Automating whole shops with two Telegram bots and GPT-4

I kept noticing the same thing at local shops: the owner's phone buzzing all day with the same questions. "Is this in stock?" "What's the price?" "Are you open?" Every message pulls them away from actually running the place. So on a few weekends I built them an assistant — a dual Telegram bot setup. One bot faces customers, one faces the vendor. Behind them, n8n workflows and a Node.js/Express backend do the coordinating. The brain is GPT-4 for intent detection: figure out what the customer actually wants, even when they ask sideways. I added Hinglish sentiment analysis too, because real customers here don't type in clean English — they mix Hindi and English, and tone matters. And inventory syncs live with Google Sheets, so the shopkeeper keeps using the tool they already trust instead of learning new software. Across 25+ shops it resolved about 70% of queries on its own — roughly three staff-hours saved a day per shop — and cut customer resolution time by 60%. The thing I learned building it: automation lands when it meets people where they already are. Telegram they already use, Sheets they already trust, Hinglish they already speak. The tech was the easy part. Fitting into their day was the real work.