Support scales by hiring

The channel that produces the tickets is the channel that is growing

The demand side is documented. On August 18, 2026 the U.S. Census Bureau put second-quarter e-commerce sales, adjusted for seasonal variation, at $340.2 billion — 17.1 percent of all retail sales in the country. The same release put the year-over-year increase at 12.2 percent, against 6.7 percent for total retail sales in the same period.

Read the two numbers together and the direction is unambiguous. The online channel is growing at roughly twice the rate of retail as a whole, which means its share of the whole rises every quarter it does so. Orders placed online are the orders that generate a written question — a tracking number that stopped moving, a size that did not fit, a box that arrived open.

What follows from that is a direction, not a coefficient. The ratio of tickets to orders is not published by any statistical agency, and it varies by category, by price point, and by how well a brand's shipping works. Any article that multiplies revenue by a ticket rate has invented the rate. This one does not need it: the claim is about the shape of the curve, not its slope.

Every other cost line has a rate card

Sort a brand's costs by what the next unit of capacity requires and they fall into two groups.

The first is step-shaped. Storefront software, the data warehouse, the analytics stack, the checkout itself — these are built or licensed once and then serve many times the orders with a change to a plan tier. The cost per order falls because the denominator grows.

The second is proportional. Payment processing takes its percentage of every order. Freight is charged per package. Cost of goods is charged per unit. None of these ever flattens, and none of them needs to, because the capacity behind them is bought at a published price, in any quantity, with no notice. A brand that doubles its orders on a Tuesday does not negotiate for more payment capacity. It sends more transactions.

Support sits outside both groups. It is proportional like the second — demand scales with orders — but there is no rate card behind it. The only way to buy the next increment is to acquire a person, and a person is not a line item that can be ordered. They are recruited, interviewed, trained on a catalog and a policy, put on a calendar, supervised, covered when they are sick, and replaced when they resign.

The labor pool refills; it does not expand

The Bureau of Labor Statistics tracks the occupation directly. Customer service representatives held 2,814,000 jobs in the United States in 2024, at a median wage of $20.59 an hour and $42,830 a year — the Handbook’s projection base, which differs from the survey estimate the Bureau publishes separately. Employment in the occupation is projected to decline 5 percent between 2024 and 2034.

The next line is the one that matters to anyone building a support team. Despite that decline, the Bureau projects about 341,700 openings a year across the decade — and states that all of them are expected to come from the need to replace workers who move to other occupations or leave the workforce entirely.

A shrinking occupation that still turns over a third of a million positions a year is a treadmill described in federal statistics. A brand hiring into it is not filling a role once. It is entering a queue it re-enters on a schedule it does not control, competing for the same candidates as every other employer in the same queue.

The ramp is documented too. The Bureau describes on-the-job training for the occupation as short-term, typically two to four weeks. That is fast for a labor market and slow for a business: weeks between deciding capacity is needed and having it, and the wait restarts with every departure. Against a channel that grew 12 percent year over year, a supply line with weeks of latency and a built-in replacement cycle is the binding constraint, not the wage.

Demand arrives in peaks; people are hired in whole units

The Census Bureau publishes its e-commerce estimate twice, once adjusted for seasonal variation and once not, and the gap between the two is the point. The same quarter reads $340.2 billion adjusted and $329.5 billion unadjusted. Quarter over quarter, the adjusted series moved 3.8 percent while the raw series moved 9.4 percent. The statistical agency separates the two because most of what a brand feels in the raw number is season, not growth.

A support team cannot be seasonally adjusted. It is bought in whole people, on fixed windows, months before the peak it is meant to cover. Staff to the peak and the brand pays through the trough. Staff to the average and the peak is where its customers meet the worst version of the operation — the launch week, the sale, the shipping delay that arrives as four hundred simultaneous messages.

The same arithmetic governs the hours nobody wants. A Saturday afternoon, a holiday, two in the morning for a customer in another timezone: each is a real interval of demand that in-house coverage buys only by paying a person to be awake inside it. The cost of that hour has nothing to do with how many tickets arrive in it.

The curve bends only when the unit of supply changes

Nothing above is fixed by working the labor market harder. A better recruiter shortens the queue; it does not remove it. The line bends only if some part of the volume stops being answered by a headcount.

Razviel's architecture draws that line by the nature of the case, so informational volume resolves at the AI tier by design. Capacity there is software capacity. It has no calendar and no notice period.

Cases that need judgment inside a policy the brand has already written route to a human. That human is an independent service provider under contract with Razviel, working a defined coverage module inside Razviel's own console, with the case already assembled: the order, the history, the policy cell, what was attempted. The brand does not recruit that person, certify them, schedule them, cover their absence, or replace them. The unit it buys is coverage, not a hire.

And every case, wherever it resolves, is designed to leave the same record behind — the same fields whether the AI closed it, a Razviel-certified Human Node closed it, or the brand's own team did. Curated patterns accumulate per brand, so the answer to a case is designed to be material the next similar case reads. That is the only mechanism in the whole model whose cost per ticket has a reason to fall as volume rises.

None of that is a claim to be cheaper. Nothing here has been measured, and this article makes no comparison it cannot support. What the architecture changes is the shape of the line, not a number on it.

The takeaway

Support is the one cost line in a direct-to-consumer brand whose next unit of capacity is a hire. A hire carries a lead time, a training curve, a calendar and a replacement cycle, while the online channel that produces the tickets grows at roughly twice the rate of retail as a whole. For a Shopify Plus brand, the planning question is not what support costs this year. It is what the next increment of capacity requires.

What to check on next year’s plan:

1. What buys the next unit. For each cost line, the next unit of capacity is a plan tier, a published rate, a purchase order or a hire. Support is usually the only hire on the list, and that row carries the lead time.
2. The lead time, end to end. From the decision that capacity is needed to a trained person in the queue: recruiting, interviewing, and the two to four weeks of on-the-job training the Bureau describes. The clock restarts with every departure.
3. The calendar the plan staffs against. A brand feels the unadjusted series — the launch week, the sale, the shipping delay — not the seasonally adjusted one. Staff to the peak and the trough is paid for; staff to the average and the peak meets the thinnest coverage.
4. The hours nobody wants. Weekends, holidays and other time zones are real intervals of demand. In-house coverage buys each one by paying a person to be awake inside it, whatever the number of tickets that arrive.
5. Which volume needs a person at all. Informational cases are settled by a lookup; in Razviel’s design they resolve at the AI tier, on capacity with no calendar and no notice period. The headcount curve bends only by the share of volume that stops being answered by a hire.

If the next increment is a purchase order, the cost grows with the business; if it is a hiring plan, the cost grows with the business and brings an organization with it, permanently.