Skip to content

Service4 of 6

AI and automation

We automate tasks in your company with language models connected to your data. Each workflow comes documented, evaluated and with an estimate of what it costs to run.

Category
AI and automationVerified
Detailed capabilities
Pending verification
Engagement models
  • Fixed-scope project
  • Maintenance and operations retainer
  • Block of hours
Handover record
Repository · Documentation · Access · Infrastructure in your name

A workflow with human approval

The model proposes; it does not decide. Pick a scenario and see what the evaluation checks, when a person steps in and what happens if the model is unsure or does not answer.

Illustrative diagram. It does not represent any client’s architecture.

Task: classify a support e-mail and draft a reply from the knowledge base.

  1. InputThe e-mail and its context
  2. ProposalThe model classifies and drafts
  3. EvaluationAgreed criteria, before anything is shown
  4. PersonApproves, corrects or rejects
  5. ResultSent and recorded
  6. FallbackManual queue with the original data

Every state

Confident

The proposal meets the criteria and reaches a person, who approves it.

Evaluation criteria
  • The category is on the allowed list. Meets
  • The reply cites the knowledge base. Meets
  • It promises nothing outside the policy. Meets
Proposal
Proposes the category and a draft with its source.
Evaluation
All three criteria are met.
Person
Reads the draft and approves it unchanged.
Outcome
The reply is sent and whoever approved it is recorded.

Unsure

One criterion fails: the proposal arrives flagged and the person corrects or rejects it.

Evaluation criteria
  • The category is on the allowed list. Meets
  • The reply cites the knowledge base. Fails
  • It promises nothing outside the policy. Meets
Proposal
Proposes a draft that cites no source.
Evaluation
The citation check fails: the proposal is flagged as unsure, with the reason.
Person
Corrects the draft, or rejects it and replies by hand.
Outcome
The case joins the evaluation set to review the workflow.

Unavailable

The model does not answer in time: the work moves to the manual queue and is not lost.

Evaluation criteria
  • The category is on the allowed list. Not evaluated
  • The reply cites the knowledge base. Not evaluated
  • It promises nothing outside the policy. Not evaluated
Proposal
The provider does not answer within the time limit.
Fallback
The e-mail enters the manual queue with its original data, and whoever runs the workflow is notified.
Person
Handles it as before the automation.
Outcome
The process carries on without the model. The failure is logged with its time.

Related scope: LLM workflowsKnowledge basesProcess automation

01What it solves

What it solves

Common situations, not client cases.

  • P-01

    Your team spends hours on tasks that repeat the same steps every time.

    What changesAutomated workflows that carry out those steps and keep a record of what they did.

  • P-02

    To answer a question, your team digs through documents kept in several places.

    What changesA knowledge base that answers from your own documents and cites where each answer comes from.

  • P-03

    You don’t know what an AI model would cost you or whether its answers would be reliable.

    What changesA quality evaluation and an estimate of usage costs before you commit.

02Scope

What’s included and what isn’t

Included

Detail pending verification
  • Steps where a language model classifies, summarises or drafts within an existing process.

  • Assistants that query your systems and act on them with limited permissions.

  • Your documents organised so they can be searched and cited.

  • Repetitive tasks connected across systems, without a manual step each time.

  • Tools that help your team work with its own information.

Not included

Proposal pending approval
  • EX-01

    Computer vision

    Not offered.

  • EX-02

    Training proprietary models

    Not offered.

  • EX-03

    Machine learning model operations (MLOps)

    Not offered as a specialty.

Boundaries proposed for review (A-07). They are not approved exclusions.

03Deliverables

What you receive at the end

This is what you receive. Each row shows whether it passes to you, depends on the contract or is not transferred.

KorhyYour company

  • Passes to you
  • Per contract and model
  • Not transferred

Final ownership is set by each project’s contract.

Base · in every service

  1. EN-01

    Repository

    The source code and its history, in a repository your company owns.

    Passes to you

  2. EN-02

    Documentation

    How the system is built, why it was built that way and how to get it running.

    Passes to you

  3. EN-03

    Access

    Admin credentials and permissions, handed to the people you name.

    Passes to you

  4. EN-04

    Infrastructure in your name

    Cloud provider accounts in your company’s name. The exact arrangement is set by the contract.

    Per contract and model

Specific to AI and automation

  1. EN-05

    Documented workflows

    What each workflow does, what data it works with and what permissions it uses.

    Passes to you

  2. EN-06

    Quality evaluation

    How the answers were checked and the criteria for accepting them.

    Passes to you

  3. EN-07

    Estimated usage costs

    An estimate of what each workflow will cost to run with the chosen provider.

    Passes to you

What is not transferred

Proposal pending approval
  1. NT-01

    Our own tools from before the project

    Internal accelerators we already had. You use them under a licence; they are not assigned.

    Licensed for use per contract

  2. NT-02

    Licensed third-party technology

    Services and software from other vendors that the solution uses.

    Subject to the vendor’s licence

04Engagement

Engagement models

We don’t publish prices: they depend on the scope. Response times, where they apply, are set in the contract.

  • A

    Fixed-scope project

    When it applies
    When you want to automate a specific task and check that it works before extending it.
    How it works
    We agree on scope, exclusions, milestones and price. The quality criteria are set before building.
    Deliverables
    Deployed workflows, quality evaluation, estimated usage costs and a handover record.
    Contract considerations
    The price is set in the proposal. Model provider usage is billed separately, in your name.
    SelectedRemove: Fixed-scope project
  • B

    Maintenance and operations retainer

    When it applies
    When the workflows are already in use and need ongoing adjustment and follow-up.
    How it works
    A monthly fee covers an agreed set of adjustments, quality reviews and improvements.
    Deliverables
    A periodic report and up-to-date workflow documentation.
    Contract considerations
    The monthly scope and the SLA are agreed in writing in the contract.
    Enquire with this model: Maintenance and operations retainer
  • C

    Block of hours

    When it applies
    When you want to explore a one-off case or adjust an existing workflow.
    How it works
    Prepaid hours used task by task, with a log of what was done.
    Deliverables
    Hours log and the documentation for each task.
    Contract considerations
    How long the hours stay valid, and on what terms, per the contract.
    Enquire with this model: Block of hours

05Questions

Frequently asked questions

What people ask us most before starting.

  • It depends on the scope. After the diagnosis we send you a proposal with the scope, timeline and cost. If you go ahead with the project, the diagnosis fee is credited against the proposal.

  • Your company does: the model provider account and its usage are contracted in your name and billed separately. The proposal includes an estimate of those costs.

  • We can sign a confidentiality agreement before we access it. The handover documents which data each workflow uses, and with which permissions.

  • Yes. The code, the workflow documentation and the access pass to your company at handover. The exact ownership terms are set in each project’s contract.

  • Yes. Through a maintenance and operations retainer with an SLA agreed in writing, or a block of hours for occasional adjustments.

06Next step

Tell us which task you want to automate.

Book a call

Online scheduling is not available yet.

Your enquiry

This comes with you to the contact form. The link only carries the service and the model.

Service
AI and automation
Model
Fixed-scope projectRemove or change the model
Continue to the contact formSee the other services