Integration & automation
The tools are not the bottleneck any more. Billions have gone into models that are, for most business problems, already good enough. What is stopping you is that your data sits in an ERP from 2011, a shared drive of PDFs and a process nobody has written down.
That gap is the work. It is unglamorous, it does not demo well, and it is where the return actually is.
Tell us what you're runningThe hard part
Most automation work stops at the first rung, because it is the only one that does not require writing software. The systems worth automating are rarely on the first rung.
Modern SaaS and anything built this decade. Straightforward, well-trodden, and the only rung most automation shops can reach.
Rarely where the valuable data actually lives.
No API, but the schema is reachable. We read directly, with change-data-capture so we are not polling a production database into the ground.
Read-only by default. We do not write into a system we did not build unless there is a supported path.
Scheduled CSV, XML or fixed-width drops over SFTP or a network share. Common in ERP estates that predate REST entirely.
Unglamorous and completely reliable once the edge cases in the format are pinned down.
Terminal emulators, thick clients, vendor-locked systems. Driven at the interface level, or via whatever export the vendor grudgingly supports.
Fragile by nature. We say so up front, isolate it behind a boundary, and monitor it closely — or advise against it.
What we automate
Each of these has a tell. If you recognise the tell, the work is probably worth automating.
Invoices, purchase orders, delivery notes, claims, applications. Extracted, validated against your existing records, and written where they belong — with anything uncertain routed to a person rather than guessed.
The tell
Someone is currently retyping these.
Inbound email, tickets, forms and requests classified and sent to the right queue on arrival, by the same criteria every time regardless of who is on shift.
The tell
Urgent things sit unnoticed in a shared inbox.
Matching records across systems that disagree, and surfacing only the ones that genuinely need a human decision instead of a full manual review.
The tell
Someone reconciles two exports in a spreadsheet monthly.
Years of documents and records turned into something staff can query in plain language, with the source attached so the answer can be checked.
The tell
The knowledge exists but nobody can find it quickly.
Pulling from the systems of record and drafting the recurring reports, with a person approving rather than assembling.
The tell
A skilled person spends a day a week producing the same document.
Why these die
Automation projects rarely fail at launch. They fail in month four, when a provider deprecates a model, a document format shifts slightly, or an overnight job stops running and nobody notices for a fortnight.
The difference between a demo and something you can depend on is entirely in what gets built around the model.
What we build around it
Name the systems and describe the process that is eating time. We will tell you which rung it sits on, roughly what it takes, and whether it is worth doing — including when the answer is that a rule would serve you better than a model.
Worth reading first: what AI can actually do, and how we work.
Start a conversation