Invoice routing
A supplier emails a PDF. The system reads it, codes the lines, routes it to the right approver by amount and posts the result to Sage or Xero.
AI consultancy · Automation
AI and workflow automation for the work that repeats
Most admin repeats: invoices to route, orders to place, prices to update, reports to build. We turn it into a process that runs the same way every time. Rules take the routine steps, AI reads the messy inputs, and your people keep the judgement calls.
Automation or an AI agent?
Most processes mix three kinds of step, and each needs a different tool.
- Fixed steps are a job for a rule. Routing an invoice by amount or sending a reminder on day three. Rules are cheaper, faster and easier to check.
- Messy inputs are where AI earns its place. A supplier’s email, a scanned form or a question in plain English. AI can read it and work out the next step. When it takes that step itself, it’s called an agent.
- Judgement calls stay with your people. Anything that spends money, speaks to a customer or changes a record waits for a person to approve it.
An agent is useful, and it needs care. We use one only where a rule can’t do the job, and we’ll tell you which is which.
Where automation pays off
The best candidates run often, follow a pattern and take time nobody enjoys spending.
Supplier orders
Fill in and submit a supplier’s web order form from your own system, once and only once.
Price and catalogue updates
Apply a supplier’s price list across your website or shop, with a report of every change.
Stock alerts
When an item drops below its reorder point, the system drafts a purchase order with the preferred supplier for you to approve.
Chasing paperwork
Notice what’s missing, ask for it and file it when it arrives.
Keeping systems in step
Compare two systems, fix what’s safe and explain the rest.
Approvals
Contracts, expenses, holiday requests and change orders, routed by rules you set, with a full audit trail.
Lead handoffs and onboarding
A new enquiry goes to the right person on the rota. A new client gets a project folder, a welcome pack and a kick-off call booked.
Reports and month-end packs
Figures gathered from several systems and ready first thing on Monday. At month-end, the commentary is drafted and left for review.
Example: a supplier price update
Before
- The supplier sends a spreadsheet
- Someone updates hundreds of pages over two days
- Mistakes go live
After
- The agent reads the spreadsheet and matches it to your products
- It proposes the changes with an audit report
- A person approves
A person checks - The agent applies them and refreshes the pages that depend on them
- The report is kept
Guard rails we build in
Only the tools it needs
An agent gets the narrowest permissions that do the job.
Approval where it matters
Money, customer messages and record changes wait for a person.
A budget and an off switch
Every agent has a spending limit and can be stopped instantly.
Every action logged
What it did, when, and what changed, before and after.
Practice first
New agents run on copies of your data before they touch the real thing.
Automation we’ve built
Price updates in three languages
For a client with a website in three languages, an AI agent applies the client’s price spreadsheets across every page and writes an audit report of each change.
A robot that places supplier orders
For a client, a robot fills in and submits a supplier’s web order form from the client’s own system, with a lock so an order can never be sent twice.
Our own coding agent
It works on one task at a time and must pass an acceptance test it isn’t allowed to edit.
What not to automate
Automation costs money to build and to look after, so we’ll tell you which of your processes aren’t worth it.
- One-off jobs. A migration, a year-end task, anything you do once. Doing it by hand is fine.
- Low-volume work. A task that happens five times a month and takes ten minutes adds up to under an hour. The build will never pay that back.
- Work that needs real judgement. A tricky contract, an upset customer, a hiring decision. Keep people in charge where the stakes are high.
- Processes you haven’t settled. If your team can’t agree on the steps, automation bakes the confusion into code. Fix the process first.
The honest test: count how often it happens, how long it takes and how many people touch it, then multiply. If the answer isn’t hours every week, automation is the wrong tool. We estimate the payback before you commit, and show you the sums.
You choose where the AI runs
Tier 1
A hosted model, through a UK or EU endpoint
The best quality for the least effort. Before any of your data goes to a provider, we check that its business terms rule out training on it, and we tell you which region does the processing.
Tier 2
Inside your own cloud account
The model runs in your AWS or Azure account, under your access controls and your logs.
Tier 3
A private model on hardware in the UK
Open-weight models on your hardware or ours. Nothing goes to an outside AI provider. Our own research and document reading run this way.
Storage location and processing location aren’t the same thing. We’ll tell you both, in writing.
Related pages
- AI pilot
Prove one automated job against a pass mark before you build more.
- Integrations
The connections underneath every automation.
- Data and dashboards
Reports and forecasts that build themselves.
- Security, data and responsible AI
Where your data goes, and who checks the output.
Questions
What does it cost?
It depends on how many systems we connect and how complex the rules are. A single workflow, such as invoice routing into Sage, sits at the lower end, from £8,000; work across several systems with approval chains costs more. Where AI does part of the job, we often start with a fixed-price AI pilot on that one job. No hourly billing.
How long does it take?
We agree the timeline with your fixed price, before any work starts. One workflow between two systems is quicker than a change that touches several systems or needs approval chains. You see working software every two weeks, not just at the end.
What stops an agent going wrong?
Narrow permissions, approval steps, budgets, logs and testing on copies first. When a check fails or something doesn’t match what it expects, it’s built to stop and ask a person. You stay in charge of the decisions; we’re responsible for building it to the agreed rules and showing you the logs.
What systems do you work with?
Sage, Xero, QuickBooks, HubSpot, Salesforce, Pipedrive, Monday, Asana, Shopify, WooCommerce, Microsoft 365, Google Workspace and custom databases. If it has an API or a database, we can usually connect it. See integrations.
Can it work with old systems that have no API?
Often, yes. A robot can use a web screen the way a person would. We’ll tell you how fragile that is before you rely on it.
Will it break when our processes change?
We build with change in mind. Rules and thresholds live in settings you can edit yourself. Bigger changes, such as a new system or a new step, we handle as a small change request.
Are you going to replace our people?
No, and we’d tell you if that were the goal. Automation takes the repetitive admin off your team so they can get on with the work they were hired for, including the judgement calls it leaves to them.
We’re not technical. Can we still use this?
Yes. Most of it runs in the background. The parts you do see, such as approval screens and settings, are designed for people who aren’t technical, and we train your team before go-live.
Get in touch
Tell us the manual process you’d hand over first.
Bring the most time-consuming process in your business. We’ll tell you honestly whether automation is the right fix, and where AI would help.