the whole model — and the last quarter you'll plan support by feel.