Your quality standards and values in customer contact, our work. Maximal reads the order, applies your policy and asks you first above an amount you choose. Every decision comes with its record.
New, built in the Netherlands. No customers yet, no SOC 2 or ISO. A free shadow week instead: nothing reaches your customers.
Most AI agents can refund a customer without asking anyone. Above a limit you set, Maximal has to ask.
Try it, you are the owner. This receipt comes from a real run in our test store. Nothing actually happens.
Return of order #1001
real run · 28 Jul 2026 · test store
Above your limit a person decides. What would you do?
See how this decision endedAbove an amount you choose, a person decides. The amount comes from the order, not from the email. Someone writing about a five euro order changes nothing about the figure the approver sees.
You write your returns, cancellation and shipping rules in plain sentences. Every draft names the rule and the version it applied, so you can see why it did what it did.
You describe once how you speak to customers. Every draft follows that description, and during the shadow week your team reads everything before anything goes out.
A real run from 28 July 2026, not a drawing. An email arrived in Trengo, the agent read order #1001 out of Shopify, and these are its own timings. The shop is a test store, so the products are Shopify sample data. Everything else is what happened.
| Time | Step | What | Context |
|---|---|---|---|
| +00.00s | TICKET | Inbound conversation 969636220 | email, via Trengo |
| +03.85s | FIND | 1 order matches #1001 | Shopify |
| +03.85s | POLICY | returns v3, refunds v4, cancellation v2 | loaded |
| +06.75s | RETRIEVE | #1001 fulfilled, placed 7 days ago | inside the 30-day window |
| +25.90s | PROPOSE | Start a return on #1001 | €885.95, senior approval required |
| +44.07s | ESCALATE | Over the €100 agent limit | confidence 0.90 |
The policy allowed the return and it was sure, at 0.90. The amount is €885.95 and the limit was €100, so it wrote the reply, attached it to the ticket as an internal note, and asked. Being right is not the same as being allowed.
Three more decisions, with the replies it wroteNot promises: how it is built. Each line says where it lives in the code.
During a shadow week nothing reaches your customers. Your team sends every reply.
Where it lives: A reply to a customer needs stage 2. A shadow week runs at stage 0, which sends nothing.
When you connect Shopify, Maximal requests read access only.
Where it lives: read_orders and read_products, nothing else.
It never takes an amount from an email.
Where it lives: The amount compared with your limit comes from the order.
It does not change a delivery address.
Where it lives: That action is switched off in this version, whoever asks.
It does not carry out a refund, a cancellation or a return.
Where it lives: It proposes them. So far not one has been carried out.
During the shadow week your day starts with one line: how much came in, and that nothing went out without you. If you want more, you draw three replies to check.
We time with your team how long a conversation takes today, before anything is connected. That is the baseline, and you keep it whatever you decide.
We connect your helpdesk together and, read-only, your store.
Maximal drafts a reply to every ticket. Your team does not have to use any of them.
At the end you get one page: your number, what it got wrong, and what it would cost a month. If the number is bad, you keep it and we stop.
There is no contact address yet, so shadow weeks run by introduction for now.
How the shadow week worksNo. A shadow week runs at stage 0: Maximal reads, decides and drafts, and sends nothing. Your team works as it always does.
No money. It does take half a day of someone on your team for the baseline, and an hour to connect.
The Trust page says exactly: what we read, what we keep, which parties process it, and what we commit to before your first ticket. It also says what is not settled yet.
You see it in the record, and the page at the end of the week lists what it got wrong. Above your limit a person always decides, and during the shadow week nothing goes out anyway.
Trengo and Chatwoot have been tested on a real account. Zendesk is written but not yet tested live. If you use something else, ask: a good candidate is worth a few days of work.
You see how Maximal reads and decides your tickets, next to what your team actually sent. Then you decide.
There is no contact address yet, so shadow weeks run by introduction for now.
How the shadow week works