An AI sales assistant on Telegram, the website, and a messaging app asks buyers clarifying questions, picks products by technical specs, finds alternatives, creates a deal in the CRM, and reports order status
S-GAS supplies industrial and specialty gases in a range of purity grades, gas mixtures, and the equipment that goes with them: cylinders, regulators, valves, fittings, gas panels, and cabinets. It sells to both businesses and individuals, who order on the website or through a sales rep.
You can’t pick the right product from a single keyword. You have to account for the gas and its purity, pressure, connection type, regulator material, and cylinder volume, plus equipment compatibility and regional delivery restrictions. On top of that, you need to confirm whether the buyer wants a filled cylinder or an empty one.
Sales reps clarified these details by hand in every conversation, so routine consultations ate up a lot of their time. Meanwhile, buyers were writing in on the website, on Telegram, and in a messaging app, and the answer had to be just as accurate in every channel.
The client needed an assistant that talks like an experienced sales rep: it pins down the specs before sharing any links, looks for an alternative when nothing suitable is in the catalog, and hands tricky cases off to a person on the team.
The assistant understands everyday names: ask for “laughing gas” and it offers nitrous oxide. If the required purity isn’t available, it suggests an alternative and explains the differences, and it picks compatible equipment for the gas.
A team of specialized AI agents inside the assistant answers routine questions about product selection, orders and delivery, gas mixtures, and services. Accounting questions, partnership inquiries, and complex cases go to a human sales rep.
Buyers get answers on Telegram, on the website, and in the messaging app, and in every channel the assistant follows the company’s rules and pulls its terms from internal guidelines.
The assistant creates a deal at the first stage of the CRM pipeline and reports order status by order number.
Here’s a conversation step by step, from the buyer’s first question to the order.
To start, the buyer clicks a button to consent to the processing of their personal data, then describes what they need in their own words — say, a gas cylinder for cutting or a regulator for a specific gas. The assistant identifies the topic and routes the conversation to the right specialist agent.
The assistant asks about the gas and its purity, volume, pressure, application, and connections, and whether the buyer needs a filled cylinder or an empty one. Until the buyer has answered every question, it holds back product links so it doesn’t suggest the wrong option.
The system finds matching products based on what the request means and the specs provided. Every answer is checked before the buyer sees it: products are verified against the catalog, and prices, lead times, and delivery terms come from the company’s internal guidelines.
The buyer gets links to the right products and places an order. If a question is beyond the assistant’s scope, a sales rep takes over with the full conversation history.
The assistant is tested automatically: a suite of 112 scenarios runs after every change. A separate model grades the answers against set criteria, and on 47 scenarios the answers scored 95–100%.
Also used: Google Sheets — request analytics · Store XML feeds — daily catalog export
S-GAS now has an assistant that knows the catalog, talks to customers within the rules the client set, and serves buyers in three channels.
The same approach works for companies whose sales reps ask several clarifying questions before recommending a product, and whose buyers reach out through different channels.
We build solutions like this around your needs: turnkey AI products, AI agents and automation, integrations, and data.
Our servicesTell us about your project — we’ll talk it through and show you how to make this work for your business
We use cookies and Yandex Metrica to improve our website. Privacy policy
