DELIVERY METHODOLOGY
From first conversation to a measured pilot.
A staged process that makes your effort, our effort, and the human checkpoints visible from the start — no unconditional promise on timing, no “plug and play” shortcut.
7-STAGE ROADMAP
Detailed Delivery Lifecycle
- 01STAGE 01
Assess
We identify the business objective, the workflow, expected volume, the owner, your current baseline, and any constraints or risks.
Your Team Contributes:
access to the workflow owner, sample non-sensitive materials, any existing measures you already track, and your desired result.
AI SMITH Delivers:
discovery, prioritisation, readiness findings, and a recommended starting scope.
- 02STAGE 02
Design
We define the agent's role, its inputs and outputs, its boundaries, escalation rules, the decisions that stay human, and how success will be measured.
PRIMARY OUTPUT:an Agent Role Card and a target workflow definition. - 03STAGE 03
Prepare knowledge and access
We collect the approved documents, rules, templates, source owners, and update dates needed for the pilot — and request only the access that pilot actually needs.
PRIMARY OUTPUT:an approved knowledge set and an access plan. - 04STAGE 04
Configure and test
We prepare workflows, examples, and expected behaviour, and test normal, uncertain, and high-risk scenarios against agreed evaluation criteria.
PRIMARY OUTPUT:a pilot-ready workflow with its known limitations documented. - 05STAGE 05
Pilot with people in control
We run the workflow within defined boundaries, with an accountable human owner reviewing outputs and correction and exception data captured throughout.
PRIMARY OUTPUT:pilot evidence and an issue log. - 06STAGE 06
Measure and decide
We compare quality, time, throughput, service, commercial, or risk measures against the baseline you started with, and decide together whether to improve, expand, pause, or stop.
PRIMARY OUTPUT:a value review and a next-step recommendation. - 07STAGE 07
Improve and expand
We refresh knowledge, refine the workflow, add a related agent, connect another approved source, or scale to another team — only once the evidence supports it.
A good first workflow is:
- Repeatable
- Valuable
- Bounded in scope
- Measurable
- Backed by enough examples to learn from
- Owned by a named person
- Manageable in risk
A poor first workflow is:
- Rare or infrequent
- Ambiguous in what "done" looks like
- Missing a clear owner
- Missing approved data
- An irreversible, high-stakes action
- Impossible to measure
Human checkpoints exist at every point where a decision, record, or communication has real consequence:
- Preparation of any material before it's used.
- External communication before it's sent.
- Any change to a business record.
- Any financial action.
- Any decision about a person.
- Any contract or agreement.
- Any deletion.
What Affects Timing
The scope you choose, how ready your knowledge is, how quickly access can be arranged, how many exceptions the workflow involves, integration complexity, stakeholder availability, and your review requirements.
Pilot Deliverables
An Agent Role Card, an approved knowledge set, a pilot-ready workflow with documented limitations, pilot evidence and an issue log, and an empirical value review with a next-step recommendation.
QUESTIONS
Frequently Asked Questions
What do we need to prepare before starting?
A named workflow owner, sample non-sensitive materials relevant to the workflow, any measures you already track, and clarity on your desired outcome. We confirm the exact list during the Assess stage for your specific workflow.
What data does AI SMITH need access to?
Only the access the specific pilot workflow needs — never broader access than the scope requires. This is defined together during the Prepare Knowledge and Access stage.
Will it work with our existing tools?
Only where a connection is verified or built as a custom integration for your account. We'll tell you plainly during Assess and Design if something you need isn't yet supported.
What languages are supported?
Language coverage depends on what your approved knowledge and review process can support; we confirm this for your specific case rather than assuming broad coverage by default.
Does our team need training to use this?
The workflow owner and reviewers need to understand how to review agent outputs and when to escalate — we support this as part of the pilot, scoped to your workflow.
What happens after the pilot?
We review the measured value together and decide whether to improve, expand, pause, or stop — there's no obligation to continue past a pilot, and no result is guaranteed in advance.
See what an assessment of your business would actually involve.
A consultation is a conversation about one priority workflow — not a sales pitch for every agent at once.