Capability

AI and automation

Used where it genuinely beats ordinary software, and not used where it does not.

Book a consultation

A lot of what is sold as AI is a scheduled job

The current wave has made it easy to sell almost anything by attaching the word AI to it. Plenty of what gets pitched is a rules engine, a scheduled job, or a database query, all of which are cheaper, faster and far more predictable than a language model.

The reverse mistake is just as expensive: reaching for AI on a problem where the data is not good enough to support it, then discovering after the build that the output cannot be trusted for anything that matters.

What you get

The deliverables, specifically.

An honest sort of the candidates

Which of your problems are genuinely suited to AI, which are ordinary automation wearing the label, and which are neither. The second group is usually the largest and the cheapest to fix.

Work on unstructured input

Documents, emails and free text are where language models actually earn their cost: extracting fields from invoices or contracts, classifying and routing incoming messages, summarising long threads.

An assistant that writes to your systems

A question-answering assistant is only worth having if what it learns goes somewhere. Conversations should land in your CRM as records, not evaporate inside a chat widget.

Guardrails and a human in the loop

Confidence thresholds, escalation to a person on anything uncertain, and a log of what the system decided and why. Anything touching money or a customer commitment keeps a human decision point.

How it runs

From first call to handover.

01

Sort the candidates

Separate genuine AI problems from ordinary automation and from things not worth doing at all.

02

Check the data honestly

If the input is inconsistent or incomplete, that gets fixed first. A model cannot repair bad data.

03

Build the smallest useful version

One process, in production, measured against how it was done before.

04

Instrument and review

Log what it decided, review the errors, and expand only once the first one has earned it.

Is this the right piece of work?

A good fit when

  • Staff spend hours reading documents or emails and typing what they find into a system
  • Enquiries arrive faster than anyone can triage and route them
  • You have been quoted for an AI project and want an independent view before committing

We will say no when

  • The process is fully structured and rule-based. Ordinary automation will be cheaper, faster and more reliable, and we will say so.
  • The underlying data is not trustworthy yet. Fix that first, or the output will be confidently wrong.
  • You want AI because the board asked for AI. That is a bad reason and an expensive one.

The assistant on this website answers visitor questions from a knowledge file we control, and logs every conversation to our dashboard as a record. It is a deliberately small example of the pattern, running in production on our own site.

Other capabilities

Software Testing & QA

Automated testing that tells you something true, on the paths where being wrong is expensive.

Web Application Development

Web applications built to work on a mid-range phone on a bad connection, because that is how most South Africans will use them.

Cloud & Infrastructure

Deployments that are small, cheap and unremarkable to run, because infrastructure should be the least interesting part of your system.

Not sure this is the right starting point?

A short call costs nothing and usually makes the answer obvious. If a different piece of work suits you better, we will say so.

Book a consultation