AI where it earns its place.
On the reading order, not the decision.

We build AI into systems where it removes a specific piece of human drudgery: reading a document, drafting a first version, deciding what a person should look at first. We do not build AI that makes the decision, because that is both worse and harder to defend.

Start a project

Where it actually works

Reading documents nobody wants to read

Invoices, statements, CVs, reports. Extracting the fields, summarising against what you care about, and flagging the ones that need a human. This is the clearest return we see.

Putting the queue in the right order

Forty applications ranked so the five worth reading first are at the top, with the reasoning shown. The recruiter still decides; the afternoon is just no longer spent sorting.

Drafting, never sending

A first draft of a reply, a description, a summary, with a person approving it. Anything that goes out under your name should have been read by somebody who works for you.

Answering from your own content

A question answered out of your own documents, with the source shown, rather than a model’s general knowledge about your industry.

How we build it so it can be trusted

Every call’s cost is logged

Tokens and cost per call, on the record. AI features that nobody is metering are how a small monthly bill becomes a surprise, and how a runaway loop goes unnoticed.

The output is checkable

Where a figure is extracted, the source is kept so a person can see where it came from. An unsourced number from a model is worse than no number.

It fails to a human, not to a guess

When confidence is low or the service is down, the work routes to a person with a plain explanation rather than inventing something plausible.

Your data stays yours

Keys are per project and never shared between clients, prompts carry only what the task needs, and what is sent is a decision rather than a default.

What we will talk you out of

A chatbot on a brochure site

Almost nobody wants to chat to a website with six pages. A clear phone number and an honest FAQ outperform it, and the FAQ also feeds search.

Automated decisions about people

Hiring, credit, eligibility. The regulatory and reputational exposure is real and the benefit over ranking-plus-human is marginal.

AI-written content at volume

Google’s own spam policies target content generated primarily to manipulate rankings, and the search quality guidelines score originality. Pages written to fill a sitemap do not earn anything.

Questions

Before you ask.

Will our data be used to train a model?

Not by default on the arrangements we build, and it is a question we check per provider rather than assume. We send only what the task needs and we say what that is.

What does it cost to run?

Per call, and we log every one with its cost so you can see it rather than discover it. For most business workflows it is pennies a document; we will measure it against your real volume before you commit.

Can it make decisions for us?

It can rank, summarise and draft. We will not build it to decide about people, and we will say why rather than quietly agreeing.

Does adding AI help our search rankings?

No. Google states plainly that there are no additional technical requirements, no special markup and no AI text files needed to appear in its AI features, and that the content itself is what earns the citation. Anyone selling you AI markup for rankings is selling the absence of a thing.

Ready to get started?

Tell us what you need. We will scope it, price it and build it.

Start a project →

AI where it earns its place.On the reading order, not the decision.