Sanddo reads the thread, finds order and return data, drafts replies, and only sends automatically in categories you have approved.
4
support modes: draft, approval, auto and human
8
known support categories from the routing layer
0
auto replies without QA, trust score and workspace rules
Inbox
Support queue
Reply draft
Order #NF-48319
Hi Anna, your order is already on its way.
The carrier scanned the parcel this morning and expects delivery tomorrow between 10:00 and 14:00.
Kind regards, Nicolai
92
trust
0
flag
18s
ready
Product
Sanddo is not a loose chatbot. It runs a controlled support pipeline on top of Outlook, order data, RMA, QA, review and audit.
Inbox
The mail processor fetches unread Outlook emails, groups them by sender, locks message IDs against duplicates and retrieves the full conversation thread. If the customer has sent multiple unread emails, they become one case.
Context
The agent gets the thread, attachments, order lookup, return/RMA status, shipping rules, product knowledge and previous human edits. It is instructed to reply in the customer’s language and look up the order from the customer’s email.
Routing
Order status, delivery time and return guidance can run in Auto. Product questions, claims, refunds and angry customers stay in manual approval in the current setup. Unknown cases start as drafts.
Reply + control
Relevance QA removes irrelevant sections, such as payment, accessory or image copy if the customer did not ask. Pre-send QA blocks internal tags, AI mentions, empty replies and risky wording before Outlook is allowed to send.
Review
The dashboard stores the draft, shows routing and QA, and lets the reviewer edit directly or give an internal instruction. The agent can then rewrite the reply with the same thread and tools.
Audit + learning
Sanddo stores created, classified, draft, QA, auto-blocked, sent, edited, approved and rejected events. Approvals, edits and rejections become trust scores per category and examples for the next draft.
Safety model
Category mode is only the first gate. Sending is also stopped by escalation, relevance warnings, low trust score, QA errors, dry-run, or if Auto is not enabled for the workspace.
| Order status | Auto | Tracking data + no policy risk |
| Delivery time | Auto | Known answer, low variation |
| Return guidance | Auto | Standard process without money movement |
| Product question | Approve | Requires product nuance |
| Claim | Approve | May require documentation or replacement |
| Refund | Approve | Money movement is controlled |
| Angry customer | Approve | Legal signals and escalations stay out of Auto |
| Other | Draft | No safe category matched |
Auto readiness
Order status
91
Recommendation
Ready for Auto, but keep dry-run for 12 more cases.
Learning loop
Sanddo records what reviewers approve unchanged, what gets rewritten, and what gets rejected. After at least five reviews, a low trust score can automatically block Auto, and edited replies are used as internal examples for the next draft.
Approvals increase trust. Unchanged replies show that the category is stable.
Edits become examples. The agent sees both the draft and the final human-approved version.
Rejections block Auto. Trust below 75 after enough reviews keeps the category in manual approval.
Pricing
The same safety model on every plan. Prices are in DKK per month.
For small shops that want to start with drafts and auto replies.
1.999 kr
/mo
For growing shops — the step between small and large.
3.999 kr
/mo
For growing webshops with review learning and multiple markets.
5.999 kr
/mo
For large organizations with high volume and special requirements.
Custom solution
ROI calculator
The calculation uses the same model as the reference: 1,000 tickets per employee, 80% automation, and a plan selected from total ticket volume.
Enter support numbers
We assume approx. 1,000 tickets per employee / month
Monthly support cost per employee
Automation
80%
tickets automated
Team
1 / 2
save 1 position
Estimates are indicative. The actual effect depends on category mix, data quality and how much Auto you choose to enable.
Pilot note
“We liked being able to choose exactly how much control we had over which emails were sent automatically.”
Camilla Nielsen
Customer service team lead, Nordic Flame
82%
less support time on average
5→1
customer service in peak season
4
control levels per email category
Book demo
We review your categories, current software, ticket volume and where Auto can realistically be enabled without losing control.
30-minute personal walkthrough
Get a concrete automation map for your first week
Choose a time
Selected time:
Demo booked.
We use your ticket categories as the first automation map.
No credit card. No automatic sending before you actively enable it.
