Table of Contents
- Key takeaways
- Who is the customer in this ecommerce chatbot case study?
- What problem did the ecommerce chatbot solve?
- How did the retailer set up its ecommerce chatbots?
- Chatbots on the product pages
- One chatbot per language store
- A daily review of chat histories
- A route to a person
- What results can an ecommerce chatbot like this deliver?
- What does a five-chatbot setup cost on SiteGPT?
- How do you decide whether this setup fits your store?
- How to repeat this ecommerce chatbot setup on your store
- Frequently asked questions
- About the case study
- Results and revenue
- Setup and cost
- Languages
- Sources
Created by
Do not index
Created time
Mar 10, 2026 10:09 AM
This ecommerce chatbot case study follows a lighting and electrical retailer that runs five SiteGPT chatbots, one per language store, to answer technical pre-sale questions on its product pages instead of by phone.
TL;DR
- The customer is a group of online stores selling lighting and electrical components in five languages. Its shoppers ask compatibility and specification questions before they buy.
- Those questions arrived by phone and email, often when the answer was already on the product page.
- The group runs five SiteGPT chatbots, one per language store. Each is trained on that store's own product content and sits on the product pages.
- The team reads chat histories every day and fixes wrong answers at the source. Unusual questions go to a person.
- Five chatbots is more than the 3 that Scale includes, so a setup like this sits on SiteGPT Enterprise, which has custom pricing.
Key takeaways
Question | Answer in this case study |
Who is the customer? | An online lighting and electrical components group with five language stores |
What problem did the chatbot solve? | Repeat technical pre-sale questions that tied up the phone and email queue |
How is it set up? | Five SiteGPT chatbots, one per store, each trained on that store's product pages |
How is it run? | A daily review of chat histories, with fixes made in the source content |
Which plan fits five chatbots? | Enterprise, at a custom price. Scale tops out at 3 chatbots. |
Who is the customer in this ecommerce chatbot case study?
The customer in this ecommerce chatbot case study is a group of online stores that sells lighting and electrical components. The range runs from household lighting to fixtures and components used in construction and renovation work.
The group sells in five language markets. Each store has its own language and, in part, its own product mix. A fitting that is standard in one market can differ in another.
This story is one of several in the SiteGPT chatbot case studies collection. For a single-store retail story, see the AI chatbot for retail case study.
What problem did the ecommerce chatbot solve?
The ecommerce chatbot solved a queue of repeat pre-sale questions. Lighting and electrical parts are not impulse buys. Shoppers want to know whether a part fits their wiring, meets the local standard, and how it differs from the next model.
Those questions came in by phone and email in five languages. Many of the answers already sat on the product pages and spec sheets. The team still had to answer each one by hand.
Three things made the load heavy:
- Technical products. A product title and a price do not answer "will this work with my fitting?"
- Five languages. Each repeat question came in again in every market.
- Pre-sale timing. A shopper who cannot get an answer on the product page either calls or leaves.
That last point matters for revenue. Baymard Institute averages 50 studies to a 70.22% online cart abandonment rate. Among the top reasons it lists are extra costs (40%), slow delivery (20%), and a returns policy shoppers did not like (13%). Each of those is a question a chatbot can answer while the shopper is still on the page.
How did the retailer set up its ecommerce chatbots?
The retailer set up five SiteGPT chatbots, one for each language store. Each chatbot is trained on that store's own product content and speaks that store's language.
Chatbots on the product pages
The team placed the chatbots on product pages, where the questions start. A shopper looking at a fitting can ask about that fitting without leaving the page. Typical questions:
- What is this product for?
- Is this compatible with my setup?
- What is the difference between these two options?
The answers come from the store's own product pages and documentation. SiteGPT trains on website URLs, sitemaps, files, and raw text. A store can also connect Google Drive, Dropbox, OneDrive, Box, Notion, GitHub, or YouTube.
One chatbot per language store
The group chose one chatbot per store over one chatbot for all five. Each store sells a partly different range under partly different standards. A separate chatbot keeps a Spanish-market answer from drawing on an English-market spec sheet.
Each chatbot also has its own name and persona, matched to the tone of its store. SiteGPT supports 95+ languages, so a store with a single translated catalog could run one chatbot instead.
A daily review of chat histories
The team reads chat histories every day. The goal is to find technical errors and gaps in the training content, then fix them in the source pages. The next answer on that topic then comes out right in every conversation.
SiteGPT's Growth plan and above add a knowledge gaps view and a weekly email digest that lists the questions the chatbot could not answer. Those two features cut the time a daily review takes.
A route to a person
Unusual questions still go to a person. SiteGPT's human support works on every plan. By default, the AI escalates when the visitor asks for a person, when the situation is urgent, or when the visitor is frustrated after an answer that did not help. The team can replace those default rules with its own in plain language.
What results can an ecommerce chatbot like this deliver?
An ecommerce chatbot like this one delivers results in two places: fewer contacts on the channel it relieves, and more shoppers who finish a purchase after getting an answer. This retailer moved its repeat compatibility questions to the product page. That leaves its team more phone time for complex installations and larger accounts.
Revenue is harder to pin on a chatbot. Seasons, pricing, and marketing all move it at the same time. Measure shoppers who used the chatbot against shoppers who did not, over the same weeks.
Metric | What it shows | How to read it |
Contact volume on the relieved channel | Whether phone and email load fell | Compare the same weeks before and after launch |
Resolution rate | Share of chats that end without a person | Track it per store and per topic |
Handoff rate and reasons | Which questions the bot cannot answer | Each reason is a page to write or fix |
Conversion for chat users | Whether answers lead to orders | Compare chat users with non-users in the same period |
For published revenue and conversion figures from named retailers, such as Lowe's and Walmart, see the ecommerce chatbot examples roundup.
What does a five-chatbot setup cost on SiteGPT?
A five-chatbot setup costs a custom price on SiteGPT, because only the Enterprise plan includes more than 3 chatbots. The self-serve plans cap the chatbot count as follows.
Plan | Chatbots | Messages a month | Price billed monthly | Price billed yearly |
1 | 4k GPT-4.1-mini or 400 GPT-4.1 | $59 a month | $39 a month ($468 a year) | |
2 | 10k GPT-4.1-mini or 1,000 GPT-4.1 | $129 a month | $79 a month ($948 a year) | |
3 | 40k GPT-4.1-mini or 4,000 GPT-4.1 | $429 a month | $259 a month ($3,108 a year) | |
Custom | Custom | Custom | Custom |
For smaller stores, the per-chatbot math is short. On Scale billed yearly, 3 chatbots cost $1,036 each a year ($3,108 divided by 3). On Growth billed yearly, 2 chatbots cost $474 each a year ($948 divided by 2). A store with two or three language versions fits a self-serve plan. Every self-serve plan starts with a 7-day free trial, and a credit card is required.
SiteGPT also has limits a store should know about. It has no Shopify app listed on its integrations page, so it installs on a store by embed script. It answers from content, so order status and returns that live in a store backend need a separate helpdesk app. The AI chatbot for Shopify guide covers that pairing.
How do you decide whether this setup fits your store?
You decide by three questions: how different your language stores are, how many chatbots you need, and where your repeat questions get their answers.
If you need | Pick | Why |
Answers to technical pre-sale questions in one language | One chatbot on Starter | One store, one content set, 4k GPT-4.1-mini messages a month |
Separate answers for 2 or 3 language stores with different ranges | One chatbot per store on Growth or Scale | Each chatbot only reads its own store's pages |
Five or more stores, as in this case study | Scale includes 3 chatbots at most | |
One translated catalog sold in many languages | A single chatbot | SiteGPT answers in 95+ languages from one content set |
Order tracking and returns from store data | A native store helpdesk next to SiteGPT | SiteGPT answers from content, not from order records |
For more on running one assistant across markets, see the guide to multilingual AI chatbots.
How to repeat this ecommerce chatbot setup on your store
To repeat this setup on your own store, follow these five steps. For a store that also has to decide which jobs need order data, the ecommerce chatbot setup guide covers the full eight steps.
- Add your store content. Enter your store URL or sitemap, or upload spec sheets and manuals. Use one chatbot per store if your ranges differ.
- Place the chatbot on product pages. Install the embed script on the product templates, where pre-sale questions start.
- Test with real questions. Pull 20-30 questions from your phone and email logs and check each answer against the spec sheet.
- Turn on human support. Set the escalation buttons and rules so unusual or urgent questions reach a person.
- Review chat histories on a schedule. Fix each wrong answer in the source page, then retest it.
The chatbot best practices guide covers testing and review routines in more depth.
Frequently asked questions
About the case study
What does this ecommerce chatbot case study show?
It shows a lighting and electrical retailer running five SiteGPT chatbots, one per language store, to answer technical pre-sale questions on product pages. The team reviews chat histories daily and routes unusual questions to a person.
Why did the retailer use five chatbots instead of one?
Each language store sells a partly different range under partly different electrical standards. A separate chatbot per store answers only from that store's pages.
Results and revenue
Do ecommerce chatbots actually increase revenue?
They can, when unanswered pre-sale questions are a reason shoppers leave. Baymard Institute puts average cart abandonment at 70.22%. Measure conversion for chat users against non-users over the same period before crediting the chatbot.
What should you measure after launching an ecommerce chatbot?
Measure contact volume on the channel the chatbot relieves, resolution rate, handoff rate with reasons, and conversion for shoppers who used the chatbot.
Setup and cost
Which SiteGPT plan covers five chatbots?
Enterprise, with custom pricing. Starter includes 1 chatbot, Growth 2, and Scale 3.
Do you need a developer to add the chatbot to a store?
No. SiteGPT installs with an embed script and trains from your store URL, sitemap, or files. Order data from a store backend needs a separate helpdesk app.
Can the chatbot hand a shopper to a person?
Yes. Human support works on every SiteGPT plan. The AI escalates when the visitor asks for a person, when the situation is urgent, or when the visitor is frustrated, unless you write your own rules.
Languages
Can one ecommerce chatbot answer in several languages?
Yes. SiteGPT supports 95+ languages. Run one chatbot per store when the stores sell different ranges, and one chatbot when every store sells the same translated catalog.
Ready to see how a content-trained chatbot handles your own product questions? Start a SiteGPT free trial.
Sources
Checked on 16 October 2026.
Written by
