Skip to content

How it works

A safe path from idea to working agent.

Six stages, each with a clear deliverable and an approval point. You always know what happens next, what we need from you and where a person stays in control.

  1. Stage 1

    Automation Assessment

  2. Stage 2

    Workflow Mapping

  3. Stage 3

    Agent Design

  4. Stage 4

    Integration

  5. Stage 5

    Testing & Safeguards

  6. Stage 6

    Launch & Optimisation

The blueprint

Six stages. One approval point after each.

Open a stage to see exactly what we deliver, what we need from you, the safeguards that apply and a sample of what you will see.
  1. Deliverables

    • A map of one workflow: where enquiries arrive, what your team asks and what happens next
    • An automation-fit review: which steps suit an agent and which should stay with people
    • A recommended first workflow and starting package
    • A written summary of risks, dependencies and open questions

    Client inputs

    • A 30-minute discovery call with the person who owns the process
    • Examples of real enquiries, emails or call notes
    • The tools you use today: inbox, calendar, CRM, phone system
    • What a successful outcome looks like for you

    Safeguards

    • We recommend against automating steps that need human judgement
    • No system access is requested at this stage
    • Scope is agreed before any price is confirmed

    Visual sample

    Automation assessment

    • Where do enquiries arrive?WhatsApp and the website form
    • What does the team ask first?Budget, area, timing
    • What may the agent use?FAQ and price list
    • When does a person take over?Complaints, high-value enquiries
    • Which system is updated?CRM pipeline
    Good fit for a first agent
    Suggested: Pro

    Example assessment checklist: five discovery questions answered, with the recommendation to start with a Pro package

  2. Deliverables

    • A workflow diagram covering triggers, questions, decisions and actions
    • The qualification questions and the data the agent should capture
    • Handoff points: where a person takes over, and why
    • The systems to connect and the fields to update in each

    Client inputs

    • Review and approve the workflow diagram
    • Confirm the qualification questions and decision rules
    • Name the colleague who receives escalations, and how

    Safeguards

    • Every handoff point is agreed before we design a single response
    • Regulated or high-stakes decisions are mapped to a person, never to the agent
    • Personal data captured is limited to what the workflow needs

    Visual sample

    Example workflow diagram: WhatsApp and web chat enquiries flow to the agent, which books in the calendar, updates the CRM or hands over to the team

  3. Deliverables

    • Agent instructions: role, tone, scope and an explicit list of what it must not do
    • A knowledge base built only from content you have approved
    • Example responses for common and difficult situations
    • Handoff and escalation rules written in plain language

    Client inputs

    • Approved FAQs, documents, price lists or website content
    • Tone preferences and the languages to support
    • Sign-off on the instructions and the handoff rules

    Safeguards

    • The agent answers from approved knowledge only, and says so when it does not know
    • Out-of-scope topics are listed explicitly and routed to a person
    • Instructions are versioned, so every change is traceable

    Visual sample

    Example agent instructions card: role and tone, approved knowledge sources switched on, open web search switched off, and three handoff rules switched on

  4. Deliverables

    • Channels connected as agreed: website chat, WhatsApp, email or phone
    • Calendar and CRM connections with agreed field mapping
    • Team notifications by email, Slack, Telegram or your preferred tool
    • Automation flows documented so they can be maintained

    Client inputs

    • Access to the chosen channels, such as your WhatsApp Business account
    • System access or API credentials at the minimum level the workflow needs
    • A test calendar, pipeline or mailbox where available

    Safeguards

    • Least-privilege access: only the permissions the workflow requires
    • Test environments or sandboxes are used where the platform offers them
    • You keep ownership of every account and can revoke access at any time

    Visual sample

    Example connected-systems panel: website chat, WhatsApp, Google Calendar and Slack connected, CRM field mapping in progress, phone marked as not in scope

  5. Deliverables

    • Test scenarios covering common, edge and ‘should not answer’ cases
    • A test report listing results and the corrections made
    • Handoff and escalation tested end to end with your team
    • Defined fallback behaviour for when the agent is unsure

    Client inputs

    • A testing session where your team tries realistic scenarios
    • Feedback on tone, accuracy and handoff timing
    • Written approval to go live

    Safeguards

    • Nothing goes live without your approval
    • Failed scenarios are fixed and re-tested, not waived
    • The agent is instructed never to guess; unknowns are logged and handed over

    Visual sample

    Example test-scenario list: five scenarios passed, one failed and then fixed, with a summary that all scenarios pass after corrections

  6. Deliverables

    • A controlled launch: test mode → team review → live
    • Monitoring at the agreed level, with alerts for handoffs and failures
    • Basic documentation for your team
    • Optimisation reviews where your plan includes them

    Client inputs

    • Tell us what to improve after the first real conversations
    • Keep approved content current — we help you spot gaps
    • Decide when to expand to more channels or workflows

    Safeguards

    • The human handoff switch stays on at all times
    • Conversation logs are available to your team
    • Changes to instructions or knowledge go through the same review and test step

    Visual sample

    Example launch control: test mode and team review completed, agent live, human handoff switch on, and a small monitoring dashboard with example figures

Your side

What we need from you.

Less than you might expect. Most clients spend a few focused hours across the whole process.

A process owner

Someone who knows how enquiries are handled today and can make decisions.

Approved information

FAQs, documents, price lists or website content the agent may use.

System access

The chosen channel and business software, at the minimum level needed.

Escalation contacts

Who receives handoffs, and how they prefer to be notified.

Sample conversations

Real enquiries, emails or call notes so we design for reality.

Privacy and compliance requirements

Anything the agent must never say, collect or decide.

A definition of success

The outcome that tells you the agent is working.

Our side

What you receive.

Every engagement ends with the same set of tangible outputs, whatever the workflow.

Configured workflows

The agreed process, built and running.

Integrations

Channels, calendar, CRM and notifications connected.

Agent instructions

Role, tone, scope and explicit limits, versioned.

Test scenarios

The cases we tested, with results and corrections.

Handoff rules

When and how a person takes over, in plain language.

Basic documentation

How the agent works and how to request changes.

Launch support

A controlled go-live with us watching closely.

Agreed monitoring

Ongoing review at the level your plan includes.

Simple agents can often be launched quickly once we receive the required content and access. More complex, voice, custom or regulated workflows require additional design and testing. After the discovery call, we provide a clear implementation plan.

FAQ

Questions about the process.

Honest answers about timing, tools and control.

Simple agents can often be launched quickly once we receive the required content and access. More complex, voice, custom or regulated workflows require additional design and testing. After the discovery call, we provide a clear implementation plan.

Usually not. We connect to the systems you already use whenever technically possible. If a tool has no usable API, we discuss the options with you before anything is built.

That is exactly what testing is for. Failed scenarios are corrected and re-tested, never waived. Nothing goes live until your team has approved the behaviour.

Yes. The agent can be switched back to test mode at any time, and the human handoff switch is always on. Your team keeps ownership of every connected account.

You do. The configuration is built in your accounts wherever possible, and you receive documentation describing how it works. We can continue to maintain it under your plan, or hand it over.

Ready to map your first workflow?

Bring one repetitive process to a 30-minute call. We’ll tell you honestly whether an agent fits—and what the first version would look like.

Free 30-minute consultation · We’ll identify the best first workflow