How it works
From first conversation to a system you own
Most disappointing AI projects skip the first stage. They start with a tool and look for somewhere to apply it. We start by writing down how your process actually runs, including the parts that are messy.
Discovery
We map how enquiries reach you today, what your team does with them, and where the repetitive work actually sits.
- Walk through the customer journey from first contact to closed job
- Identify the channels enquiries arrive on and who handles each
- List the questions your team answers repeatedly
- Document the tools already in place — CRM, calendar, phone, forms
- Agree what the agent must never do without a person involved
Agent Design
We define the agent’s goal, conversation logic, knowledge sources, escalation rules and the integrations it needs.
- Define the outcome each agent is responsible for
- Write the qualification criteria and the routing rules
- Assemble the approved knowledge base the agent answers from
- Set explicit escalation triggers and handover behaviour
- Confirm which integrations are available in your stack
Build & Test
We build the agent, connect it to your systems and test it against real scenarios, including the awkward ones.
- Build the conversation flows and connect the integrations
- Test against real transcripts and edge cases from your business
- Check escalation paths actually escalate
- Review tone and wording with you before anything goes live
- Run a limited pilot before full rollout where it makes sense
Launch & Optimise
We deploy, monitor real conversations, and refine prompts, knowledge and workflows based on what actually happens.
- Go live on an agreed scope, usually one channel first
- Review real conversation logs together in the early weeks
- Refine wording, knowledge gaps and escalation thresholds
- Expand to further channels or workflows once it is proven
- Hand over documentation so you know how the system works
What we need from you
The build depends on your input more than on ours
An agent is only as good as the knowledge and rules it is given. These are the things that make the difference between a useful agent and an embarrassing one.
How enquiries reach you
Which channels, roughly what volume, and who currently handles each one.
The questions you answer repeatedly
The ones your team could recite in their sleep. These become the knowledge base.
What makes a lead worth pursuing
Your real qualification criteria, including the ones nobody has written down.
Where a person must be involved
The conversations, values or customer types that should never be handled automatically.
Your tools
CRM, calendar, phone system, forms — what you use and who administers them.
Someone to make decisions
One person who can approve wording and rules. Builds stall on approval, not on code.
Honest limits
What an AI agent will not do
Being clear about this early prevents the disappointment that follows an oversold implementation.
It will not replace your sales judgement
An agent can qualify and route. Reading a hesitant buyer, negotiating, and knowing when to push are human work.
It will not fix a broken process
Automating a workflow that does not work simply produces the same bad outcome faster. Where we find that, we say so.
It will not integrate with everything
Some systems have no usable interface. Where that is the case, the agent captures structured data for a person to enter rather than pretending the integration exists.
It will not be perfect on day one
Real conversations always surface gaps that testing does not. The first weeks after launch are a refinement period, and we plan for that rather than declaring victory at go-live.
Questions
About implementation
What happens when the agent cannot answer something?
It escalates, using rules you define. That might be a live transfer, a structured message to your team, or a scheduled callback. An agent that says it will find out and hand over is far better than one that guesses, and we design for that.
How is the agent trained on my business?
From material you supply and approve — services, policies, pricing rules you are happy to publish, opening hours, FAQs, common objections and how you want them answered. The agent answers from that approved knowledge rather than from general assumptions about your industry.
How long does implementation take?
It depends on scope. A single-channel agent answering questions and capturing enquiries is a much shorter build than a multi-agent system integrated with a CRM, a calendar and a phone system. We give you a timeline against your specific scope after discovery rather than quoting a standard number.
Will I be able to update the agent later?
Yes. Knowledge, wording, qualification criteria and escalation rules are meant to change as your business does. We agree at the outset which parts you will maintain yourself and which we handle, and hand over documentation either way.
What do you need from me to get started?
A conversation about where customer conversations or repetitive work are slowing your team down, and access to enough detail about your current process to map it honestly. Nothing needs to be documented perfectly before we start.
Discuss Your Workflow
Bring us the part of your process that eats the most time. We will tell you what an agent would and would not change about it.