Why do AI chatbots always miss the point?
Explaining hallucination from a technical angle, and how RAG fixes it. 8-minute read.
Open any AI chatbot on the market and ask "What's your return process?" — it'll make up an answer that sounds reasonable, butthat's not your company's actual process, just the statistically most likely response.
Our AI doesn't work that way. We feed your website, product manuals, SOPs, support logs, and past FAQs into a custom knowledge base,training an AI that only knows your company。
It doesn't know the universe — it only knows your business.
Your website, product copy, internal SOPs, and past support conversations all go into a vector knowledge base. Ask him "What's the difference between Plan A and Plan B?" and he can answerbecause he's read it. Not AI improvising — answers grounded in the real data you provided.
Not bland, generic AI pleasantries. We design custom tone, vocabulary, stance, and even taboos for every client. Responses read like you wrote them by hand, not from a template。
Every successful response gets reviewed and added to the knowledge base. Human-corrected conversations flow back as new training data, After a year, he'll understand your company better than a new hire. Hire a colleague who grows, not a tool that hasn't improved in three years.
LINE, Telegram, on-site chat, email — one AI working everywhere, with unified knowledge and unified personality. Customers never get contradictory answers from different channels.
When unsure, he says "I'll need to bring in a human for this," rather than inventing a plausible-sounding but wrong answer— this is the biggest problem with most AI chatbots, and where brand trust starts to collapse.
We come from advertising and games,
and we know one sentence can destroy a brand.
So our AI isn't an all-purpose assistant —
it's a professional colleague focused only on your brand。
No competitor stands at all four of these positions at once.
Xiao Ai will pop up in the bottom right — click to chat. The scenarios below cover the core capabilities of the AI colleagues we build for clients.
Click any question below to try itNot every shop needs a website. Hanging an AI colleague on a LINE business account is the same as hanging one on a website.
And even if you have a website, if all your customers find you on LINE — booking, inquiries, appointments, support — LINE is your real storefront。
Has a website, but visitors almost all come in through LINE — bike inquiries, time slot booking, payment, confirmation, all handled in the LINE chat. The website just shows the brand to people who haven't added LINE; LINE is the real storefront.
/admin/ shows orders, updates statuses, and surfaces KB gaps.
Production running, 24 hours, no sleep — ebike.saomin.tw ↗
有什麼車?租一天多少?" — unlike the demo above, this isa LINE actually running a real business, not a sandbox.We're not arguing which is better — we're saying we and generic AI chatbot SaaS are two different models. You'll see which side you belong on.
| Dimension | Generic AI chatbot SaaS | Satsuma AI Colleague |
|---|---|---|
| Business model | Software subscription | Creative custom service |
| Suited for | Generic, standardized companies | Mid-sized brands with personality |
| Setup | You upload FAQs yourself | We interview + build the knowledge base |
| AI source | Off-the-shelf general LLM | Dedicated vector knowledge base |
| When wrong | Mostly makes things up | Says "let me get a human" |
| Answer traceability | ❌ | ✅ Source citations |
| Long-term growth | Stuck at day-one quality | Understands your company more over time |
| Marketing integration | You wire it up yourself | We're a marketing agency to begin with |
| Starting price | From NT$3,000/month | From NT$3,000/month Custom from NT$50,000 setup + NT$5,000/month |
| Buying mindset | "Installing a tool" | "Hiring a colleague" |
Every tier is customized — the differences are in **interview depth**, **number of channels**, and **how deeply we stay involved**.
Custom (games, subscription, specialized industries) by quote · Multilingual / Avatar / API SDK add-ons negotiated separately
8 perspectives and case studies on AI colleagues, AI chatbots, RAG technology, and integrated marketing.
Explaining hallucination from a technical angle, and how RAG fixes it. 8-minute read.
Vectors, chunking, retrieval, reranking — from principles to engineering pitfalls. 10-minute read.
Three anchor questions to help you figure out which side to pick. 10-minute read.
Backend data from Satsuma's own Xiao Ai after one month live, plus 3 unexpected learnings. 9-minute read.
Five structural problems with the SaaS model, and why they can't be fixed. 11-minute read.
Three-tier plans + hidden costs + ROI, all laid out. 12-minute read.
5 real cases + technical breakdowns. 10-minute read.
The natural next step for integrated marketing, and why the last mile of the funnel needs to be custom. 9-minute read.
What you need isn't a tool,
it's an AI colleague that belongs only to your company.