Ten areas, explained in plain English first. If you already know
exactly what you want, the technical detail is there too — but you
should not need it to work out whether we can help.
AI & Automation
Put AI where it saves real time — and nowhere else.
Most businesses have a handful of jobs that eat hours every week: reading documents and typing what they say into another system, answering the same customer questions, moving data between tools that will not talk to each other. Those are the jobs worth automating. We work out which ones actually pay for themselves, build them, and leave the rest alone.
Technically:
LLM-backed workflows, retrieval over your own documents, structured extraction with validation and human review steps, scheduled and event-driven pipelines, queue-based workers with retries and dead-letter handling.
Typical examples
Pull key fields out of invoices, quotes or contracts and file them automatically
Draft first-pass replies to routine customer emails for a person to approve
Summarise and route incoming enquiries to the right person
Answer internal questions from your own documentation, with sources cited
Off-the-shelf software makes you change your process to fit it. Sometimes that is fine. When it is not — when the workaround spreadsheet has become the real system — a purpose-built application is usually cheaper than the accumulated cost of working around the wrong one.
Technically:
Server-rendered and API-driven applications, domain modelling, background processing, role-based access control, audit trails, reporting, automated test suites.
Typical examples
Internal tools that replace a fragile spreadsheet
Job, order or inventory tracking that matches your actual workflow
Quoting and estimating tools with your own pricing rules
AI that runs on your hardware and keeps your data yours.
If your work involves client records, medical information, legal documents or anything else you would not paste into a public chatbot, you can run capable AI models on your own machine or server instead. Nothing leaves your network. No per-seat subscription, no usage meter, no vendor reading your documents.
Technically:
Local LLM deployment and quantisation, GPU and CPU inference sizing, private vector search over your own corpus, containerised inference stacks, air-gapped deployment where required.
Typical examples
A private assistant that only knows your documents
Document search across years of files, on your own server
On-premise drafting and summarising for confidential material
Applications your team and your customers actually use.
Not a brochure site — a working application. Somewhere your customers log in to see their own information, or your staff log in to get work done, on any device, without training.
Technically:
Authentication and authorisation, server-rendered and progressive interfaces, responsive layouts, accessibility to WCAG 2.2 AA where practical, caching, and a straightforward deployment story.
A lot of businesses have the same problem: the real numbers are spread across three spreadsheets, an old accounting package and someone's inbox. Nobody is sure which copy is right. We design the structure, get the data in cleanly, and make it possible to answer a question once rather than reconciling it every month.
Technically:
Relational schema design, normalisation with deliberate exceptions, constraints and indexing, migrations under version control, ETL and import pipelines with validation, backup and restore procedures that have actually been tested.
Typical examples
Consolidate records scattered across spreadsheets and systems
Design a schema that will still make sense in five years
Make the systems you already pay for talk to each other.
You already have tools that each do their job. The expensive part is the person copying data between them. Integration removes that person from the loop — and with them, the typos and the forgotten updates.
Technically:
REST and webhook integrations, OAuth and API key handling, idempotent sync with conflict resolution, retry and backoff, rate-limit handling, and adapters for systems whose APIs are best described as characterful.
Typical examples
Sync orders, customers or inventory between two systems
Push data into accounting software automatically
Connect a website form to your internal tooling
Build an API so someone else's system can talk to yours
Stop doing by hand what the computer should be doing.
Every business has a process that runs on someone remembering to do something. Those are the ones that break when that person is on holiday. Automating them is usually less about clever technology and more about writing down what actually happens, then making the boring parts happen on their own.
Technically:
Workflow orchestration, scheduled jobs, event-driven triggers, approval chains with audit trails, notifications and escalation, exception reporting for the cases that still need a human.
Typical examples
Automatic reminders, follow-ups and escalations
Approval workflows with a record of who approved what
Scheduled reports that arrive without being asked for
Exception reports that surface only what needs attention
Old systems are not automatically bad — a system that has run the business for fifteen years clearly does something right. The risk is when nobody left understands it, it will not run on current hardware, or it cannot be secured. The answer is rarely a big-bang rewrite; it is usually understanding it properly, then replacing it in pieces while it keeps running.
Technically:
Codebase and schema assessment, incremental strangler-pattern migration, data extraction from legacy stores, compatibility layers, containerising applications that predate containers.
Typical examples
Get an old application running on supported infrastructure
Extract and preserve data from a system being retired
Run it on your own infrastructure, and own it outright.
Subscriptions add up, and they keep adding up. For a lot of tools there is a self-hosted equivalent that runs on a single server you control. You pay for the setup once instead of per user per month forever, and your data stays on your hardware.
Technically:
Docker and Docker Compose deployments, reverse proxy and TLS termination, persistent volumes, health checks, backup and restore scripts, upgrade procedures, and documentation written for whoever inherits it.
Typical examples
Self-hosted alternatives to recurring SaaS subscriptions
Containerise an existing application for reliable deployment
Set up a server that you own and we document
Migrate from a hosted service onto your own infrastructure
Someone who knows the system, on hand when you need them.
Software is not finished when it ships. Dependencies need updating, backups need checking, and occasionally something breaks at an inconvenient moment. A support plan means the person who fixes it already knows how your system works.
Technically:
Monitoring and alerting, dependency and security updates, backup verification, performance review, incident response, and a documented change process.
Typical examples
Keep an existing system patched and healthy
Verify that backups actually restore
Small changes and improvements as needs shift
A known point of contact when something goes wrong
You do not need to know what to call it. Describe what is going
wrong and we will work out which of the above — if any — is the
right answer.
The same information gets typed into three different systems.
Integration or automation removes the retyping, and with it the typos and the version that never got updated.
The business actually runs on a spreadsheet nobody dares touch.
A purpose-built application with real validation, proper access control and a history of who changed what.
Staff spend hours on work a computer could do.
Automate the repetitive part, keep a person on the judgement calls, and measure what it gave back.
We want to use AI but cannot send our data to a third party.
Private AI running on your own hardware. Nothing leaves your network.
Our software subscriptions cost more every year.
Where there is a credible self-hosted equivalent, we will tell you — and where there is not, we will say that too.
The system works, but the person who built it is long gone.
Assessment, documentation, and incremental modernisation that does not require switching everything off.
We need reports, but pulling them together takes a day.
Get the data into one structure, then let the reports build themselves.
We are not sure whether AI would help us at all.
That is a fair question and a short one to answer. Sometimes the honest answer is that conventional software is the better tool.
Tell us what you are trying to fix
The first conversation is thirty minutes, free, and has no pitch in it.
We will tell you whether we can help, roughly what it would cost, and
if the honest answer is that you do not need us, we will say that too.