Ecommerce Customer Service Outsourcing vs AI: Costs for a Small Brand

For most small brands under about 300 tickets a month, an AI helpdesk plus the owner is cheapest, because outsourced teams often charge a monthly minimum. Near 2,000 tickets, outsourcing the complex tickets and letting AI take repetitive ones usually beats either alone. The deciding number is your share of repetitive questions, not your total volume.

Those thresholds are rules of thumb built on illustrative prices, so do not take them as quotes. The sums below show how to run the comparison yourself with numbers from your own support inbox and from real vendor proposals. The helpdesk prices come from Gorgias's pricing page in November 2026; the outsourcing figures are assumptions you should replace with quotes.

Follow me on Instagram@sagnikteaches

Start with what your tickets actually are

Total ticket count is a poor guide, because AI and outsourcing are good at different tickets. Spend an hour with your last 100 tickets and sort them into four piles:

Connect on LinkedInSagnik Bhattacharya

1. Look-ups: "where is my order", "what is your returns window". The answer is in a policy or a tracking page.
2. Simple actions: change an address before dispatch, cancel an order, resend an invoice.
3. Judgement calls: damaged goods, wrong item, refund exceptions, upset customers.
4. Pre-sale advice: sizing, compatibility, "which one should I buy".

Say a small brand selling skincare supplies finds 45 look-ups, 15 simple actions, 25 judgement calls and 15 pre-sale questions. AI can plausibly take most of the first pile, some of the second, and a minority of the fourth. The third pile stays with people, however you staff it. If 25 percent of your tickets are judgement calls, no tool changes the fact that you need a person for a quarter of them.

Subscribe on YouTube@sagnikteaches

The cost per ticket, at 300 tickets a month

Here is a small store comparison, in USD, using Gorgias's list prices for the AI side. These figures are illustrative and the assumptions are stated, so you can swap in your own.

Note on the AI line: Gorgias's page lists 30 automated interactions with Basic and $1.50 for each one above the allowance you have bought. If only those 30 came at $0.90, the month would be nearer $280 ($77 for Basic with its 30 included, plus 135 at $1.50). Treat $199 as the lower end and ask the vendor which rate applies. Add the owner's time on the escalated 135 tickets (about 11 hours, or $340 at $30) to compare fairly: the AI option is about $539 to $620 all-in against $750 for the owner alone or $600 plus review time for an outsourcer. The difference is modest at this size, and the quality risk is where the decision lies.

The cost per ticket, at 2,000 tickets a month

The same allowance caveat applies here. Pro lists 190 included automated interactions; if the other 910 were billed at $1.50, the AI line would be about $1,840 ($471 for Pro with its 190 included, plus $1,365), and the hybrid would rise to about $4,090, above the all-outsourced placeholder. Buying an AI allowance that matches your expected volume is the difference between those two answers, so get the rate for your volume in writing.

Two observations. First, the AI cost per ticket stays flat while the human share dominates the total, so the cheapest path to a lower bill is raising the automated share, not negotiating the outsourcer's rate. Second, outsourcing only the leftover tickets can cost about the same as outsourcing everything, because the leftovers are the harder and slower tickets, and an outsourcer will price them higher. Ask for a quote for a "hard tickets only" scope before assuming the saving.

Quality and escalation: the part a price table hides

An outsourced agent represents your brand to customers they have never met, using your policy document. An AI agent does the same with your knowledge base. Both fail in predictable ways.

• Outsourced: inconsistent tone between shifts, refunds that exceed your rules, slow responses outside their working hours, and agents who cannot see your stock or order data.
• AI: confident wrong answers when the policy is out of date, failure to recognise an angry customer, and answers that quote a policy you changed last month.

The fix for both is the same: a written escalation rule. An illustrative one: "Anything involving a damaged item, a chargeback, a legal threat, a medical or safety claim, or a customer who has written three times goes to the owner within four working hours." Put it in the contract with an outsourcer and in the automation rules for the AI. See AI ticket triage for how to tag and route by category.

Contract and data points for outsourcing

Read these before signing and negotiate the ones that matter for a small brand.

1. Minimum monthly fee and term. A twelve-month term with a minimum that you cannot reduce in a slow month is a real cost.
2. Service levels. First-response targets by channel, with what happens if they miss them.
3. Refund authority. A cap per ticket, and who approves above it.
4. Tools and data. The team should work in your helpdesk, so the conversation history is yours. If they use their own system, ask for a full export on exit.
5. Customer data handling. What they can see (names, addresses, order contents), who can access it, and the data-protection terms if you have customers in the EU.
6. Training and changes. How fast they adopt a policy change, and whether a change costs extra.
7. Notice period. Thirty days is far easier to live with than ninety.

Contract and data points for an AI helpdesk

The same discipline applies, with different items. Check the billing definition (Gorgias, for instance, bills as a normal ticket if a human replies within 72 hours of an AI resolution), the overage rate above any included allowance, whether the conversation data is used to train the vendor's models, how to switch the AI off at once, and what the monthly bill does in a busy week. The detailed per-platform pricing sits in what each platform charges per resolved chat.

A hybrid set-up that a small brand can run

For an illustrative online homeware brand with 800 tickets a month, a workable hybrid looks like this:

1. AI answers look-ups and takes simple actions, with a visible "talk to a person" option on every reply.
2. Judgement and pre-sale tickets arrive in one queue, tagged by type.
3. The owner handles the most sensitive tags (damage, safety, legal) the same day.
4. A part-time outsourced agent or virtual assistant handles the remaining judgement tickets in set hours, using macros the owner wrote.
5. Every Friday the owner reads ten escalated conversations and updates the knowledge base with anything the AI got wrong.

That Friday review is what makes the hybrid improve. Without it, the AI's share stays flat and the human share keeps its cost.

A sample brief that makes outsourcing quotes comparable

Send the same brief to three providers so the prices line up. An assistant can help you draft it, but the facts must come from your own data:

We are a small online brand selling [product type]. Monthly tickets:
about [number], by channel: email [x]%, chat [y]%, social [z]%.
Mix (from our last 100 tickets): look-ups [a]%, simple actions [b]%,
judgement [c]%, pre-sale [d]%.
Needs: working hours [hours], first response within [time], tone guide attached.
Tools: your agents work inside our helpdesk; we keep all data.
Please quote: monthly minimum, per-ticket price at [300] and [2,000] tickets,
onboarding fee, notice period, refund authority limits, and training fee for policy changes.

Strip any real customer names before you share sample tickets. The responses will show who understands your business and who sends a generic price list.

Sorting your tickets with an assistant first

Doing the four-pile sort by hand on 100 tickets takes about an hour. If you have thousands, an assistant can do the first pass, as long as you strip names, emails and addresses before pasting anything. Use a plain spreadsheet with one ticket subject and first line per row.

Classify each customer message into exactly one category:
LOOKUP (answer is in a policy or tracking page), ACTION (a simple change to an order),
JUDGEMENT (damage, wrong item, refund exception, complaint), PRESALE (advice before buying).
Return a table: row number, category, one-line reason.
If you are unsure, use JUDGEMENT.
Messages:
[paste 40 rows, personal details removed]

An illustrative result: out of 40 rows, 17 LOOKUP, 6 ACTION, 11 JUDGEMENT and 6 PRESALE. That is 58 percent automatable, with the "if unsure use JUDGEMENT" rule pushing doubtful cases towards a person. Read ten of the model's labels yourself. If you disagree with more than two, tighten the category definitions and re-run. The proportions, not the individual labels, are what feed the cost model above.

The in-house alternative nobody prices

The real comparison for a growing brand is sometimes a third option: a part-time employee. Suppose an illustrative online gift shop considers hiring someone for 20 hours a week at $18 an hour. That is $360 a week, about $1,560 a month before payroll costs, and at 5 minutes a ticket the person could close roughly 240 tickets a week, far more than the shop needs. Spare capacity can go to returns and stock questions, but the salary is fixed in a month with 150 tickets. Compare it with a usage-based AI bill that falls with your volume, and with an outsourcer that charges a minimum. The in-house option wins on knowledge and loyalty; it loses on cover for holidays and sickness. Cost it honestly alongside the other two, including training time at the start.

A seasonal swing, and which option copes

Many small brands live with a peak: Black Friday week, a Mother's Day push, a product launch. An illustrative brand has 300 tickets in a normal month and 1,400 in its peak month. Fixed-minimum outsourcing is wasteful in the quiet eleven months. An in-house hire cannot flex. AI helpdesk billing that moves with volume copes best on the look-ups, though the judgement tickets still need people. A sensible pattern is to keep the AI on all year, line up a temporary outsourced or freelance agent for the peak six weeks with the same macros and rules, and pay the extra only when needed. Ask providers whether they sell short seasonal contracts, and what notice they need.

A quality scorecard you can reuse every month

Whatever you choose, score it. Sample 30 conversations a month, 10 from each of three buckets: handled entirely by AI, handled by the outsourced team, handled by you. Score each from 0 to 2 on four things:

An illustrative month: AI conversations average 6.1 out of 8, outsourced 6.8, owner 7.5. The AI score is dragged down by escalation misses, which you fix with a rule. The outsourcer's is dragged down by correctness on a policy that changed, which you fix with a briefing. The owner's is high but costs the most time. The point of the scorecard is that the fixes become specific.

Before and after: rewriting a macro so both options work better

A saved reply (macro) is the cheapest improvement for every option. Here is an illustrative before and after for a late-delivery message.

BEFORE:
Hi, sorry about the delay. We are looking into it and will get back to you.

AFTER:
Hi [first name], thanks for your patience. Your order [order number] left our
warehouse on [dispatch date] and the carrier's latest status is "[status]".
Delivery is now expected by [date]. If it has not arrived by [date + 2 days],
reply to this message and we will open a claim with the carrier, with no need
for you to chase them. [your name]

The after version answers the three questions the customer has: where is it, when will it come, and what happens if it does not. An outsourced agent can send it unedited, and an AI agent can fill the fields from your order system. A better macro also reduces repeat contacts, which is where much of the cost in any support model hides.

Switching safely: a four-week plan

1. Week 1: sort 100 tickets, write the top 30 answers, and set the escalation rule.
2. Week 2: trial the AI helpdesk on look-ups only, with every answer visible to a person for review, or send the outsourcer a pilot batch of 50 tickets.
3. Week 3: score 30 conversations with the scorecard, correct the knowledge base, and expand scope one category at a time.
4. Week 4: compare cost per ticket and repeat contacts with the previous month, then decide whether to commit, extend the pilot or stop.

Do not sign a long contract before week 4. A pilot that costs a few hundred dollars is far cheaper than a twelve-month term you regret.

A realistic mistake and how it showed up

An illustrative online stationery brand outsources support in a hurry before a busy season. The provider is cheaper per ticket than any alternative, but the knowledge base is two pages. Agents guess. Refunds go out at the wrong rate. Within a month repeat contacts (customers writing again about the same issue within seven days) rise, and the owner spends the hours saved on cleaning up. The lesson is that neither option is a substitute for documentation. Spend your first two weeks writing the answers to your fifty most common questions, and both an AI helpdesk and an outsourced team become cheaper and better.

Start by building that content with the FAQ a chatbot needs before launch, then test AI on look-ups only. If you want help weighing the options against your own ticket mix, that is the kind of problem my AI implementation consultation covers.

Hidden costs on each side of the table

The headline price leaves out items that show up in month two. For outsourcing, count the owner's weekly review time (an hour or two is realistic), the cost of writing and updating training documents, and refunds issued outside policy before you set a limit. For AI, count the hours spent building and correcting the knowledge base, a monthly review of wrong answers, and any extra charge when volumes exceed the plan's allowance. For both, count the cost of a bad experience: a customer who receives a wrong answer about a return window and then disputes the payment costs far more than the ticket price, often including a lost sale and a fee.

A quick sum shows the scale. Say an illustrative brand has 2,000 tickets a month and a one percent rate of costly errors (20 tickets) with an average cost of $45 each in refunds, fees and goodwill. That is $900 a month, more than the AI helpdesk plan itself. Cutting the error rate from one percent to half a percent saves $450 a month, which is why the Friday review of escalated and wrong conversations earns its hour. Set a target error rate, measure it with the scorecard, and treat the trend as part of the cost comparison.

Finally, consider what you will do with the time you win back. A brand that saves ten hours a week on tickets but spends none of it on product, marketing or supplier relationships has swapped one chore for slack. Decide in advance where those hours go, and the saving becomes visible in revenue rather than only in a smaller invoice.

How to decide, in four checks

1. Share of repetitive tickets above 50 percent: start with AI.
2. Share of judgement tickets above 30 percent: plan for human capacity from day one.
3. Volume swings more than 3x between quiet and peak months: prefer usage-based pricing over fixed minimums.
4. Knowledge base not written: write it first, before paying anyone.

Review after 90 days. Compare cost per ticket, repeat contacts within seven days, and the owner's hours. If the numbers have moved the wrong way, change the scope rather than the vendor.

Questions before you outsource or automate support

Is outsourcing ever cheaper than AI?

Yes, when tickets are few and complex, when most of them need judgement such as damage claims, or when your knowledge base is not ready. An AI helpdesk with thin documentation resolves little and still costs money. Outsourcing wins on judgement and coverage hours. AI wins on repetitive questions at volume.

What should a support outsourcing contract include?

Response-time targets with remedies, a named account lead, a clear escalation path back to you, ownership of the conversation data, and a notice period you can live with. Add rules on what they may refund without approval, and a requirement to use your tools so you keep the history if you leave.

Can I start with AI and add outsourcing later?

Yes, and many small brands do. Build the knowledge base and macros first, because both an AI agent and an outsourced team need them. If you later outsource, those documents cut the training time and make the handover cheaper.

How do I compare quality, not only cost?

Sample 50 conversations from each option and score accuracy, tone and whether the issue was really closed. Track repeat contacts within seven days, refund errors and customer satisfaction if you collect it. A cheaper option with a higher repeat-contact rate often costs more once your own time is added.

Further reads

• Your First 30 Days of AI Customer Support for an Online Shop — A launch plan for AI support in the first month.
• How to Automate Returns and Refunds With Clear AI Rules — Which returns decisions rules can safely make.
• Can an AI Chatbot Handle Order Tracking and Returns? — Whether a bot can handle tracking and returns.
• How Much Does It Cost to Add AI to a Shopify Store? — The wider cost of adding AI to a Shopify store.
• Best AI Tools for a Small Ecommerce Store: 7 Worth Paying For — Seven paid tools by job, with plan prices, what Shopify already includes free, what to skip and three monthly budgets.
• Best AI for Ecommerce Operations: Orders, Stock, Returns, Suppliers — Back-office AI for an online store: Sidekick for orders, Gorgias or Tidio Lyro for returns, and an assistant for supplier emails and reorder checks.
• Faire vs Ankorstore: Which Wholesale Marketplace Suits a Small Brand? — A worked fee comparison on one sample wholesale order, plus reach, payment terms and minimums, and the figures to confirm on each vendor page.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Gorgias pricing page and its FAQ (November 2026) for helpdesk plans, included automated interactions, the $0.90/$1.00/$1.50 AI Agent rates and the 72-hour rule; outsourcing figures are illustrative assumptions, not quotes.