Most of what gets sold as AI should be a rule. We tell you which is which.
We have been automating businesses since long before a model could write a sentence, and we run our own software on the results. That is the useful part: we are not selling you AI, we are selling you the hours back. Sometimes the answer involves a model. More often it is an integration and four rules, and that version is cheaper to build and does not drift.
The first thing we do is work out whether your problem needs judgment. If the task follows rules that you could write down on a page — move this record when that field changes, send this when that is signed, sync these two systems — then it is ordinary automation, it costs less, it runs the same way every time, and it never invents an answer. That is our business automation service, and for a lot of businesses it is the honest recommendation.
AI earns its place at the point where the task needs judgment: reading messy input that arrives in fifty different formats, deciding what a customer actually wants, pulling the relevant clause out of a document, drafting something a human then approves. That is genuinely hard to do with rules, and it is where a model pays for itself.
Enquiry intake that reads what came in, classifies it, routes it and drafts the first reply. Document processing that pulls structured data out of invoices, quotes and forms. Customer service triage that answers the repetitive eighty percent and escalates the rest with context attached. Internal search across the documents your team keeps asking each other for. And AI features built directly into custom software, where the model is one component of a system rather than the whole product.
A demo is easy. A model that behaves on the tenth thousand input is the actual work, and it is mostly unglamorous: deciding what happens when the model is wrong, keeping a human in the loop where being wrong is expensive, testing against real examples from your business rather than the happy path, and measuring the error rate honestly before it touches a customer.
There is also a running cost, which rules-based automation does not have. Every task a model handles costs a fraction of a cent to a few cents. That is almost always trivial next to the labour it replaces, but it is a real line item and it belongs in the numbers before you commit, not after.
Three quick questions and we'll come back within one business day with a clear next step and a fixed price. If it turns out we are not the best people for it, we will point you at who is.
Start here →Our clients see measurable growth because we focus on strategy, not just execution.
A focused automation handling one workflow typically runs $6,000 to $20,000 to build. Broader systems that touch several processes run $20,000 to $50,000. On top of the build there is a running cost for model usage, usually $50 to $500 a month depending on volume. We quote a fixed build price before work starts and show you the running cost estimate alongside it.
Regular automation follows rules you could write down — it is cheaper, it runs identically every time, and it never invents anything. AI automation handles work that needs judgment, like reading messy input or drafting a reply. We scope the rules-based version first, because for a lot of businesses that is the whole answer and it costs less.
It will, occasionally, which is why that is a design question rather than a hope. For anything where being wrong is expensive we keep a person in the approval step, so the model drafts and a human sends. For low-stakes work it runs unattended with a confidence threshold that escalates anything it is unsure about. We measure the error rate on your real data before it goes anywhere near a customer.
Often not, and we will say so. If your bottleneck is that two systems do not talk to each other, that is an integration. If it is that nobody follows up on quotes, that is a workflow. Both are cheaper and more reliable than adding a model. We look at where your hours actually go before recommending anything.