How to Compare Different Customer Support Options for Online Stores Without Guessing

Compare different customer support options for online stores on cost per resolved ticket, not on sticker price. Every vendor page leads with the monthly fee and every hiring calculation leads with the hourly wage, and neither number tells you what a support lane actually costs you once training, tooling, rework, and escalations are counted. The lanes that look cheap at the top of the funnel are frequently the most expensive at the bottom.

Support volume tracks sales almost mechanically in an e-commerce store. That makes support a variable cost that grows with success, which is why the comparison matters more the faster you grow. A number you can run on your own data beats a comparison table someone else built.

The Short Version Before You Compare Anything

The only comparison that survives contact with a real November is cost per resolved ticket, measured on your own case mix, because every other number in the category is either a sticker price or a vendor's definition of success. Everything below explains how to build that number and what it reveals about each lane.

Each has a legitimate best case. The mistake is choosing between them on a single axis, because they differ on several at once and those differences only show up under load.

A quick word on vendor claims, because they pollute this comparison more than anything else. "Automation rate" is the number most AI support vendors publish, and it is routinely inflated, because a vendor can count any ticket that did not receive a human reply as automated regardless of whether the customer got an answer. If you want to see how that number gets manufactured, the breakdown of what automation rate really measures is worth reading before you take any vendor's headline figure at face value.

What Cost Per Resolved Ticket Actually Measures

Cost per resolved ticket is total support spend divided by tickets that reached a real resolution, which means a lane that closes everything and satisfies nobody scores badly. There is no way to make that number flattering without either reducing real cost or improving real outcomes, which is exactly why it is the right comparator.

The numerator includes more than wages or subscription fees. Add agent hours, the tooling stack, onboarding and training time, the management attention tickets consume, and the cost of tickets that had to be handled twice because the first pass resolved nothing. That last category is where cheap lanes quietly destroy their own advantage.

The denominator is where most comparisons cheat. A chatbot that deflects a question by linking a help-center article has not resolved anything if the customer comes back the next day. A lane's cost per resolved ticket should count reopens and repeats against it, not against the customer's patience.

Two lanes can post the same headline cost and differ enormously on this metric. That is the whole argument for building the number yourself rather than trusting a published one.

Why Sticker Price Misleads

A subscription fee is a fixed monthly number that does not move with volume. A wage is an hourly number that moves with volume but only after a training period you pay for and can lose when the person leaves. A managed service is usually priced against case volume, which makes it the only lane whose cost scales in step with the work it absorbs. Comparing a fixed fee to a variable wage is comparing two different shapes, not two different prices.

How Each Lane Actually Runs

Each lane has a mechanism, and the mechanism is what determines where it breaks. Understanding the moving parts saves you from buying a lane that fails for a reason you could have predicted.

Handling It Yourself

You are the cheapest agent at low volume because your marginal hour is free until it is not. The failure mode is that support competes directly with merchandising, buying, and marketing for the same hours, and the queue wins that competition by being loud. This lane is rational while support is under a few hours a day and stops being rational the moment it displaces revenue work.

Hiring

A hire adds capacity and judgment, which is what you need for complex returns, disputes, and anything involving an angry customer. The cost structure is front-loaded: recruiting, on-boarding, and weeks before the person is productive, and the whole investment walks out the door when they leave. Hiring works best when your ticket mix is genuinely complex. It is an expensive way to answer order-status questions.

Self-Serve Help Centers and Chatbots

A help center scales a single well-written answer to unlimited readers at near-zero marginal cost, which makes it the best value in the category for predictable questions. Its limit is that it only deflects; it never resolves the exceptions, and the exceptions are where your cost concentrates. A cheap chatbot plugin fails on a related axis, which is knowing when to stop.

Managed AI Support

This lane answers routine tickets, things like order status, returns, refunds, delivery changes, order edits, and discount codes, and routes everything else to a person. It differs from a chatbot in that the answering is bounded: answers come from approved sources, and the system transfers to a human when it is out of its depth rather than improvising.

What the honest versions of this lane look like mechanically: the teammate operates as a user inside the helpdesk you already run, with no migration and no new platform to manage, and it pulls from approved policies and data through read-only, least-privilege tools. The comparison question is not whether such a system can answer a question. It is what it does at the edge of its knowledge, and whether you can read every conversation it had. For a fuller breakdown of the per-ticket arithmetic, including what the cost side looks like at different volumes, the analysis of what AI support really costs per ticket is a useful reference.

A Comparison Process You Can Run This Month

The process below produces a number you own. Step order matters: you cannot cost a lane you have not sized, and you cannot judge a lane until you have counted its failures.

1. Export 90 days of tickets and tag each one by type, then split resolved-versus-escalated. You now know your case mix, which is the input every lane comparison needs.
2. Estimate the manual minutes each type consumes today, including time spent searching for context. Multiply by the fully loaded hourly cost of whoever handles it. That is your baseline cost per resolved ticket.
3. For a hire, add recruiting, onboarding, and tooling costs over a realistic tenure. Confirm the resulting number against your actual ticket mix rather than a generic productivity figure.
4. For a self-serve tool, estimate how many of your top ticket types it can actually close, then re-price only those. Leave the rest at baseline.
5. For a managed service, ask what happens to a ticket outside its scope, and count only the types it genuinely closes. Escalation is a feature of the comparison, not an asterisk.
6. Re-run the arithmetic with a 40% volume spike and see which lanes still hold. If a lane only works at your current volume, it is a seasonal fix, not a permanent one.

The lane with the lowest cost per resolved ticket at your normal volume is the starting candidate. The lane that also holds up under a spike is the one worth buying.

What to Ask Every Vendor

Ask which ticket types the system is authorized to answer, what happens when a case falls outside that scope, and how you audit conversations after the fact. Ask whether the price is fixed or scales with case volume, and what the stated upper limit of the service is. A vendor who cannot answer the escalation question clearly has not thought about the tickets that cost you the most.

Mistakes That Make Comparisons Useless

The pattern underneath nearly every bad support decision is the same: applying a pre-AI cost model to a category that has changed. An hourly cost multiplied by headcount made sense when every ticket needed a person. It is now an incomplete formula for the routine slice of your volume.

Counting a deflection as a resolution is the most expensive error, because it flatters the cheapest lane and hides the tickets that come back. A customer who reads a help-center article, finds it does not match their order, and writes in anyway has cost you the self-serve lane plus the human lane, and your comparison counted only the first.

Buying on the automation rate alone fails in the same direction. The number is defined by the vendor and measures what they chose to measure, which is why a lane can look twice as efficient on paper and deliver no improvement to the customer. The failure shows up as reopens, not as a dashboard line.

The opposite error is treating AI support as a bolt-on to a tool you already dislike. If your helpdesk is the constraint, replacing it does not fix the ticket mix; it just moves the same unresolvable cases to a new interface. Fix the routing underneath the tool, and any lane you choose performs better.

Finally, ignoring the ceiling of a lane. Many services cap the account size they support, and a lane that stops fitting at higher volumes is one you will have to re-run the whole comparison to replace. Knowing the ceiling before you buy saves you the migration later.

When to Act, and When to Wait

You are running this comparison too early if your support load is under roughly an hour a day and your ticket types are genuinely varied. At that stage, your own attention is the cheapest response, and buying a lane adds a system to manage without removing real cost.

You are running it too late if you are answering the same order-status question for the fifteenth time this week, or if a Friday night queue is still open on Monday morning. Repeated, predictable questions are the signal that you have a routable slice of volume, and that slice is what every lane is priced against. Volume alone is not the trigger; repeatable volume is.

The decision point usually arrives in September or October, before the seasonal spike, because every lane except refunding margin away is closed by mid-November. If you are choosing between building in-house capacity and buying a managed lane, ask whether you want a hiring project or a finished output. A hire is a project with a recruiting phase, a training phase, and a person who may leave. Everything you can buy instead is a project you are paying someone else to run.

The signal that you picked the wrong lane is not a cost overrun. It is a queue that got shorter while your reopens went up. That pattern means tickets stopped reaching a conclusion, which is worse than a ticket that sat unanswered, and it means re-run the comparison rather than tuning the tool.

How We Approach This

Loqum is a managed AI teammate that works inside the helpdesk your store already uses. No migration. No new platform to run. It is trained on your policies, your customer orders, and your catalogue, and it answers the routine tickets: order status, returns, refunds, delivery changes, order edits, and discount codes. Answers come from approved policies and data through read-only, least-privilege tools, and everything outside that scope goes to a person.

The distinction readers should test is whether a system knows when to stop. A chatbot generates a plausible sentence and moves on. Our teammate answers from approved sources and hands the ticket to a human teammate when it is out of its depth. Every conversation it has is a readable, overrideable ticket inside your queue, not a black box behind a vendor dashboard. Each teammate has an accountable AI engineer behind it with a monthly audit, a report, and a capability plan. We build, train, and run it, so you get the output instead of the upkeep.

We are direct about the limits. Judgment stays with human agents. We are not positioned for stores handling more than 4,000 support cases per month, and capability expands over time from simple tickets rather than arriving complete. We accept a limited number of stores and may be full. Our pricing is a fixed monthly case-based fee with no claim of unlimited scale, and the current numbers live on our pricing page rather than in this article, because prices change and articles do not.

If your comparison keeps pointing at routing rather than at tooling, getting ticket routing right is the fix that makes any lane cheaper. If you want to see where we sit against one specific category of tool, our comparison to a helpdesk-native AI agent is here. Either way, run the numbers on your own case mix first.

Frequently Asked Questions

What are the 12 different types of customers?

The exact count varies by framework, but the categories that matter for an online store cluster around behavior: the first-time buyer, the repeat loyalist, the deal hunter, the returner, the silent browser, the complainer, the advocate, the confused order-changer, the fraud risk, the wholesale or B2B buyer, the gift purchaser, and the customer who only ever contacts you at a spike. What matters for your support comparison is which of these generate predictable, routable tickets and which generate judgment calls. Order-changer and order-status tickets are routable. Disputes and fraud are not.

What are the top 10 best customer service practices?

The practices that survive contact with real volume are few. Answer the question the customer actually asked rather than the one nearest your FAQ. Resolve in one contact where the case allows it. Never make a customer repeat context they already gave. Tell the customer what happens next and when. Measure reopens, not just first-response time. Route by case type instead of by whoever is free. Staff for your spike, not your average. Make every automated conversation readable by a human. And revisit the cost per resolved ticket every quarter.

What is a 5 star customer service?

Five-star service is a resolution that leaves the customer with no reason to write again. That usually means the first response solved the problem completely, the answer matched their actual order rather than a generic policy, and any exception was handled by a person who had full context. Speed matters but is not sufficient: a fast wrong answer earns one star. For stores comparing lanes, five-star service is the standard each lane should be measured against, because the cheapest lane can only be cheap if it resolves rather than deflects.

What are the five customer types?

A common working split is the loyal advocate, the price-sensitive shopper, the impatient transaction-focused buyer, the hesitant researcher who needs reassurance before purchase, and the dissatisfied customer who arrives already frustrated. Each generates a different ticket shape. Advocates rarely contact support at all, price-sensitive buyers generate returns and discount-code questions, transactional buyers want order status and delivery changes, researchers want product detail, and dissatisfied customers generate the complex cases no lane should automate. Counting your own tickets by these types tells you how much of your volume is genuinely routable.

---

The comparison that matters is the one you run on your own case mix. If you want the number for your store rather than the category, book a call and we will build it with you.