What is a probation period for an AI worker?

A probation period for an AI worker is a fixed period at the start of a deployment in which the worker does the real job, in your own systems, and is measured against targets agreed in writing before it starts. At The AR Dept it lasts 90 days. At day 90 the results are set against the KPIs in the worker's Agent Spec. At day 90 you can walk away with no ongoing commitment, whatever the figures show. If you stay, it rolls into a standard 12-month managed service.

Why does an AI worker need a probation?

Employers use probation because an interview cannot show how somebody does the job. The same is true of AI. A demonstration shows what the technology does on a good day with tidy examples. It does not show what happens with the tenant who sends half a form, the supplier whose invoices never match the order, or the week when three people are off and everything arrives at once.

The only fair test is the job itself, for long enough to take in a month end, a busy week and the awkward cases. It means the decision to commit for a year is made on evidence from your own business, counted in your own figures, rather than on a pitch.

What is agreed before the first day?

The terms are written down first, in the role specification, and signed before the worker starts:

• The job. What the worker does, step by step, in which systems, and what it never does.

• The KPIs. A small number of measures in the business's own terms. For a letting agency running tenant referencing that might be days from application to completed reference. For a wholesaler it might be orders entered correctly on the day they arrive.

• The volumes. How much work a month both sides expect, so that the KPIs are set for the job as it really is.

• Who checks its work. A named person at your business who approves what needs approving and hears about anything the worker cannot settle.

What happens during the 90 days?

From the first day the worker drafts, chases, files and reports. Actions with consequences, such as sending an email in your name or changing a compliance record, wait for a person to approve them. Everything it does is logged, so when a result looks wrong you can see what it did and why.

Through the 90 days there are regular reviews with us, where the KPIs are looked at together. If a measure is being missed, that is ours to fix inside the probation. The fix might be a rule tightened, a piece of wording rewritten, or a step handed back to a person because it needed judgement after all. A miss in the third week is what the probation is there to find.

Your team has a part in it too, as they would with a new starter. They use the worker, they check what it sends for approval, and they say when something is wrong. What they report is what we fix in the reviews, so a problem nobody mentions can go unfixed.

What happens if the AI worker does not work out?

On day 90 the worker's figures, which come from your own business, are set against the KPIs you signed. A measure being missed is ours to fix inside the probation, and at day 90 you can walk away with no ongoing commitment, whatever the figures show. If you stay, it rolls into a standard 12-month managed service.

The date and the measures are both fixed before the work starts, so the worker is judged on what it did in the job.

How is a probation different from a free trial or a pilot?

• A free trial. The trial period for an AI worker is the probation, and it is not free. Your first AI worker starts on a 90-day probation contract: the worker is built for your role and does live work from the first day. For the price, the homepage has our starting monthly fee; a one-off build fee comes first.

• A pilot or proof of concept. A pilot usually asks whether the technology works at all, often on sample data and beside the real job. A probation asks a harder question: does this worker do this job, to this standard, in live work?

• A demonstration. A demonstration is run by the seller. A probation lasts three months and is judged on your numbers.

• Probation in employment law. The word is borrowed from employment law; the law is not. An AI worker is not an employee and has no employment rights. The idea of a standard set in advance, a period of real work, and a date on which the results are judged carries over.

What should you have ready before one starts?

Know how the job performs today, because the KPIs are set against it. If you cannot yet say how long the job takes now, or how many arrive in a month, measure that first: it costs nothing and it will be the first question in any review. Then choose one named person who approves what the worker sends for approval and joins the reviews. Describing the job itself is the work of the role specification, and our explainer on that covers what to bring.

Your first AI worker starts its working life on this probation; a worker added later joins without one. Our explainer on what AI staff are sets out the rest of the model.

In your business that would be one defined job done in your own systems for 90 days, against KPIs you have signed, with a named person of yours approving anything that has consequences. If you would put one job on probation tomorrow but not sign for a year today, talk to our agent about which job it would be.