Adopting AI-native RPA your staff build and grow

The people doing the work build, fix, and grow their own automations by talking with AI. This is how to adopt it: prove it in a trial and a pilot, set the rules, and spread it from department to department.

Connecting systems with no API is a different engagement, which Descarty builds and maintains, with its own steps.See how system integration runs

The whole path

  1. 1Trial60 minutes, online
  2. 2Pilot plan1 to 2 weeks
  3. 3Pilot (one-month joint verification)One month
  4. 4AdoptionAbout one month
  5. 5Rollout1 to 2 months per department
  6. 6Run and growOngoing
  1. 1

    Trial

    60 minutes, onlineFree
    You
    A member of your staff joins with one task of their own. Kitewell runs on your PC or in an environment Descarty provides.
    Descarty
    We sit beside them while they build that task by talking with AI.
    What you get
    Whether your staff can build it themselves
  2. 2

    Pilot plan

    1 to 2 weeksFree
    You
    Choose the department, and give us its task list and procedure documents. Your DX lead and someone from IT join.
    Descarty
    We list and rank the tasks, and decide with you who builds what, the success criteria, and where it runs. We answer your security questionnaire.
    What you get
    A pilot plan (tasks, team, success criteria, setup) and a quote
  3. Free up to here. This is where you find out whether your staff can build it, and settle the pilot plan.
  4. 3

    Pilot (one-month joint verification)

    One month¥500,000 before tax, free if the success criteria are not met
    You
    Sign an order form and NDA, and give the department's staff time for training and building.
    Descarty
    After training, we stay beside your staff as they build their own tasks, and build the hard parts ourselves. At the end, we measure the effect and report it with a rollout plan.
    What you get
    Measured time saved, how many workflows staff built and fixed themselves, and a rollout plan
  5. 4

    Adoption

    About one monthLicenses (quoted). The verification fee is credited to the first year
    You
    Provide shared computers or employees' PCs, and set your internal guidelines.
    Descarty
    We set up with you the permissions, the review before anything goes to production, the robot ledger, and the training program.
    What you get
    The environment and rules to spread it across the company
  6. 5

    Rollout

    1 to 2 months per departmentTraining and launch support (quoted)
    You
    Choose the builders in each department; the DX team reviews what they build, totals the effect, and reports it to management.
    Descarty
    We support each department's training and launch, and help with cross-department work and hard automations.
    What you get
    Automation spread from department to department, with its effect totaled
  7. 6

    Run and grow

    OngoingLicenses and maintenance support (quoted)
    You
    Staff fix their own workflows as the work changes and keep improving them. The DX team keeps track of everything in the ledger.
    Descarty
    We keep up product support, finding the cause of stops AI can't fix, checking updates first, and help with hard automations.
    What you get
    Automation your staff keep and grow

Plans for companies are invoiced, priced in yen, and contracted yearly.

Who builds, and who fixes

Who builds depends on the work. The split is set in the pilot plan and becomes your internal rules at adoption.

The people doing the work

Build, fix, and grow their own automations by talking with AI.

  • Entering data into Excel or web systems
  • Handling documents that arrive by email
  • Getting information from customers' and suppliers' sites
  • Changing steps when the work changes

The DX team and IT

Build and review what crosses departments or has a large impact.

  • Work that crosses departments
  • Work that writes to core systems
  • Anything that runs on shared computers
  • Reviewing what staff build before it goes to production

Descarty and partners

Support the hard parts and the launch.

  • Connecting systems with no API
  • Work with many exceptions, and hard automations
  • Launch and training
  • Product support

Keeping orphan robots from appearing

  • A workflow AI builds is saved only after a person checks it
  • The DX team reviews what staff build before it goes to production
  • Every workflow is listed in one robot ledger
  • Who changed or ran what, and when, stays on record
  • Each project sets who can build and who can use

Choosing what the pilot covers

We list the department's tasks and choose by effect, together with who builds each. The choice comes from all of the department's work, not one task.

Example: a sales department's tasks

TaskItems a monthHours a monthSystemsWhat to automateBuilt byOrder
Order entry1,200100Email, sales systemAI reads each order and registers it; a person checks only those whose amounts differStaff1
Checking customer portalsDaily20Five customers' sitesCollect new orders and date changes every morningStaff1
Invoicing30030Sales system, accounting softwareCreate and send invoices right after orders are registeredDX team2
Shipping instructions1,00040Sales system, warehouse siteEnter shipping instructions on the warehouse siteStaff2
Payment matching30025Online banking, accounting softwareMatch statements to invoices; a person checks only the mismatchesDX team3

An example rollout plan

Pilot (one month)
Order entry and checking customer portals
Months 2 to 3
The department's other tasks (invoicing, shipping instructions, payment matching)
From month 4
Other departments

An example. Your actual plan is built from the task lists and procedure documents you give us, and from what we hear.

Settling the setup

In the pilot plan, we decide these with your IT team, with the company-wide rollout in view.

Scale

This sets the number of computers and licenses.

  • How many tasks to automate, and how many items a month
  • How many run at once, and when (overnight, month-end)
  • Which departments use it, and how many people build

Where it runs

Run records, results, and credentials stay on the computer that runs it.

  • Employees' PCs
  • Shared computers (VMs in your Azure, AWS, or Google Cloud, or your own servers), which run with no one signed in
  • A closed network, with no outside traffic
  • A VM Descarty hosts for you alone

What starts it

Set to fit the work.

  • On a schedule (every morning, at month-end)
  • When an email arrives (Gmail, Microsoft 365, IMAP)
  • By staff, when they need it
  • For each row of an Excel list

AI model

Screen content goes to the model you choose.

  • Azure OpenAI, Amazon Bedrock, or Google Vertex AI under your contract
  • Major AI model APIs
  • In-house or local models that are OpenAI-compatible

Data and security

Set to match your rules.

  • Which AI the screen content may go to
  • Whether to use team sync, or keep everything inside your environment
  • How long to keep the record of changes and runs (audit log)

Support

Who to ask when something goes wrong.

  • Whether the DX team takes staff's questions
  • The DX team's contact at Descarty
  • The response time for incidents (SLA)
  • Whether an SIer or partner takes first-line support

How the pilot runs

In one department, we prove that staff can build and fix their own work.

  1. Week 1

    Training and organizing the work

    We train the staff and map the business flow from your procedure documents, as it runs today and as it will run automated. Who builds what and the success criteria are set here too.

  2. Weeks 2 to 3

    Staff build

    Your staff build their own tasks by talking with AI, with Descarty beside them; we build the hard parts. A 30-minute weekly check-in tracks progress.

  3. Week 4

    Measure and report

    We run it on production screens and data, measure the effect, check it keeps working when part of a screen changes, and report with a rollout plan.

Example success criteria

  • The time the tasks take falls by at least the agreed share
  • Staff can fix the workflows they built themselves
  • Processing continues without changing the workflow when part of a screen changes
  • AI usage cost stays within the agreed limit

Results and cost

If the criteria are met
We invoice ¥500,000 before tax. If you go on to a contract, it is credited to the first year.
If they are not met
There is no charge. The report tells you why it fell short and how to deal with it.

Example: the order entry flow

Today

  1. Open the emailed order and check itHuman task
  2. Type the order into the sales system by handHuman task
  3. Check stock and prices by eyeHuman task
  4. Write the order confirmation and reply to the customerHuman task
  5. Copy the order into the Excel order ledgerHuman task

Automated

  1. Receive the order emailEmail
  2. Read the order (PDF or Excel)AI judgment
  3. Match stock and pricesAutomatic
  4. A person checks only orders whose amounts differApproval
  5. Register it in the sales systemWeb
  6. Reply with the confirmation and record it in the ledgerEmail

An example. Your actual business flow is mapped from the procedure documents and screens you give us.

Run and grow

The work keeps changing. What happens, and who handles it.

What happensWhat is done
The work changesThe person doing the work changes the steps by talking with AI. Why it changed stays in the knowledge.
A screen changesAI reads the screen again and keeps processing. Which steps it relearned stays in the run record.
Something new should be automatedStaff build it, and the DX team reviews it before production. Descarty helps with the hard ones.
The person in charge changesEach workflow's purpose and how to fix it stay in its knowledge, so the successor can take over by asking the assistant.
It stops and AI can't fix itDescarty finds the cause: a Kitewell bug, or a change on the other site or in the work. If it is a Kitewell bug, we deliver a fixed version.
Kitewell is updatedBefore the update, we check its effect on your workflows and tell you.
Every monthThe DX team checks everything's runs and effect in the ledger. Descarty reports runs as part of maintenance support.
An incidentWe respond within the time agreed in the contract (SLA).

See whether your staff can build it

In a 60-minute trial, a member of your staff builds one of their own tasks with AI. It costs nothing.