Support cost per order: the e-commerce math

Most stores can quote their ad cost per order to the cent and their support cost per order not at all. That asymmetry is expensive: support is usually the second or third largest operating cost that scales with volume, and what isn't measured per unit quietly grows per unit. This guide gives you the formula, honest benchmark ranges, and the two levers that actually move the number.

Why per order — not per ticket or per customer

In e-commerce, support demand is generated by orders. Every shipment can produce a delivery question, every delivery a return, every return a refund thread. That makes the order the natural unit:

• Per ticket hides growth. Cost per ticket can look flat while ticket volume doubles — the budget still exploded.
• Per customer misleads. Repeat buyers generate contacts per order, not per account; a loyal customer base makes per-customer figures drift.
• Per order composes. Support cost per order sits next to acquisition cost, shipping cost and margin in the same unit economics table. You can finally see whether a product line or market is profitable after its support burden.

The formula, fully loaded

Support cost per order = total monthly support cost ÷ orders shipped that month.

The discipline is in "total." Include:

• People. Loaded salaries (with taxes and benefits) of everyone doing support work — including the founder's hours, prorated. Two hours a day of a founder's time is a real cost even if payroll doesn't show it.
• Tools. Helpdesk seats, chat software, phone lines, and AI fees.
• Outsourcing. Any agency or contractor invoices.
• Training and QA time. Prorated if someone spends part of their week on it.

A worked example. A store shipping 10,000 orders a month runs one full-time agent at $4,200 loaded, a founder spending a quarter of their time on escalations (0.25 × $6,000 = $1,500), and $600 of tooling. Total: $6,300 a month — $0.63 per order.

Healthy figures for direct-to-consumer stores usually land between $0.30 and $1.50 per order, with high-AOV and complex-product categories at the top of the range. The absolute value matters less than the trend: this number should fall as you grow, and in unmanaged support operations it does the opposite.

The two ratios inside the number

Support cost per order decomposes cleanly:

Cost per order = contact rate × cost per contact.

• Contact rate — conversations per 100 orders. For most stores it sits between 8 and 20, dominated by delivery and returns questions.
• Cost per contact — total support cost divided by conversations handled. A human contact typically costs $3–6: an agent at $4,200 a month handling 800–1,000 conversations lands at $4–5 each, before tooling.

The decomposition tells you where your problem lives. In the example above: $0.63 = 12 contacts per 100 orders × $5.25 per contact. Every improvement belongs to one of the two ratios — which keeps the strategy conversation honest.

Lever 1: cut the contact rate at the source

The cheapest conversation is the one that never starts. Each point of contact rate at 10,000 orders is 100 conversations — roughly $400–500 a month at human cost:

• Proactive delivery notifications. Most "where is my order?" contacts happen because the customer wasn't told first. Shipment and delay notifications remove a large slice of the biggest category.
• A self-service order tracking page linked from every email.
• A returns portal that answers "how do I return this?" by doing it, not describing it.
• Product page fixes. If sizing questions are 10% of contacts, the highest-ROI support tool is a better size chart.
• A post-purchase email answering the top three questions for that product category before they're asked.

Run a monthly top-drivers review: the three biggest contact reasons, each with one root-cause fix. Stores that do this consistently push contact rates down 2–4 points in a quarter.

Lever 2: lower the cost per contact with an AI front line

Root-cause work has a floor — some questions will always arrive. That's where the blended-cost math changes the picture.

An AI agent answering from your documentation and order data handles the routine share of contacts at a fixed, known cost. On ReplyPool's plans the arithmetic is explicit: $99 a month buys a pool of 1,000 AI answers ($0.099 each at full use), $219 buys 2,500 ($0.088), $499 buys 7,500 ($0.067). A top-up of 1,000 more answers is a flat $49 — $0.049 each — bought in one click, only when you choose.

Blend it. Take 1,200 monthly contacts at $5.25 each: $6,300. Let an AI front line resolve 60% at roughly $0.09 per answer and route the rest to your team:

• 720 AI-resolved contacts ≈ $65
• 480 human contacts × $5.25 = $2,520
• New total ≈ $2,585, a blended $2.15 per contact

Cost per order drops from $0.63 to about $0.26 — with people freed for the judgment calls that actually deserve them. The human cost won't fall to zero and shouldn't: the point is that the routine majority stops consuming $5 a piece.

Watch the denominator trap in metered pricing

One caution before you shop for the AI layer: per-resolution pricing puts a floor under your cost per contact. At $1–1.50 per AI resolution, the metered fee is 10–20× the pooled cost above, and it grows linearly with your volume forever — the exact property you're trying to remove from the cost structure. A fixed pool amortizes: the same plan price spread over more answers as you grow. When you model vendors, model your peak month, and check what the marginal answer costs at 2× your current volume.

Make the number fall as you grow

Support cost per order is one of the few costs that should benefit from scale, because so much of it can be fixed rather than metered:

• Fixed-price tooling amortizes over rising order counts.
• Seat-free pricing matters: if adding a teammate is free, seasonal helpers don't move the tool bill. ReplyPool includes unlimited teammates on every plan for exactly this reason.
• A quarterly ritual keeps it honest: recompute the formula, decompose into the two ratios, pick one root-cause fix and one automation expansion.

One caveat to keep the metric from being over-optimized: not every contact is a cost. Pre-sales conversations convert — a sizing question answered in seconds is revenue, not burden — and a generous human touch on a high-value order pays for itself in reviews and reorders. The goal isn't the lowest possible support cost per order; it's knowing the number, splitting it into its two ratios, and spending deliberately where the contact creates value while automating where it doesn't.

Do that four times a year and support stops being a cost that "just grows with the business" — it becomes a unit cost with a downward slope you can show your finance spreadsheet, and your investors.