SOFTWARE

A marketing site is typically four to six weeks from the first session to launch. A single automation is two to four. What moves that most is not engineering — it is whether the copy, the images and the access to the systems we are integrating with are ready when we start. Projects where those arrive during the build run roughly a third longer, and we will tell you which of the two yours looks like before we quote it.

Websites start at R48 000, automations at R24 000 per process, applied-intelligence work at R64 000, and retainers at R18 000 a month. Those are starting points, not opening positions: the scope moves, the rate does not. You get a fixed price against a written scope before anything begins, and it changes only when you change the scope in writing. Payment is half up front and half on your approval of the finished work, and anything outside the quote is R550 an hour, agreed in writing first.

You do, from the first commit. The repository is created in your GitHub organisation rather than ours, and the domain is registered to you. There is no licence to keep paying for and nothing that stops working if we part ways. If a supplier will not work that way, ask them why.

The people building it. TUMO OLO is a small team, so you brief the developers writing the code rather than an account manager relaying it — which removes both the coordination layer that makes up much of a typical agency invoice and the telephone game that comes with it. The trade-off is real and we would rather say it now: a small team has a ceiling on how much can run at once. If you need four workstreams in parallel, you will hear that in the first call rather than after you have signed.

You get a walkthrough, a written handover and thirty days of fixes for anything that was in scope. After that you can take it entirely in-house, come back to us per piece of work, or add website care from R1 500 a month: hosting and security looked after, backups, monitoring and up to two hours of changes, Monday to Friday, 08:00–17:00. Bigger backlogs go on a retainer.

It is the opposite of it. Roughly a third of the time we look at a problem, the honest answer is that a lookup table beats a model, and we say so before quoting. Where a model does earn its place, it ships with a labelled evaluation set, a confidence threshold that routes uncertain cases to a person, and a deterministic path underneath — so the business does not stop on the day the model is wrong.