AI agents for SaaS back offices

The requests keep coming. Dockett handles them.

Agents pick up refunds, plan changes and invoice chases from email, Slack and your helpdesk, do them in your own tools, and ask a person before anything risky.

Book a demoWatch it work

148,000 requests closed for 36 SaaS teams. Illustrative

  • Lumen & Co
  • PAYLOOM
  • northgrid
  • Tessellate
  • ORBIT/HQ

Placeholder marks · fictional companies

The problem

Requests don't arrive in one place. They arrive everywhere.

A pile of unsorted request cards on a desk in front of the idle sorting machine.

Unsorted · 47 requests · Monday 09:12

Illustrative

  1. 01

    It lands in four inboxes.

    Support email, the #billing channel, Zendesk and a founder's DMs. Nobody sees all of it.

  2. 02

    The same 20 tasks, every week.

    Refunds, seat changes, missing invoices, W-9s. Each takes eight minutes and four tabs.

  3. 03

    Nobody owns the in-between.

    Support thinks finance has it. Finance thinks it's a product bug. The customer waits two days.

  4. 04

    Risky ones slip through.

    A €1,240 refund looks like an €84 one at 6pm on a Friday.

How it works

Pick up. Read. Do. Check.

  1. 01

    Pick up

    Requests drop in from every channel.

    Email, Slack and your helpdesk feed one queue. Dockett picks up each request the moment it lands, with the thread, the customer and the account attached.

  2. 02

    Read

    Each card is read, field by field.

    The agent works out what is being asked, who is asking, and which record it touches, then writes it down where your team can see it.

  3. 03

    Do

    The work happens in your own tools.

    A refund is issued in Stripe, a seat change lands in HubSpot, the reply goes out in Zendesk. When it's done, it's stamped.

  4. 04

    Check

    Anything risky waits for a person.

    Over a limit, outside policy or unusual for this customer: the card stops at the gate with a red tag until someone on your team approves it.

Try it

Send a request through the machine.

Pick one. Watch it get read, done and stamped, or stopped at the gate for you.

Agent logidle
  1. Choose a request to start a run.

Sample data · illustrative run

The product

One queue. Every decision written down.

Your team sees what came in, what the agent did and what is waiting for them. Nothing happens in a tool that isn't in the log.

Results

Counted in tickets closed, not tokens used.

Illustrative figures for a fictional product

Requests handled / month
12,400Across support, finance and people ops at a 180-person SaaS company.
Hours saved / month
310 hOf copy-paste work between the helpdesk, billing and the CRM.
Median time to resolve
4 minDown from 19 hours, for requests that need no approval.

Integrations and security

Works in the tools you already pay for.

Dockett doesn't add a new place to work. It does the work where it already happens, with the permissions you give it.

Tool names show integrations, not customers

  • Gmail

    Reads and replies to shared inboxes

  • Slack

    Picks up requests from any channel

  • Zendesk

    Works tickets, writes internal notes

  • Stripe

    Refunds, plan changes, invoices

  • HubSpot

    Keeps deals and contacts in sync

  • Notion

    Follows your policies and playbooks

  • Approvals you define

    Limits by amount, customer and action. Anything over the line waits for a named person.

  • Audit log of every step

    What was read, what was done, in which tool, and who approved it. Exportable.

  • Role-based access

    Agents get the narrowest scopes each tool allows, per team and per workflow.

  • Your data stays yours

    No training on customer data. EU or US hosting. SSO and SCIM on every plan.

Use cases

Start with the team that's drowning.

  • Tray 013 request types

    Support ops

    • Refunds and credits
    • Account and email changes
    • Bug reports routed to the right team

    61% of tickets closed without a humanillustrative

  • Tray 023 request types

    Finance ops

    • Missing and corrected invoices
    • Failed payment follow-ups
    • Matching payouts to invoices

    9 days faster month-endillustrative

  • Tray 033 request types

    People ops

    • Onboarding paperwork
    • Tool access requests
    • Equipment orders

    Day-one laptops, every timeillustrative

The Dockett sorting machine as a thin ink line drawing.

Fig. 1 · sorting machine, ink pass

How it's made

A toy machine, built to run at 60 fps.

Framework
React 19 · TypeScript · Vite · React Router v7, pre-rendered
3D
Three.js through React Three Fiber and Drei
Motion
GSAP ScrollTrigger and Lenis on one shared ticker
Machine
Built in code from simple parts; ink edges as geometry, one baked shadow pass
Cards
One instanced mesh per layer; every label drawn at runtime from copy
Weight
No model download; the whole 3D chunk loads after the hero text

Dockett is a design concept by Nova Digital Studio: strategy, copy, design and build.

Book a demo

Send Dockett your first request.

Bring your ten most common requests. We'll run them through a sandbox connected to copies of your tools and show you every stamp.

PricingPer resolved request, never per seat. Published page coming soon.

30 minutes, with your own requests. No slides.