Shipment status
Where the parcel is, when it arrives, what to do about a delay. From your delivery rules, not from the model's general knowledge.
Industry
An agent that knows your returns policy better than a new hire
In a shop, most tickets are four questions in different wordings. The answers already exist: in the terms, in the delivery description, in the complaints policy. What is missing is somebody to answer at two in the morning.
Four areas that take off the most
Returns and complaints
Deadline, conditions, return cost, how to file. The return itself is accepted by your team, not the agent.
Sizing and variants
Questions about the size chart and differences between variants land exactly where the buying decision wavers.
Changing order details
Delivery address, pickup point, invoice data. The customer hands the request to the team with the „Talk to a human” button, and the team sees the whole conversation and makes the change.
A sample conversation
A demonstration conversation. The answers come from a returns policy and delivery rules loaded into the knowledge base.
Customer
I ordered a dress in M but it is too big. Can I exchange it for an S?
Agent
Yes. A size exchange is possible within 30 days of delivery and is free with us, return shipping included. Source: Terms, point 5.
Customer
Wants the exchange: clicks „Talk to a human” below the message field, leaves an e-mail address and clicks „Send to the team”.
Operator
Sees the whole conversation and the customer's latest orders, found by the address from the handoff. Makes the exchange in the shop, and the label goes out to the customer.
conditions from the termshandoff at the customer's requestexchange by the team
Shop and helpdesk connected with keys from your own admin
Ada reads orders from your shop and opens tickets in your helpdesk. Connecting means keys generated on your side and pasted into the AIvatary panel. No passwords, no installation, no middleman.
- Shopify and Zendesk, and also WooCommerce and Gorgias
- In Shopify and Zendesk you create an app in your own admin and paste its client ID and secret, and WooCommerce and Gorgias connect with API keys from their settings. Every connection shows in the panel with its probe result: alive or not, in one word, with no vendor messages.
- The customer checks an order in a form, not in the chat
- The order number and the e-mail address from the order go into a separate widget form. That data never reaches the language model: Ada gets only the facts about this one order, for the length of the conversation, and answers „where is my parcel” and „when will it arrive” without guessing.
- What Ada reads from an order
- Number, order and fulfilment status, carrier, tracking link, date placed, number of items and the total. It never reads the delivery address, payment details or the basket lines.
- What Ada does not do
- It does not change orders. With the „Talk to a human” button in the widget, the customer hands your team a request to cancel an order, change the address or return an item. Your team sees the whole conversation and makes the change. Switching a connection off is one click; deleting it removes the keys from our side.
Repeat questions on one list
Ada groups customer questions from every channel into topics and shows which ones it handles alone, which end in a handoff and which have no answer in your knowledge base yet.
Topics instead of single questions
„Where is my parcel”, „when does my order arrive” and „shipment status” are one topic. The list shows the number of questions, the trend over the last week and whether an article already covers the topic.
An article draft in one click
From customer questions and your own sources Ada composes a draft answer. The draft enters the editorial cycle like any other article: a person publishes it, never an automaton.
Publishing closes the gap
When publishing an article, the editor closes the knowledge gaps it addresses, from the panel's suggestions. From then on Ada answers that topic with a citation, and the list shows how the number of questions and handoffs in the topic changed over the last week.
The team gets context, not a raw chat
Harder cases still reach people. The difference is what they arrive with.
- A summary at handoff
- When a conversation goes to a person, a note stands next to the thread: what the customer wanted, what was settled, what is left to do. The agent starts from the point, not from reading twenty messages.
- The customer's orders next to the conversation
- By the contact's e-mail address the panel shows the latest orders from the shop beside the conversation, with no window switching. The data is fetched live and is not stored on our side.
- A suggested reply with citations
- One click and the agent gets a draft reply from your knowledge base, with the sources marked. They send it their own way: edited, shortened or discarded.
- A ticket in Zendesk
- A handoff opens a ticket in your helpdesk with the note and a link to the conversation. One handoff is one ticket, also after a retry.
Measures you count yourselves
Instead of percentages from other people's deployments: three numbers from your own conversations, in the panel, from day one.
Share of conversations without a handoff
How many conversations Ada closed without the team. Counted from conversations that really happened, not from a survey.
Conversation rating
The customer rates answers with a thumb and the whole conversation at the end. The panel shows the average and the number of ratings, so one bad rating cannot pose as a trend.
Time saved
Conversations without a handoff times your minutes per case. An estimate on your own assumptions, which you set yourselves, and it is labelled as such.
How many conversations this takes off your team
The starting values describe a mid sized shop. Replace them with yours.
Questions about status, policy, setup and pricing, where the answer already exists somewhere in your company.
Total employer cost divided by the hours actually worked. The default of 13 EUR is converted from our Polish figure, so replace it with your own.
0
conversations a month taken by the agent
0
team hours returned to other work
0
returned to the budget every month on these assumptions
An estimate on your figures, not an offer and not a forecast of results. The calculation uses only the fields above: it knows nothing about your contracts, your seasonality or how your team works.
Why you can put this on a shop
- Answers only from your content
- The citation gate acts before the customer sees the first character. An answer without coverage is rejected whole: the customer is told plainly that there is no confirmed answer, and the question goes to the gap register.
- Operations only with human consent
- A data change becomes a confirmation card. There is no path by which the agent performs it alone.
- A trace on every turn
- The audit log records what was retrieved and what was decided, without the conversation content.
- Your data separated from everyone else's
- Organisation scope is a condition of the knowledge base query, not a filter over its result.
Start with the policy you already have
Load your returns and delivery rules, then ask the agent last week's questions from your own mailbox.