Stella for Support
The customer support AI agent built for retail and e-commerce
Cut support costs in half.
Without compromising customer experience.
Stella, our AI agent, rethinks support costs that balloon with headcount. First-line inquiries are resolved on the spot — improving response speed and resolution rates.
In demand with CS teams at leading retail and e-commerce companies
Problems
Is that outsourcing bill really buying you quality?
Note: The same applies if you run your own in-house center. Hiring, training, and shift management weigh on you as fixed costs, while quality still hinges on each operator's experience — the structure is identical.
The cause isn't a negligent vendor or a team that isn't trying hard enough.
A structure where a human handles every single case, one at a time — that is the real problem.
As long as humans are in the loop, unit costs won't fall and quality stays tied to individual people.
Cost Structure
Pay per seat or pay per call —
your cost still scales with inquiry volume
Call center outsourcing is priced in one of two ways. A fixed monthly (per-seat) plan that reserves dedicated agents typically runs ¥300K–500K per seat per month in retail. Pay-per-use plans run roughly ¥300–1,000 per call. Even on fixed plans, exceeding the contracted volume triggers per-call overage fees, and nights and weekends carry surcharges.
Either way, the structure is the same: one agent can only handle so many cases, so more inquiries mean more seats. Whether you pay per seat or per call, total cost keeps growing in proportion to inquiry volume. And most retail and e-commerce inquiries are routine questions whose answers are already sitting in your order data. If rate negotiations can shave off a few percent at best, the only real lever left is changing the structure of support itself. An AI agent handles a case for just a few hundred yen — with no fixed seat costs in the off-season.
Typical market rates for call center outsourcing
Reserves dedicated agents. The going rate in retail
The going rate for inbound support
Calls beyond contract are billed per case; nights and weekends cost extra
Fully usage-based. No seat fees, no after-hours surcharges
Seats × seat rate ≈ Cases × per-call rate
Based on publicly available industry benchmarks. Actual rates vary by scope and specialization.
Cost per case over time (conceptual)
Human support cost
+6.3%/yr
Japan's FY2025 minimum wage increase (largest on record). Government target: ¥1,500/hour by the mid-2030s
AI support cost
×1/10/yr
Annual decline in the cost of equivalent AI capability. GPT-4-level quality now costs ~1/1000th of what it did 3 years ago
Sources: Ministry of Health, Labour and Welfare (FY2025 regional minimum wages); a16z, "LLMflation"; Epoch AI. Curves are a conceptual illustration of the trends.
And this gap will only
keep widening from here.
Rising labor costs are locked in by policy. Japan's minimum wage rose ¥66 (+6.3%) to a national average of ¥1,121 in FY2025 — a record increase for the second year running — and the government has set a target of ¥1,500/hour by the mid-2030s. Annual spring wage negotiations are landing above 5% as well. With a shrinking working-age population intensifying the competition for talent, neither agent wages nor outsourcing rates have anywhere to go but up.
Meanwhile, the cost of equivalent AI capability is falling to roughly 1/10th every year. GPT-4-level answer quality now costs about 1/1000th of what it did three years ago. This is no temporary price war — it is the product of three structural forces working at once: smaller, more efficient models; new generations of chips; and competition among providers, including open models. And the capability you get for the same spend keeps going up.
On one side, increases of around 5% a year are scheduled by policy; on the other, costs keep falling by about 90% a year. Staying with human-only support means holding on to an operation that is guaranteed to get more expensive every year. Moving to an AI agent is an investment whose returns grow the longer it runs — and every year you postpone the decision, that year's savings are lost for good.
Solution
Put AI agents at the center
of your customer support.
What we offer isn't another chatbot. It's an AI agent trained on your product information, policies, and order data — one that understands intent, makes decisions, and completes procedures end to end. It takes over the full job of a seasoned operator, at just a few hundred yen per resolved case. However much inquiry volume grows, you can scale top-quality support without limits.
Understands intent, then answers
"The thing I ordered yesterday still hasn't arrived" — even from a vague request, it identifies the order, grasps the intent, and responds.
Completes the procedure, not just the reply
It doesn't stop at answering — it processes returns, size exchanges, delivery date changes, and cancellations.
Consistent quality, always
A post-sale inquiry rush or a midnight order — zero wait time, and the quality never wavers.
✆Stella
Hello, this is Stella support. Am I right that this is about the cordless vacuum you ordered yesterday?
✆Stella
It's currently out for delivery, arriving around 6 p.m. today. If that doesn't work for you, I can reschedule the delivery right now.
Built for Retail
An AI agent built solely
for retail and e-commerce.
This isn't a general-purpose AI with retail knowledge bolted on. Returns and exchanges, delivery, repair intake, cancellations — it's a purpose-built agent with decision criteria built in, trained and tested against every inquiry type that actually occurs in retail and e-commerce. That's why it doesn't stop at pointing customers in the right direction — it completes the procedure itself.
Returns & Exchanges
It weighs your return policy — window, item condition, receipt — and decides eligibility on the spot. If eligible, it handles everything in one pass: return instructions, issuing the return label, and confirming when the refund will arrive.
Built into this inquiry type
Decision logic
Eligibility decisions based on return window, item condition, and receipt, plus triage of exceptions like sale items and used goods
Procedures it executes
Issuing return labels / processing refunds / arranging replacements
Training & validation
Trained on real return inquiry patterns and tested against expected scenarios
The three foundations behind that purpose-built design.
One Agent, Every Channel
One AI agent,
resolving support at every touchpoint.
This dedicated agent can be deployed anywhere. It draws on all of your product information, order data, return policies, and member profiles to deliver the right answer and handle the process for each individual customer. And web chat is just the start: it extends to LINE and even the phone — covering every channel your customers reach out on, at the same consistent quality.
+ Personalization
On every channel, answers addressed to you — powered by member data.
For signed-in customers, the agent personalizes every answer using order history, delivery status, and loyalty points. "Where's my order?" gets answered on the spot — in chat, on LINE, or over the phone.
Learning Loop
Smarter with every conversation.Learning that never stops running.
Chat, LINE, phone — every interaction becomes training material. Was the issue resolved? Where did the conversation stall? How did the customer rate it? The agent analyzes its own results and automatically generates improvements to its answers and knowledge.
Approved improvements take effect immediately, and the next cycle begins. Because this loop never stops turning, day one is the weakest your agent will ever be — it keeps getting smarter on its own.
The only human task is approving the improvement reports the AI drafts. Quality keeps rising with no added operational workload.
∞
Runs
continuously
01
Respond
The AI handles real inquiries while conversation logs accumulate
02
Self-analyze
The AI reviews resolution rates, sticking points, and customer ratings
03
Generate improvements
Proposed updates to answers and knowledge are compiled into a report
Humans only here
04 Approve
Just review and approve the report. Changes go live instantly
Before / After
Here's how the cost structure changes.
Model case: an e-commerce / direct-sales company handling millions of inquiries per year
* Figures vary depending on your inquiry mix and knowledge-base readiness. We'll run a free simulation to estimate the impact for your business.
What would the numbers look like for your business?
Enter your inquiry volume and current costs to get a rough estimate on the spot.
Onboarding
Start small, without the risk.
STEP 1
Current-state analysis (free)
We analyze your inquiry logs to map what AI can automate — delivery status, returns, order changes — and estimate the cost-reduction impact.
STEP 2
Knowledge integration & build
We connect your FAQ, return policies, cart/order management, and membership systems to build an AI agent dedicated to your business.
STEP 3
Limited pilot run
We start on select channels and time slots to validate answer quality. Human support runs in parallel, so operations never stop.
STEP 4
Full rollout
We expand channels and coverage step by step as the results are confirmed.
FAQ
Answers to the questions we hear most.
Q.What if the AI gives a wrong answer?
Responses are generated strictly from your approved knowledge base, and low-confidence cases or critical procedures are automatically escalated to a human agent. The system is designed to keep the AI from improvising answers that could put your brand at risk.
Q.How is customer data kept secure?
Access to member data is limited to the authenticated user's own information. The scope of data handling is designed around your security policies during onboarding.
Q.Does this make all of our agents unnecessary?
No. We recommend a setup where the AI absorbs routine inquiries so your people can focus on the complex cases and escalations that genuinely require human judgment.
Q.Can it integrate with our existing cart and order management systems?
We support API integration with major e-commerce carts, order management systems (OMS), and CRMs. For in-house systems, we design the integration approach during onboarding. And you decide how far the AI goes — order lookups, return intake, delivery changes — to match your operations.
Q.Will it work for our category (electronics, apparel, food, etc.)?
We build on a foundation shared across retail — returns, delivery, defects, and member support — and layer in your specific product knowledge, policies, and workflows. Category-specific rules (non-returnable food items, manufacturer warranty handling for electronics, and so on) are incorporated at the design stage.
Q.How long does implementation take?
It depends on how organized your knowledge is, but a pilot can start in as little as a few weeks. We begin with a current-state analysis and a proposal.
See What Stella Can Do
See everything Stella can do,
up close.
We'll show you how Stella uses AI to keep support costs down while delivering a customer experience that still feels human — tailored to your products and your team.
Learn MoreAfter you submit the form, our team will get back to you within one business day.