AI for customer service that drafts the answer, so your agents only have to check it
Half the inbox is the same question. Your team spends the day looking things up in one screen and typing them into another. The lookup and the first draft can be done for them. The decision to send stays with a person.
Start with a free 20-minute call. Paid discovery is $1,000 and takes 1 week. It is credited in full toward the build if you sign within 90 days.
What does a support inbox full of repeat questions cost a team?
Your agents answer "where is my order?" all morning. The customer with the real problem waits until after lunch.
For example, imagine a 60-person Shopify brand using Gorgias. A large share of tickets ask some version of "where is my order?" For each one, an agent opens the order in Shopify, copies the tracking number, checks the carrier site, and types a reply. The answer was sitting in two systems the whole time. Meanwhile, a customer whose package arrived damaged waits behind forty status checks.
That brand is an illustration of a common pattern, not a past engagement. The same shape shows up in clinics answering appointment questions. It shows up in service firms answering "when is my technician coming?"
How does a support build run, step by step?
- 01Discovery and ticket export
- 02Sort the reasons, find the answers
- 03Pick ticket types, approve reply rules
- 04Build and test on past tickets
- 05Agent sign-off, launch, 30-day check
All ten steps, in detail
- Paid discovery kickoff. Raquel meets your support lead and agrees which channels and ticket types are in view.
- Ticket export. A recent sample of tickets comes out of Gorgias (or your help desk). Customer details are limited to what the review needs.
- Contact reason breakdown. Tickets are sorted by why the customer wrote. You see which questions repeat most.
- Answer source map. For each repeat reason, where the true answer lives: the Shopify order, the carrier, the returns app, or your third-party warehouse.
- First ticket types chosen. You and Raquel pick the reasons to automate first. Usually that means a clear answer in a system and low risk if delayed.
- Reply rules approved. Your support lead signs off on written rules for tone, what a draft may promise, and what must always go to a person.
- Build. A contracted technical delivery partner builds inside your help desk to the spec Raquel writes. Raquel reviews it before testing.
- Testing on past tickets. Drafts run against historical tickets, including the awkward ones: split shipments, stalled tracking, address changes after fulfillment.
- Agent acceptance and launch approval. Agents work the drafts on live tickets, your support lead approves, and it goes live.
- Thirty-day review. Handle time and first-response time on the automated ticket types are compared with the before number.
What do you walk away with?
- A contact reason breakdown from your own ticket history
- An answer source map for each repeat question
- Drafted replies for the chosen ticket types, filled from live order or account data, waiting for an agent to approve
- Routing rules that send exceptions to the right person or team, including operations for address changes and damaged goods
- Written reply rules your support lead approved
- Updated macros for the ticket types that stay manual
- A before number and a 30-day review against it
- A short agent guide to the new workflow
What do you get, and what is this not?
What you get
- Agents who check and send instead of look up and type
- Exceptions that reach a person faster because routine tickets no longer bury them
- A clear record of which ticket types were automated and how they performed
What this is not
- No reply reaches a customer without an agent approving it.
- No refunds, credits, or replacements are issued by AI. Anything involving money is decided by your staff.
- Not a help desk migration. The build lives in the tool you already use.
- No promised satisfaction score or cost reduction. The 30-day review shows what changed, measured honestly.
What does your team need to bring?
- A support lead who owns the decisions and approves the reply rules
- Admin access to your help desk and a staff account in your store or order system
- Two agents who will test drafts and say plainly what is wrong with them
- Your current macros, refund policy, and shipping policy
Five questions that set the price
- How many contact reasons are in scope for the first build?
- How many systems hold the answers: store, carrier, warehouse, returns app?
- How many channels: email, chat, SMS, social messages?
- How many escalation paths and teams do exceptions need to reach?
- Do replies need to work in more than one language or brand?
What does it cost?
- What moves it
- Number of contact reasons, number of connected systems, channels, and escalation paths
- Payment
- Discovery is paid on its own first. Then half the build fee at signing, and the other half when agents accept the handover
What if it isn't worth it?
Discovery shows you the contact reason breakdown before any build is priced. The 30-day review compares handle time and first-response time with the before number.
The founding client discount does not apply to support builds.
See also: Sales and marketing
Questions people ask before booking
Will the AI answer customers on its own?
No. Every reply is a draft until an agent approves it, and that holds after launch too. You are right to be wary of a bot talking to your customers unsupervised. What changes is the agent's job: reading and sending instead of searching and typing.
What happens when the tracking data is wrong or missing?
The reply rules cover it. Say the carrier shows no movement for longer than your policy allows, or the order record is incomplete. The ticket skips the draft and goes straight to a person. Testing on past tickets proves those rules before launch.
Will this replace our support agents?
It is not designed to. The aim is to take the lookup work off their plate. Then damaged orders, upset customers, and tricky returns get attention sooner. How you use the freed time is a staffing decision that stays with you.
Our help desk already has AI features. Why not just switch them on?
Sometimes that is the right call, and discovery will say so. Built-in features often need your data sources connected, your rules written down, and testing on your own tickets first. Only then are they safe to use. Raquel does not resell help desk software or earn referral fees, so you get whichever option fits.