Consulting service
AI Workflow Integration
Assessment-led integration of AI into a specific workflow, with the human decision preserved, the output checked, and honest limits about what it can be relied on for.
What you need to know first
- What problem does this solve?
- There is pressure to "use AI" without an agreed answer to where it would genuinely help. The risk is not that it does nothing; it is that it is placed at a step where a plausible wrong answer is costly, and nobody notices because the output reads well.
- What exactly do I receive?
- AI integrated into a defined workflow step where it demonstrably helps, with the human decision preserved, a way of checking the output, measured results against the current process, and a written statement of what it cannot be relied on for.
- How is it priced?
- Assessment requiredThe fee is set after a scoped assessment, because the work cannot be priced responsibly before the current situation is understood.The fee is set after a scoped assessment, and the assessment is genuinely capable of concluding that AI is not the right tool for the step in question. Pricing an AI integration before that conclusion exists would create an incentive to reach the other one.
- How long does it generally take?
- Confirmed after the assessmentNo indicative duration is published. It depends on which step is chosen, what evaluation the output needs, and what model and data access is available to you.
- What happens next?
- Send an enquiry describing the requirement. We confirm whether an assessment is the right starting point, then set out its scope and fee in writing before any work begins. An enquiry is not an order: an engagement becomes binding only when it is expressly accepted, as set out in the Terms of Service.
Deliverables
- Written assessment of candidate workflow steps, with a recommendation and the reasoning for rejecting the others
- The integration itself, at the agreed step
- A human review point, positioned where a wrong output would matter
- An evaluation method: how the output is checked, and what triggers escalation to a person
- Measured comparison against the current process
- A written statement of limitations: what this must not be relied on for
Included
- Assessment of where AI would genuinely help in the workflow, and where it would not
- Model and provider selection appropriate to the task and to your data constraints
- Integration into the agreed workflow step
- Human review and escalation design
- Output evaluation and measurement against the current process
- Two adjustment rounds and written limitations
Not included
Stating the boundary is part of the offer. Anything here can usually be arranged separately.
- Model training, fine-tuning or custom model development, unless separately scoped
- AI provider or model usage fees, which stay in your accounts
- Any guarantee of output accuracy, of a quality level, or of a cost saving
- Automating a decision the assessment concluded must stay with a person
- A guarantee that a third-party model or API will remain available, priced or behave unchanged
- Regulatory or compliance sign-off for AI use in your sector
Customer prerequisites
- A specific workflow with a step that is a plausible candidate
- Agreement on what an acceptable output looks like, and who judges it
- Clarity on what data may and may not be sent to a third-party provider
Information and access we need
We will ask for these. Where the work needs access to a system, access is arranged and scoped with you rather than assumed.
- The workflow and the step you have in mind, and why
- Real examples of inputs and of good and bad outputs
- Your constraints on data leaving your environment, and any provider restrictions
- Who reviews the output, and what they do when it is wrong
- What the current process costs in time, so a comparison is possible
Expected process
Enquiry
You describe the workflow and where you think AI might help.
Assessment
A scoped assessment identifies the candidate steps, recommends one or none, and states the data constraints.
Written scope and fee
The integration is scoped and priced from the assessment, in writing.
Build with review point
Integration at the agreed step, with the human review and escalation path built in from the start.
Measure and adjust
Measured against the current process, then two adjustment rounds, then written limitations and handover.
Meetings, revisions and scope
- Meetings
- Meeting required, during the assessment (about 90 minutes)
- Revisions
- 2 revisions included. Two rounds of adjustment after the integration is running against real cases, including tuning of the review threshold.
- Scope
- Deliverables are limited to those expressly included in the agreed scope. Additional revisions, meetings, research or services outside that scope may require additional fees — see the Consulting Terms.
Frequently asked questions about this service
What happens if a proposed implementation is not feasible?
You are told, with the reason, and you are not charged for an implementation that cannot be delivered.
This is the main reason the Consulting catalogue separates an assessment from an implementation. The assessment is the paid piece of work that establishes whether something is practical, what it would take, and what it would cost. Its output is useful even when the answer is "not this way": it will say what does work, what the blockers are, and what a smaller or different change would achieve.
Where a recommendation depends on software, platforms, vendors, APIs or external suppliers, their availability and their decisions are outside our control — stated in Third-Party Dependencies.
Will you need access to our systems?
For most Consulting work, some access is needed — but it is scoped and agreed with you, never assumed. Each service page lists what it needs under "Information and access we need".
The principle applied is least privilege and shortest duration: read-only where reading is enough, a sandbox or test environment where one exists, access limited to the systems in scope, and removal at the end of the engagement.
The Consulting Services Terms record that the client provides reasonably necessary access and credentials, and that each party will use reasonable safeguards for confidential information received in connection with the engagement.
How is sensitive access handled?
Practically, and with your own controls rather than ours:
- access is requested for named systems and a stated purpose, not as blanket administrator rights
- credentials are issued by you through your own identity and access management, so you can see and revoke them
- shared credentials are avoided wherever your systems support individual accounts, because a shared one cannot be attributed or cleanly revoked
- production data is not copied out of your environment where the work can be done inside it
- access ends when the engagement does, and you should verify the removal yourself
Where a more specific arrangement is needed, an NDA or engagement agreement controls over the general terms — Confidentiality.
Is an assessment charged separately?
Yes. An assessment is its own service with its own fee and its own deliverable, and it is not a free pre-sales exercise.
That is deliberate. An assessment that is bundled into an implementation quietly pressures its own conclusion towards "yes, proceed". Paid for separately, it is free to conclude that the implementation should be smaller, different, or not done — which is frequently the most valuable thing it can tell you.
Where an assessment and a following implementation are commissioned together, any relationship between the two fees is stated in the written scope.
Next step
Reference ai-workflow-integration is added to your enquiry automatically so it reaches the right place.