AssessmentScoped individually
AI Readiness and Opportunity Mapping
A consultant-led assessment of your data, processes, systems and team, with an opportunity map and a prioritised written plan.
AI agent development
AI agent development for customer service, operations and internal work
An AI agent is more than a chatbot: it reads data, plans the next step and uses tools such as your CRM, store or inbox. We design agents with a narrow, clearly described job, minimal permissions and human approval points — so they help without taking on uncontrolled risk.
01
Task and limits
02
Prototype in a test environment
03
Evaluation and tuning
04
Release with monitoring
- Ways to work together
- Fixed-scope projectOngoing operation through AI Managed Services
- What you receive
- A description of the task, its limits and the approval points
- First step
- A short call to clarify your goals and the current state.
When you need it
When an AI agent makes sense for your business
Your team repeats the same multi-step tasks across several systems.
Your chatbot answers questions but cannot do anything about the order or request itself.
You want AI to take on routine work, but not decisions with financial impact.
You tried an agent that behaved unpredictably or left no trace of what it did.
Our approach
How we solve it
A narrow job, clear limits
The agent gets one described task, a list of permitted tools and rules for when to stop and ask a person.
Tools with minimal permissions
Every action goes through a specific API function with limited access — never through a shared admin account.
Approval and an audit trail
Actions with impact wait for approval, and every step is logged so it can be reviewed and replayed.
AI agent development
The AI agents we build
Customer service agents
Check order status, draft replies and hand over to a person when they are not sure.
Operations agents
Handle returns, stock mismatches and recurring requests according to defined rules.
Internal knowledge agents
Answer your team's questions from documents and policies, citing the sources.
Tools and integrations
Access to your CRM, store, inbox or ERP through narrow API functions — only the actions that are needed.
Rules and human approval
Actions that affect customers, money or data pass fixed checks and an approval step.
Evaluation and monitoring
Test scenarios before release, a log of every step and alerts when behaviour drifts.
Use cases
Where it helps in practice
Typical scenarios — every project starts from your own process and data.
Order status and returns
The agent checks the order, drafts a reply and a return proposal that a team member approves.
Inbox triage
Classifies e-mails, extracts the key data and routes them to the right person or system.
Quote preparation
Pulls catalogue and CRM data into a draft quote for a salesperson to review.
Technical review
What we check before building an agent
Repetition and volume
How often the task runs and how many steps it has — agents make sense for repeatable work with clear rules.
Available systems and APIs
Which systems the agent needs to reach and whether they offer APIs with narrow enough permissions.
Actions that need a person
Which steps affect customers, money or data and must therefore pass an approval step.
Cost of mistakes and escalation
What happens after a wrong answer or action, and when the agent must hand over to a person.
How it is evaluated
Test scenarios from real cases and the criteria for accepting the agent into service.
What you receive from an AI agent project
- A description of the task, its limits and the approval points
- A working agent in a test environment, checked with real examples
- A test set and quality criteria
- Logs, documentation and a monitoring plan
Ways to work together
- Fixed-scope project
- Ongoing operation through AI Managed Services
Pricing is set once the scope is clear and is provided in a written quote.
Request a quoteWorkshops and assessments
Consultant-led formats with a clear scope that we usually start with.
WorkshopScoped individually
Agent Architecture Workshop
A technical session in which we design the agents, their tools, the limits of their access and the points where a person approves.
How we work
How we develop an AI agent
What we need from you
- A description of the task and examples from real work
- Access to test environments or copies of the systems
- A team member who approves and assesses the results
- Requirements for privacy and data retention
Task and limits
We describe the agent's job, its tools, the rules and the approval points.
Prototype in a test environment
The agent works on real but safe examples, with no access to live actions.
Evaluation and tuning
Checks against the test scenarios; tuning instructions, tools and rules.
Release with monitoring
A limited release, a log of every step, alerts and a rollback plan.
AI agent development
Frequently asked questions
What is the difference between an AI agent and a chatbot?
A chatbot answers questions. An agent carries out a multi-step task — for example checking an order, drafting a reply and proposing a refund — using only the tools it has been explicitly allowed to use.
Can the agent act on its own?
Only within the limits you set. Reading, classifying and drafting are usually autonomous; messages to customers, changes to orders and payments require approval.
What happens if the agent makes a mistake?
Every step is logged, actions with impact require approval, and when the agent is unsure it escalates to a person. Quality is checked against test scenarios before release.
Related tools
Start with a quick self-assessment before we talk.
- Available now
AI process analysis
AI Automation Advisor
Describe a manual process and get a structured first assessment: automation opportunities, workflow, integrations, risks and the points that need human approval.
- For:
- Owners and managers with a specific manual process
- You get:
- A structured first assessment: opportunities, workflow, integrations, risks and approval points.
- Time:
- about 3 min
- Available now
Agents
Build your AI team
Choose agents, the tools they may use and the approvals they need — and see the system as a diagram.
- For:
- Teams planning AI agents
- You get:
- A diagram of your AI team with its agents, tools and approvals, plus indicative costs.
- Time:
- about 6 min
- Available now
Architecture
AI solution builder
Compose a solution from agents, integrations and approval points and get an architecture diagram to discuss.
- For:
- Technical leads and product teams
- You get:
- An architecture diagram of the solution with agents, integrations and approval points.
- Time:
- about 5 min
- Available now
Agents
AI agent library
Browse typical agent roles by business function: tasks, tools and limits. The catalogue is for planning, and the agents are not connected to your systems.
- For:
- Teams looking for ideas for their first agent
- You get:
- Profiles of typical agents: tasks, tools, limits and approvals.
Useful reading
- Guide5 min read
AI customer assistants: when they make sense and how to introduce them safely
When an AI customer service assistant delivers real value for an online store, what the risks are, and how to introduce one with quality control.
- Guide8 min read
How AI automation reduces manual work in small businesses
A practical guide to AI automation for small businesses: which tasks are a good fit, how to estimate the benefit in advance and how to keep control of important decisions.
Related services
AI workflow automation
Inquiry classification, CRM workflows, document processing and system integrations — with human approval where needed.
AI integrations & data systems
REST API and webhook integrations, connecting Shopify and WooCommerce with CRM and ERP, cloud deployment, monitoring and protection.
AI Managed Services & optimisation
Monitoring, maintenance, quality evaluation, security and cost optimisation for AI agents and automations that are already in production.
Discuss this service with us
Describe your task and we will suggest a suitable scope and engagement model.
- Written quote with a clear scope
- Transparent milestones and timelines
- Your data is used only to reply to you