A bounded task
“Extract these fields from approved documents” or “draft a reply from this approved knowledge” is testable. “Handle the customer” or “decide the case” is not a sufficiently bounded task.
We use AI inside a bounded workflow when it can prepare, classify, extract or draft useful material, and when a person can still inspect, correct and approve the consequential action.
AI automation services place a bounded AI task inside a wider business workflow. Durward does not sell an autonomous agent or an AI model: we design the input, task, checks, human approval and recovery route around a specific business use case, then test it before any wider release. Whether you hire an AI automation agency or a freelance specialist, ask who looks after each step once it is running and how you would take the work back yourself.
AI automation services place a bounded AI task inside a wider business workflow. Durward does not sell an autonomous agent or an AI model: we design the input, task, checks, human approval and recovery route around a specific business use case, then test it before any wider release. Whether you hire an AI automation agency or a freelance specialist, ask who looks after each step once it is running and how you would take the work back yourself.
The technology is only one part of the service. The surrounding decisions determine whether it can be used responsibly.
“Extract these fields from approved documents” or “draft a reply from this approved knowledge” is testable. “Handle the customer” or “decide the case” is not a sufficiently bounded task.
The workflow can compare required fields, preserve source links, apply a known rule or route low-confidence material for review. The check is designed before the output is relied on.
A person owns approval, ambiguous outputs, changed instructions and any action with a customer or professional consequence. Human involvement is part of the operating design.
We test the workflow around the AI component, not just whether a model can produce a plausible-looking answer.
We specify the business trigger, authorised source material, AI task, required output and the person who owns the result.
Buyer receives: A constrained use-case brief and a clear exclusion boundary.
We check source quality, permissions, system access, exception volume, change sensitivity and whether non-AI automation would be simpler.
Buyer receives: A proceed, amend or stop decision before a prototype is relied on.
The AI task is combined with input shaping, deterministic checks, source retention and a human review queue.
Buyer receives: A testable workflow path with defined inputs, checks and review.
We use agreed examples, edge cases and failure conditions to establish what the output must contain, when it is rejected and who sees it.
Buyer receives: Acceptance criteria and an exception process that the owner can inspect.
A workflow is only released after the owner can work with the records, exceptions and manual recovery route; it is reviewed as source material or requirements change.
Buyer receives: An owned operating routine and a defined review point.
These patterns reflect the areas already represented in the Durward portfolio. They show the controls a working design needs; they do not represent deployed client outcomes.
A B2B supplier wants a more consistent starting point for researching an account from authorised public market sources.
A team holds client-authorised historic quote or service records and needs help preparing a follow-up queue without handing decisions to a model.
An operational team needs to prepare recurring documents and messages for a responsible person to assess.
A useful AI workflow has to be inspectable when it works, when it changes and when it fails.
AI is one option in the implementation route. It should earn its place by handling an information task that deterministic automation alone cannot reasonably handle.
The same answers as on the full FAQ page.
Routine messages can go out without a person reading each one, but only inside a scheme your team has approved in advance. Every workflow has a named owner and a defined escalation and review point, anything outside the scheme is held for a person, and special-category data never gets a solely automated decision.
Open in the full FAQYes, and we'd rather say it early: sometimes the honest answer is fix the process first, keep the platform you have, or that a step needs a person.
Open in the full FAQThen we'd say so: the review deliberately checks what you already pay for and haven't switched on, and "keep what you have" is a legitimate outcome.
Open in the full FAQThose are legitimate tools and may end up inside what we build. The difference is designing a system whose value survives a change of model, price or product: that's the part that's ours.
Open in the full FAQThat answer has to exist before launch: critical workflows keep a manual recovery path, and enquiries are preserved rather than lost while the automation is down.
Open in the full FAQIt keeps running: where the architecture allows it, the infrastructure and data sit in your accounts, so leaving costs you our support, not your system.
Open in the full FAQEach engagement is scoped individually, so there is no price list for these products yet. The fit call is free, and if work can be scoped you get a written price before anything starts.
Open in the full FAQNothing is required in advance. The first step is a free 20-minute call to understand the current setup.
Open in the full FAQAsk who looks after it once it is running and how you would take the work back yourself. Our guide to choosing an automation agency lists the questions, and they apply to us as well.
Open in the full FAQUsually not at the start, because one defined job can be set up and tested for you, with a named person in your team reviewing the results. Everything is set up in accounts you own, so a specialist you hire later could take it over.
Open in the full FAQThese pages distinguish the business process, the implementation service and the AI-specific decision rather than treating them as one offer.
Supporting information: AI consulting: a fixed-scope review of where AI earns its cost, before anything is bought · Choosing an automation agency: questions to ask any supplier, including us · B2B Lead Generation: an approved public-source research and nurture route · Research & Evidence Workflows: source-linked internal preparation for B2B teams · Security and data: how data and access boundaries are approached · Front desk and enquiry handling: missed calls, messages and out-of-hours cover · Discuss one AI-assisted workflow
Describe the trigger, source material, output and person who would approve it. We will assess the non-AI route as well as the AI-assisted one.
With your permission, Google Analytics uses cookies to help us understand how this site is used. We remember your choice for six months; you can change it at any time using Cookie settings in the footer. How Google uses data.
Your choice could not be saved. Please check your browser storage settings.