DEV Community: Mia

Chat SDK for E-Commerce Customer Service: A Case Study with MINISO

Mia — Tue, 27 Oct 2026 03:14:38 +0000

"Where is my order?" is probably the most common question in online retail. The answer is rarely a mystery. It's sitting in an order system, one lookup away. Yet in plenty of support queues a person still types it out by hand, in the same inbox as the refund dispute that needs real judgment.

That mismatch is the heart of e-commerce customer messaging. Volume isn't the problem. The problem is that easy and difficult conversations arrive in the same place, get handled by the same people, and run through tools that weren't designed for both.

This case study looks at where retail support conversations break down as a business scales, what a well-built chat architecture does about it, and how MINISO, the global lifestyle retailer, uses Nexconn's messaging products.

The market, and why support has to keep up

Online retail keeps taking share. eMarketer forecasts global retail e-commerce sales of about $6.9 trillion in 2026, roughly one in every five retail dollars. In Southeast Asia, the e-Conomy SEA 2025 report from Google, Temasek and Bain projected $185 billion in e-commerce GMV for 2025 and found that three in five people in the region now shop online.

Support has to follow that geography. A retailer selling into a dozen markets has shoppers asking questions in a dozen time zones, and their expectations come from the best service they've had anywhere. Zendesk's CX Trends 2026 research, which surveyed 6,182 consumers and 5,115 CX professionals across 22 countries, found that 88% of consumers expect faster responses than they did a year ago. In the same research, 85% of CX leaders said a single unresolved issue is enough to lose a customer.

Messaging suits that pressure. A shopper can ask a question in thirty seconds and come back to the answer later, and the retailer ends up with a persistent thread tied to an account instead of a scatter of calls and emails.

"Where is my order?", again and again

WISMO is the industry shorthand, and the sources that measure it agree it's the most common post-purchase contact. Industry benchmarks from providers like Gorgias and Zendesk show that order-status inquiries typically account for 20% to 35% of all post-purchase volume, while peak-season vendor blogs often claim numbers as high as 40% to 60%. Each contact is quick to answer, and nothing about it needs a human. The cost comes from thousands of them arriving in the same queue as conversations that do need judgment.

Context that disappears at every handoff
Zendesk found that 81% of consumers want an agent to pick up where the last conversation left off, and 74% get frustrated when they have to repeat themselves. Retail support makes that hard by design. A question starts with a bot, moves to an agent, turns into a delivery problem, and picks up a supervisor along the way. If every transfer means closing one ticket and opening another, the history gets left behind.

Devices multiply, messages don't follow

Agents work from a desktop, a tablet and a phone, and sometimes share an account across shifts. Customers change phones and reinstall apps. When messages and conversation lists live in a local database on the device, a fresh install starts out empty. It's the least glamorous problem on this list, and it's easy to miss until real users start switching devices.

Transactional messages need real control
Once a chat carries coupons, payment prompts or refund confirmations, free-form text stops being enough. The backend has to be able to check that a message is legitimate, block one that isn't, and adjust content before it reaches the customer. Moderation adds a cost of its own if every message goes through it, so it makes sense to review only the traffic that carries risk.

Supervisors need visibility without getting in the way
Quality assurance, escalation management, and compliance audits require supervisor observability: silently joining live sessions, tagging, pinning, and exporting transcripts on demand. Many brands also run fan groups and flash-sale communities alongside support, and those put thousands of people in one room, which is a different load on the same messaging layer.

What a well-built retail chat system does

The systems that hold up tend to converge on a handful of decisions, and most of them get made before anyone designs a screen.

The first is how to model a conversation. A ticket, with one owner and one status, suits email. Chat works better as a session with members: the customer, an agent, maybe a bot, and a supervisor who can join and leave. A transfer then becomes a change in membership rather than a new record, and the history belongs to the session instead of whoever touched it last.

Automation comes next, placed where the volume is. Bots are good at lookups such as order status, delivery windows and return policies, and people are good at exceptions. What separates a decent setup from a frustrating one is the exit: clear rules for when the bot steps aside, and a handoff that carries the whole conversation along. Zendesk found that 83% of CX leaders in Asia-Pacific see AI as critical to faster resolution, but speed only counts if escalation works.

Then there's proactive messaging. Much of the "where is my order" volume stems from the post-checkout communication gap between payment confirmation and final delivery, so updates at the right milestones go after the volume at its source. ShippyPro, a shipping software provider, reports that proactive notifications cut these contacts by up to 65%. While carrier performance varies, the mechanism is sound.

Anything transactional should stay under server control. Message callbacks let the backend inspect each message of a given type before delivery and approve, block, rewrite or route it. They also give you a cheap way to manage moderation spend: check the message types that carry risk and skip the rest.

State across devices deserves a decision before launch, not after. That means choosing what syncs, how far back an offline device catches up, and what happens to a conversation list across client environments.

Finally, supervisors need a view that doesn't disturb the customer: joining a live session, filtering by status, and exporting transcripts for review.

How Nexconn approaches it

Nexconn's messaging platform has a building block for each of those decisions. Here's how they fit together.

Sessions, handoff and the agent workbench
Nexconn models a customer conversation as a group session. The customer is paired with an agent through it, agents in different roles can be active at the same time, and a merchant or supervisor can join or leave at any point to watch service quality in real time (via silent monitoring). When a conversation moves between agent groups (from sales to after-sales, say), the incoming agent can read the full history. Agents can also send targeted messages inside a session that only chosen participants see.

The workbench covers the rest of a support team's day. One agent account can be shared by several staff members across shifts, with notifications when an account is active elsewhere. Conversations can be tagged, pinned to the cloud, and organized into custom operational queues. Important notices (such as payments or delivery updates) are clearly flagged, while AI bots triage routine lookups and escalate to human agents when edge cases arise.

The shopper's side of the chat
In the app, customers get the message types a shopping chat needs: text, images, voice, short video, location, and emoji, plus inline system notices. Custom message types cover retail-specific cases such as satisfaction ratings or coupon cards. Read receipts show whether a message has been seen, and the client can search conversations by keyword. Status prompts (welcome greetings, queue progress, and transaction alerts) keep shoppers informed while they wait.

Control over transactional messages
For anything involving sensitive data or financial transactions, synchronous message callbacks put your backend in the loop. Your server receives messages of chosen types before delivery and returns instructions: let it through, block it with a custom reason, or alter the content and metadata on the fly.

Callbacks also work alongside content moderation in flexible pipelines. Teams can choose to moderate first and callback afterward, or execute callbacks first and moderate later—allowing businesses to bypass moderation on trusted system alerts and optimize review spend.

Multi-device sync and operational resilience
To eliminate context loss when agents switch between desktop, tablet, and mobile, Nexconn provides cross-device state synchronization paired with server-side offline message compensation. When users or agents sign in on new devices, missed messages and events replay automatically within a configurable catch-up window. Conversation lists restore cleanly across web and mobile clients, while message revocations sync instantly to online sessions and apply upon next sign-in for offline devices.

For community-driven retail, group channels are built for scale. Apps can deploy large customer communities supporting thousands of members per group, complete with role permissions, broadcast announcements, read receipts, and targeted @mentions.

MINISO: messaging at retail scale

MINISO is a lifestyle retailer known for affordable, IP-driven products, listed in New York (NYSE: MNSO) and Hong Kong (HKEX: 9896). Scale is the point here. The company reported record revenue of RMB 21.44 billion (about US$3.07 billion) for 2025, up 26.2%, and ended the June 2026 quarter with 8,674 stores worldwide after entering Switzerland, its 113th country or region.

At that size, even routine questions add up, and they arrive around the clock.

MINISO implemented Nexconn's chat and messaging products for communication between shoppers and customer service, along with a notification solution that keeps shoppers updated on order status while a delivery is on its way.

The two pieces cover both directions of the conversation:

The messaging layer handles exchanges that need a person, in a persistent thread that keeps its history.
The status notifications bridge the post-checkout communication gap between payment and delivery, answering "where is my order" before anyone has to ask.
While MINISO's deployment focuses on high-volume customer support and automated order status notifications, the broader retail capabilities covered above—from agent workbenches and transactional callbacks to community channels—are part of Nexconn's modular architecture for global retail.

If you're building something similar
Whether you build or buy, a few decisions are much cheaper to make early:

Pick the session model before the UI: Retrofitting handoff onto a ticket-shaped data model is painful.
Choose your multi-device behavior on purpose: Test mobile conversation lists early to ensure sync logic fits your use case.
Put transactional message types behind server-side callbacks from day one: Adding verification after coupons or orders are live is a much harder conversation.
Meter your moderation: Run it where risk exists and skip trusted internal system traffic.
Let bots take the lookups: Write down clear human escalation rules before writing bot logic.

Nexconn Startup Growth Pack: Free Chat + Cloud Discounts

Mia — Mon, 19 Oct 2026 03:23:35 +0000

Here's a number worth putting on a sticky note: 1.9 years. That was the median gap between a seed round and a Series A for US startups in Q4 2025, according to Carta's data on 9,843 rounds. Carta's data suggests companies should plan for an 18-to-24-month gap between rounds, which is another way of saying the next check is rarely as close as it looks.

When funding is that far out, runway is the one lever that sits entirely in your hands. You can't change the market, but you can change what you pay for the things every product needs. Two of those costs show up in almost every app with real users: the messaging layer, and the cloud bill underneath everything.

We've already tackled the first. Our free chat plan gives teams the full Pro feature set at no cost for up to 10,000 monthly active users. Today we're adding the second line item and putting both into one program, the Nexconn Startup Growth Pack:

**Free chat infrastructure. **The complete Chat Pro feature set, free for up to 10,000 MAU/mo.

Cloud discounts. Add a payment method to get 15% off AWS or 30% off Alibaba Cloud services.

Faster integration. Free data migration from other chat vendors, a built-in multi-channel push gateway, and an open-source AI Skill that writes most of the integration code for you.

The rest of this post covers the funding backdrop, where early-stage infrastructure money tends to leak, what's inside the pack, how it compares with the usual free tiers, and what it's worth in months of runway. Every assumption is spelled out, so you can swap in your own numbers.

The runway squeeze, by the numbers

Start with the global picture. Carta's data shows the time between rounds stretched for years before it began to ease, and even after the easing, the median seed-to-Series A gap was still 1.9 years at the end of 2025. Capital has also been gathering in fewer places.

That concentration is even sharper in emerging markets. Southeast Asia tells a sharper version of the same story. DealStreetAsia and Kickstart Ventures put H1 2026 funding at $7.25 billion across 217 deals, which reads like a recovery until you look closer. Deal volume fell to its lowest level since at least 2018, and a single $4.5 billion Series C did most of the lifting. Without that one round, regional funding fell 21% from the previous half-year to $2.75 billion. At a September summit, the CEO of Vertex Holdings described the stage after early rounds as essentially hollow, with promising companies struggling to find follow-on capital locally.

MENA looks similar. MAGNiTT counted $1.35 billion across 214 deals in H1 2026. The ten largest deals absorbed 58% of all capital, international participation fell 48%, and regional investors supplied roughly 81% of what was deployed.

None of this means money has disappeared. It means money is flowing to fewer, larger bets, and the typical early-stage team has to stretch what it raised for longer. Founders in that spot tend to do the same unglamorous thing: go through the fixed costs line by line and cut whatever the product doesn't need.

Where early-stage infrastructure money actually goes

The free tier that ends when you succeed
Most chat SDKs advertise a free tier, and most are built for a demo. The free allowance typically sits between 100 and 1,000 monthly active users, with concurrent-connection caps as low as 10. Push notifications, which bring people back to your app, usually require a separate push service or a paid add-on. At least one vendor's support documentation says its free plan is for testing and development only, and another requires applicants to its startup program to have fewer than five team members and under $10,000 in monthly revenue.

The trouble isn't that these limits are unreasonable from the vendor's side. It's that they arrive at the worst possible moment. Your first few thousand users are the signal you've been working toward, and that's exactly when the invoice appears. Published pricing varies widely, but a team growing from 100 to 10,000 MAU over its first year can spend roughly $6,600 to $16,000 on chat alone, before any engineering time.

Building it yourself is a long detour
The alternative is to build. Our build-versus-buy analysis puts a basic in-house chat system at three to six months of work, and a production-grade one at twelve months or more. Counting maintenance, that usually means two to four senior engineers for six to twelve months. The math is hard to justify for a team whose real product is something else. Delivery guarantees, weak-network recovery, per-manufacturer Android push handling and moderation all look small on a whiteboard and turn out to be projects of their own.

The cloud bill grows quietly
Cloud is the other fixed cost that sneaks up. Flexera's 2026 State of the Cloud survey found that 85% of respondents name managing cloud spend as a top challenge, and it estimated that wasted spend rose to 29%, the first increase in five years. The survey covers organizations of every kind, not only startups, but the pattern holds at small scale: cloud bills are easy to start and surprisingly hard to shrink.

Put the three together and a seed-stage team can be paying for a chat vendor, a push service and a cloud bill before it has found product-market fit.

What's in the Startup Growth Pack

Free chat infrastructure
The chat portion is the same plan we announced earlier, not a trimmed-down version of it. It's the full Pro feature set, free for up to 10,000 MAU, with no credit card required. That includes unlimited messages across direct, group, community and open channels, with community channels supporting up to 10,000 members. Read receipts, typing indicators, mentions, unread badges and multi-device sync are standard. A complete webhook system comes with the plan, along with SSO/SAML, fine-grained permissions and audit logs.

Multi-channel mobile push is built in. You'll still register your own APNs and FCM credentials, as you would on any platform, but the multi-channel push gateway is included natively, so there's no third-party push provider subscription to buy.

Each app also gets peak concurrent connections equal to 5% of MAU (500 at 10,000 MAU), 10 GB of storage, 10 GB of upload capacity and 180 days of message history. On Nexconn's standard pricing page, the equivalent Pro plan at 10,000 MAU is $860 a month, so this isn't a discount on a lesser product. It's the full-price plan at $0 for a defined window.

Cloud discounts for the infrastructure underneath
The new piece is the cloud discount. Link your AWS or Alibaba Cloud billing through Nexconn and add a payment method to receive 15% off AWS or 30% off Alibaba Cloud services, whichever matches where your backend already runs. The free chat plan itself still needs no credit card. The cloud discount is a separate, optional layer. Adding a payment method within 30 days makes you eligible for the discount.

Tools that shorten the path to launch
Two more pieces reduce integration time. If you're moving from Sendbird, CometChat, Stream, PubNub, Twilio Conversations or Agora Chat, end-to-end data migration support is included at no charge. And the open-source Nexconn Chat Integration Skill works with Cursor, Claude Code, OpenAI Codex, Windsurf, GitHub Copilot and any tool that supports the SKILL.md standard. Describe what you want, and it generates the client SDK code, a prebuilt chat interface and the backend token endpoint. It reads the live documentation through llms.txt, so the code follows the current API.

How it compares
Here's where the free tiers start across the category, based on each vendor's published plans as of September 2026:

Every platform needs your own APNs and FCM credentials. In this table, "Included" means a built-in multi-channel push gateway, so there's no separate push provider subscription to buy.

A larger ratio on a larger base is worth more than a generous-sounding ratio on a tiny one. A tier that advertises unlimited concurrency on 1,000 MAU can never support more than 1,000 concurrent users. Five percent of 10,000 gives you 500 real connections to work with.

The gap shows up in dollars once you start to grow. This is what a year of chat costs on published plan pricing, with $50 a month assumed for third-party push where a vendor doesn't bundle it:

The ranges reflect plan tiers, with the low end covering entry-level plans that leave out Pro-level capabilities such as webhooks, multi-device sync and advanced moderation. PubNub is left out because it meters transactions rather than users, so its cost depends too much on how your app behaves to put in a single cell.

For a deeper look at how each vendor bills, see our chat SDK pricing breakdown and the free tier feature comparison.

What it's worth: a runway model

Let's put numbers on this. The model below uses three illustrative teams, and the assumptions are deliberately conservative:

Chat savings: $550 a month, the cheapest comparable paid tier at 10,000 MAU in the table above. We've left out the roughly $50 a month you'd spend elsewhere on push, and the engineering time of a build.
Cloud savings: 15% of a monthly AWS bill, or 30% if the same bill runs on Alibaba Cloud.

A few honest observations. The savings are real: the pre-seed team above keeps $10,200 a year on AWS, or $13,800 on Alibaba Cloud.

The runway math also understates the engineering time you get back. Skipping a three-to-six-month build means those months can be reinvested into core product development, which is usually where the next round gets decided.

Why build on Nexconn

A free plan is only worth taking if it holds up when you grow into it, so here's what sits underneath.

Infrastructure built for uneven networks. Our SD-CAN acceleration network spans more than 3,000 nodes across 233 countries and territories and is designed for sub-120ms end-to-end delivery, with packet recovery and offline sync built in. Our guides to Southeast Asia and the Middle East explain why this matters in practice.

Features that usually sit behind a higher tier. Community channels with sub-channels and roles, open channels with message priority for live rooms, friend and group management, webhooks and audit logs are all part of the same plan. Teams tend to discover they need these a few months after launch, which is a bad time to find out they're locked behind an upgrade.

A path that doesn't require switching vendors. When you add voice and video, there's the Call API. When your community grows to the point where moderation becomes a compliance question, there's real-time AI moderation, priced by usage rather than by MAU tier.

How to get started

Sign up and create a new Chat App in the Nexconn Console.
Integrate the Chat SDK, by hand or with the AI Skill.
Send at least one successful message before December 31, 2026. This simple step locks in your 10,000 MAU free plan. Even if your app isn't live yet, we highly recommend sending a quick test message just to secure your quota today. Claim your spot now so this free infrastructure is ready and waiting whenever your next project needs it.
Add a payment method to apply the cloud discount.
Steps one to three are all you need for the free chat plan, and step four is optional.