The problem, as you feel it
Where is my order. Do you ship to Canada. What is your return policy. Can I change my appointment. Is this in stock. Do you offer gift wrapping. How do I reset my password. Someone on your team answers these all day, at the pace a human types, in between the questions that actually need a human.
Response times drift to a day. Customers ask twice. The person doing support is your best product expert, and she spends her time copying the shipping policy into emails.
You may have tried a chatbot. It answered from a script, got confused by any real question, and annoyed people. That was a different technology and it deserved its reputation.
What a grounded support assistant does
The word that matters is grounded. A modern assistant answers only from your actual material: your policies, your product information, your help articles, and, where connected, the customer’s own order or account. It does not answer from general internet knowledge and it does not improvise. When it does not know, it says so and gets a person.
- Answers instantly on every channel. Website chat, email, text, and messaging apps, with the same knowledge and the same voice.
- Looks things up for real. Order status from your store, appointment details from your scheduler, account status from your system, so “where is my order” gets a tracking number, not a policy paragraph.
- Takes actions you allow. Reschedule an appointment, start a return, update an address, resend a receipt, within rules you set.
- Escalates properly. Anything outside its knowledge, anything emotional, anything above a threshold, goes to a person with the whole conversation and the customer’s details already attached.
- Learns from your answers. When a person answers a question the assistant could not, that answer becomes part of what it knows next time, after you approve it.
- Reports honestly. What percentage it resolved, what it escalated, where it was wrong, and which questions customers ask that your material does not cover.
Take a business like this one
The business in this section is a composite of the kind of company I talk to, not a named client. The numbers are the shape of the problem, not a case study.
Picture an online specialty food shop with a small retail counter, four staff, and around 1,400 orders a month. Support arrives by email and Instagram messages, about 900 conversations a month, handled by the owner’s daughter and whoever is at the counter. When I categorise a month of their conversations, roughly 70 percent fall into ten questions: order status, shipping to a particular place, substitutions, allergens, returns, gift notes, wholesale inquiries, delivery windows, out-of-stock timing, and how to store the product.
What I set up for a business shaped like this:
- A knowledge base built from what they already have. Their policies, product pages, the daughter’s saved replies, and a transcript of an hour of her explaining how she answers the hard ones.
- Connections to the store and the shipping account, so the assistant can answer order questions with the actual tracking status and can look up whether an item is in stock and when it is due.
- Web chat, email, and Instagram, all answered by the same assistant in the same voice, with a clear handoff when a human is needed.
- Three allowed actions: add or change a gift note before shipment, resend an order confirmation, and start a return within policy. Everything else is a question for a person.
- A weekly report of resolution rate, escalations by reason, and the unanswered questions, which becomes the list of help articles to write.
In a business shaped like this you should expect the majority of conversations to be resolved instantly with no human involvement, the humans to spend their support time on wholesale inquiries and genuinely unhappy customers, response time on everything to drop from many hours to seconds, and, after a month, the report to show a short list of questions worth answering once properly on the website.
What it costs to run
Model usage for a business at this volume is tens of dollars a month. The chat and email tooling is a small monthly fee, sometimes free at small scale. The setup is the work: building a knowledge base that is complete and correct, connecting your systems, defining the allowed actions and the escalation rules, and testing it against a few hundred real past conversations before it talks to a customer.
What I need from you
Access to your past support conversations, so I can see the real distribution of questions. Your policies, in whatever state they are in. An hour with your best support person explaining how she answers the hard ones. Decisions on what the assistant may do on its own. Then two weeks of reading its transcripts and correcting it, which is how it becomes genuinely good.