From data preparation to deployment and monitoring.
We build practical models on the records you already keep, such as demand forecasts, job duration estimates, lead scores and anomaly flags, test each one against what you do today, and send anything unusual to a named person.
Each model is built on what your business already records, and each one hands a person something to act on. None of them makes the decision for you.
How many jobs, calls or bookings to expect each week, given as a likely range rather than one number.
How long a new job is likely to take, from its type, its site and the details on the work order.
Which new enquiries look most like the ones that turned into work, so they get the first call back.
Invoices, timesheets and job costs that don’t look like the rest, put in front of a person.
Where your records support it, we also build recommendations and document classification. If a report or a simple rule would answer the question, we say so before anyone builds a model.
For example, a 60-person plumbing and air-conditioning business keeps five years of jobs in Simpro and wants to know how many service calls to roster for each week of summer. The business is an illustration, not a client.
We review the quality, coverage and suitability of your records, then prepare a dataset for building and testing.
We build a model for the question you agreed: a forecast, an estimate, a score or a flag.
We compare it with an agreed baseline, on records it has not seen, and look hard at the errors that matter to you.
We connect the output to the work and set checks for changing data and performance.
Every forecast comes with its likely range and sits next to the baseline you agreed, such as the same week last year. You can see where the model adds something and where it doesn’t, and when the range is wide, the panel says so instead of hiding it.
Sample data for the example business, not a real one.
Anomaly flags never act on their own. Each one lands with the person you name, with the reason in plain words. Confirming starts your usual process. Dismissing takes a short note. Both are recorded, and the dismissals are read at the next evaluation, so a flag that keeps being wrong gets looked at.
The sample card works. Confirm it or dismiss it and see what is recorded.
Logged hours are well above the estimate range for this job type, and two of Thursday’s timesheet lines overlap.
Sample data for the example business. Names are made up.
A check that trips goes to the operations manager. Retraining, or switching the model off, is her decision.
Four written pieces of work, one for each stage. Timelines and deliverables are confirmed in writing for your project before it starts.
Anything else, ask in the evaluation. A person answers within one business day.
It depends on the task, the quality of your records and the approach. A weekly forecast needs enough history to cover the seasons that matter to you. A lead score needs past enquiries with how each one ended. We assess suitability before recommending any model, and if the answer is not yet, we say so.
No. Performance depends on the problem and the data. Before we build, we agree how the model will be judged and the baseline it has to beat, test it against that baseline, and tell you its limitations before anything is deployed.
In a copy made on terms you agree in writing: which fields, who can see it and what happens to it at the end. The copy is held in Australia and the model is trained in Australia, with no overseas compute. Your records themselves stay in the systems they live in now.
Not on its own. Its output reaches your systems through the same approved connections as everything else in SpeedOz, so anything that changes a record waits for a named person, and every request is recorded, refusals included.
New services, new prices or a new system all change the data a model sees. Monitoring checks for that, and for performance slipping against the baseline. When a check trips, you hear from us with a recommendation, and retraining or switching the model off is your decision.
Not to start. We work from the records in the systems you already use, and a well-kept spreadsheet is a fine place to begin. What matters more is a person on your side who knows what the records mean.
Tell us what you would like to predict and where the records live. A person reads it and replies within one business day, and we will tell you whether your data can support a model before anything is built.
