Agent scope document
Workflow definition, success criteria, edge case mapping
[ Outcome ]
A production-grade AI agent handling a specific workflow in your business — with full documentation and a trained team.
Custom AI agents are the highest-leverage form of AI deployment — because they’re built around the specific task, the specific data, and the specific quality standard that matters to you.
They don’t hallucinate about your products because they’re trained on your product data. They don’t give wrong policy information because they’re constrained to your policy documents. They don’t respond in the wrong tone because they’re trained on your brand voice.
[ What's included ]
Every deliverable. No surprises.
Architecture design
Model selection, tool use, integration approach
Custom training and knowledge base
Fine-tuning on your specific data
Integration build
API connection to CRM, helpdesk, website, or custom platform
Testing protocol
Output quality, edge case coverage, escalation triggers
Production deployment
Live in your environment, monitored for first 14 days
Operations runbook
Full documentation for every function and escalation
Team training
2-hour session with hands-on testing
30-day post-launch support
Bug fixes, edge cases, quality review
For businesses with a specific identifiable workflow that meets three criteria: it’s high volume (hundreds or thousands of times per month), it’s currently handled manually, and it has a definable quality standard.
Common use cases: customer service triage, lead qualification agents connected to portals and CRMs, internal knowledge retrieval, automated proposal generation, lease and contract abstraction.
Vague briefs ("we want an AI agent that does everything") or use cases without a definable quality standard. If you're not sure what the workflow is, start with the AI Launchpad.
[ FAQ ]
Common questions.
What AI models do you build on?
Primarily Claude (Anthropic) and GPT-4o (OpenAI), chosen based on the specific task requirements. For retrieval-heavy knowledge agents, we often use Claude for its large context window.
How do you handle data security and GDPR?
All agent builds operate under a signed Data Processing Agreement. We use model APIs with business tier privacy guarantees — client data is never used for training. For sensitive data, we can architect for on-premise or private cloud deployment.
What happens when the agent gets something wrong?
Every agent has defined escalation triggers — edge cases or confidence thresholds that route to a human. The runbook documents every escalation path. During the 30-day support period, we review output quality weekly and retrain on failure cases.
Is £8,000 the full cost including API usage?
The £8,000 build fee covers design, development, training, integration, testing, and handover. Ongoing API costs (typically £50–£500/month depending on volume) are billed directly to your account.
Can we expand the agent's capabilities after launch?
Yes. Agent expansion is typically scoped as a separate project once the initial build is stable. Most clients return for a second phase within 3–6 months.
You have a workflow that should be running itself.
Book a 30-minute agent discovery call. Come with the workflow in mind. We'll tell you whether it's a good agent candidate, what it would take to build, and what it would save your business every month.