From an attachment to a checked record
Capture an invoice or form, extract the relevant details and validate them. Send uncertain records to a person before the approved data reaches the operational or accounting system.
UK AI automation consulting · Founder-led · Hands-on
AI automation for the work that holds your business up.
I’m Matthew Hutchings, an independent AI automation consultant for UK businesses. Through Fewzen, I help you improve processes, connect systems and build the software needed to make the change work. You work with me directly, from the first question to handover.
The Opportunity Review is free for the first 10 qualifying businesses, by application. Applying does not guarantee selection or availability.
Business process automation, not a tool pitchFollow the work.
Follow the work.
Find the useful intervention.
When staff re-key records, chase approvals or rebuild reports, the answer may be a better process, an integration or a carefully bounded AI step. These are illustrative workflow options—not additional client delivery claims.
From an enquiry to a prepared next step
Interpret an incoming request, find the customer record and prepare a CRM update or draft response. Keep approval with the person responsible for any commitment.
From separate records to an operational view
Connect project, finance or compliance information so teams do not maintain competing copies. Define ownership and reconciliation before adding a dashboard.
AI where it helps. Rules where they matter.
Not every automation needs AI.
AI is useful for variable language, document interpretation, classification or drafting. Conventional software is better for exact calculations, permissions, validation and predictable data movement. A person remains accountable for decisions that need judgement.
Explore conventional systems integration ↗Use conventional automation
When the rules and inputs are known: synchronise approved fields, calculate totals, route an approval or check required values.
Evaluate AI deliberately
When inputs vary: interpret a document, summarise information or prepare a draft. Test quality, costs and the escalation route.
Keep people in control
For ambiguous, financial or customer-binding decisions: prepare the evidence, limit authority and require the right review.
Published delivery evidenceWorkflow automation · Accounting integrations Custom software · Vue · Laravel · AWS A2V · Compliance and project systems
Relevant work you can inspect.
Invoice processing and accounts payable automation ↗
Built invoice processing and approval workflows with PDF extraction and email intake.
Less manual invoice handling and better accounting data flow.
An automotive finance CRM connecting ten APIs ↗
Built an internal CRM integrating ten existing APIs.
A connected internal CRM, supported through production delivery and employee training.
Connecting compliance data with Sage 200 and Monday.com ↗
Created a single source of truth for supply-chain compliance and project data.
A shared source of project and compliance data in place of fragmented departmental workflows.
These are published project outcomes, not a forecast of savings for your business. Each case study identifies the limits of the available evidence.
Fit & boundaries
A real process. An owner. A checkable outcome.
A good fit when
- Your team repeats administrative work across inboxes, spreadsheets or disconnected systems.
- You can involve process owners and provide representative, appropriately handled examples.
- You want a working improvement, not AI for its own sake.
- You are willing to measure results and retain review where mistakes matter.
A different starting point when
- A supported connector or simpler process already solves the problem.
- The task requires exact results but has no reliable validation or review route.
- You cannot provide access, an accountable owner or time to test the change.
- You need a formal security audit or specialist legal/compliance advice; this service does not replace those assessments.
From discovery to handoverSmall enough to review.
01 02 03 04
Small enough to review.
Solid enough to operate.
Discover & design
Map one real workflow with its users. Establish the baseline, data owners, success criteria, risks and the smallest useful scope. Compare simplifying, buying, connecting and building.
Build & connect
Deliver reviewable increments. Connect the necessary systems, put rules around the uncertain steps and keep permissions no broader than the job needs.
Test & evaluate
Use representative inputs and awkward cases. Test duplicate events, unavailable systems and review routes, then compare accepted work and operating cost with the baseline.
Hand over & operate
Document how to use, monitor and recover the workflow. Agree who handles exceptions, how staff are trained and what support is needed before expanding.
Costs & dependencies
Budget for the complete workflow.
Builds are scoped in stages, with scope and price agreed before each stage. Effort depends on system access, API capabilities, data quality, exceptions and approval requirements. There is no universal implementation rate or guaranteed saving published here.
Separate the implementation from platform licences, hosting, model usage and ongoing support. Any estimate should state its assumptions and what your team needs to contribute.
Read the UK automation budgeting guide ↗Security & human review
Set boundaries before giving access.
Agree what data may leave your systems, provider terms and retention, permissions and credential handling. Minimise sensitive inputs and decide which actions require approval. A model’s confident output is not an authorisation to change a record or send a commitment.
Plan logging that helps diagnose problems without unnecessarily exposing records, an exception owner and a manual fallback. The controls need to match the workflow; no blanket compliance or error-free operation is promised.
Need decisions only? Explore AI advisory. Need a full custom build? Explore software development.
Practical answers
Before you take the next step.
Do we need AI to automate a workflow?
No. Known rules, calculations and moving structured data often need conventional software or an existing connector. I use AI where interpreting variable information is useful and its quality can be checked.
How much does AI automation cost?
There is no universal implementation price. Scope depends on the workflow, integrations, data quality, approvals and operating requirements. Builds are scoped in stages with scope and price agreed before each stage; subscriptions, model usage and support need separate consideration.
Do we have to start with an Opportunity Review?
No. The review helps when the opportunity is unclear. If you have a defined brief, we can discuss a sprint, integration or scoped software build directly.
Will confidential data go to an AI provider?
Some AI workflows send data to a provider. Before doing that, agree what can be shared, the provider’s terms and settings, retention, access and the requirements of your organisation. Sensitive information should not be pasted into an initial enquiry.
Can the workflow run without human review?
Only where the task, quality evidence and consequences justify the authority granted. Consequential decisions need appropriate approval; uncertain outputs need a clear route to a person. An impressive demonstration is not proof of safe unattended operation.
What happens after launch?
The delivery needs documentation, operational access and a handover your team can use. Agree monitoring responsibility, the exception owner and any ongoing support separately. Ongoing leadership is available when you need help with priorities or governance.