
Support
Customer support at scale without losing the human touch
Aug 4, 2026 · 8 min read
Updated Oct 31, 2026
Support scales without losing the human touch when the AI takes the repeat questions from your own material, a person keeps every case that needs judgment, and the handover carries the whole conversation, so the customer does not have to explain again. Watch resolution — the share of issues actually solved — and treat deflection as a side number.
At Unistaff, pre-sale questions on web chat and email are Live through the AI inbound sales manager. The AI customer support agent, which drafts answers from your docs for a person to send, is On request; no help-desk connector ships today.
Which support questions an AI can take
A question whose answer already sits in a policy, a help page or an old reply is one an AI can answer from your material. Questions that need live data — an order, an account, a booking — need a connection to the system that holds it, and decisions about money or exceptions stay with a person.
| Question | What the answer needs | Who handles it |
|---|---|---|
| Hours, location, service area, what is included | Your written material | AI, from the knowledge folder |
| Returns policy, warranty terms, how to prepare for a visit | Your written policy | AI, from your docs |
| Price of a listed service | The approved price list | AI, only from the list; anything else gets “I will check with the team” |
| Where is my order, change my booking | Live order or booking data | A person, or an on-request setup; store and calendar connectors are planned |
| Refunds, exceptions, discounts | A decision | A person |
| Complaints, emergencies, legal, tax or medical questions | Judgment and responsibility | Handed to a person with a summary |
Draft-first or auto-reply
There are two ways to put AI on a support desk. In draft-first, the AI writes the answer and a person reads and sends it; speed comes from not typing, and every reply still has an owner. In auto-reply, the AI sends on its own inside rules. The Unistaff support role runs draft-first: a person sends each answer. The Live sales manager starts the same way — as a Trainee every reply waits for the owner — and sends on its own only after you promote it.
The handover is the product
A bad handover drops a confused customer on a person who has to ask everything again. A good one arrives with the whole conversation, the reason for the handover and what is still missing.
- The trigger is explicit: the word HUMAN (also HUMAIN, ЛЮДИНА, ЧЕЛОВЕК), a complaint, an emergency such as gas or water near electrics, a legal, tax or medical question, or a customer ready to pay.
- The handover lands in the cabinet with a summary, and an email copy goes to the handover contact agreed for the pilot.
- The conversation stays attached, so the person sees what the customer asked and what was already answered.
- A booking is recorded as a request for a person to confirm; the AI never says “you are booked” on its own.
Measure what the customer feels
Handle time shows how fast a reply went out and says nothing about whether the problem was solved. Deflection counts conversations that never reached a person; resolution counts issues that were actually solved. A desk can raise deflection by making it hard to reach a human, so read it next to resolution, repeat contacts on the same issue, and how many handovers were solved on the first human touch. The full definitions are in deflection vs. resolution: how to measure an AI support agent.
On the Unistaff cabinet the weekly review gives you three conversations to read — one you rated down, one that was handed over and one picked at random — next to the results against your targets. Reading the handed-over one each week shows whether the rules send the right cases to a person.
Where to start
- List the twenty questions your desk answered most last month, with the reply you would want sent.
- Put the answers into one place — policies, prices, hours, what is included — and remove the contradictions.
- Write the never rules: refunds, discounts and promises the AI must leave to a person.
- Decide who takes handovers and how fast they reply.
- Start on pre-sale questions, which are Live on web chat and email, and scope the support desk on request.
Questions about AI customer support at scale
Let the AI answer or draft the repeat questions from your own material, hand every complaint, emergency, refund and exception to a person, and pass the whole conversation with a summary at the handover so the customer explains once. Measure resolution next to deflection.
The Unistaff support role is draft-first: it writes the answer from your docs and a person sends it. The Live sales manager starts as a Trainee whose replies wait for the owner, and sends on its own only after you promote it.
Those answers need live data from your store or account system. Store connectors are planned, and order-status drafts are built on request; until then a person answers questions that depend on live data.
On the word HUMAN, a complaint, an emergency, a legal, tax or medical question, or a customer ready to pay. The handover lands in the cabinet with a summary and an email copy goes to the agreed handover contact.
Resolution: the share of issues actually solved. Read it next to deflection, repeat contacts on the same issue and the share of handovers solved on the first human touch, because deflection alone can rise while customers get less help.
Pre-sale questions run on the free 30-day pilot of the AI inbound sales manager, then one flat monthly fee per AI employee agreed on the pilot review. The support desk is built on request and quoted after a scoping call.
- AI customer support agent: answers drafted from your docsOn request
- AI support assistant: what it drafts, what a person keepsGuide
- Deflection vs. resolution: how to measure AI supportBlog
- AI employees for e-commerce: pre-sale Live, order drafts on requestIndustry
- Integrations: every channel with its statusIntegrations
- AI employee vs. chatbot: nine differencesCompare