How does Unloq check its own numbers?

Are Unloq’s odds checked?

Yes. Whoever approves a recommendation tends to read its odds as a promise, so Unloq’s odds are checked against your own history, meaning how similar calls have gone in your business, and every set of odds comes with its explanation, the evidence behind it. Odds of 70% should mean that, of all the options Unloq has rated around 70%, about seven in ten succeed. Each recorded outcome is compared with the odds Unloq gave, so a 70% keeps meaning 70% in your business.

How does Unloq show where a number is heading?

As a range with a central point, with the reasons and sources beside it. On a made-up company’s sales, the label might read “−6% to −11% if nothing changes, centred on −9%”. The width of the range shows how sure Unloq is: a narrow range means the history behind it is steady, and a wide one means that history is short or uneven and the figure deserves more caution.

How does Unloq choose which model makes the forecast?

By how close each one has come to what then happened in your business. Several models forecast from your own history, and the one with the best record does the work and has to keep earning its place as new data arrives. We do not name the forecasting models or the measure that scores them, because that is part of how Unloq is built.

How does Unloq measure what a decision was worth?

Against a baseline set before work starts. You and Unloq agree what would have happened without the decision, using your own rules, such as last year’s run rate or a measure your team already trusts, and the result is measured against that baseline. A person signs off every decision, and Unloq records the approval with the outcome you expected, so Prove can later set what happened beside both the expectation and the baseline.

A CFO at an insurance broking group asked us in August how we strip out inflation and the changes that would have happened anyway. The agreed baseline is where that gets settled, before anyone has a result to defend. The baseline itself stays an estimate of how the business would have run without the decision, a version of events that exists only on paper.

Where do Unloq’s checks run out?

Each check depends on your own history. The check on the odds rests on how similar calls have gone in your business, and that record grows slowly, because a business makes only so many decisions of any one kind and the result of each can take a long time to arrive.

The checks have these limits:

• A thin history. When there are too few similar decisions in your business to check a set of odds against, Unloq says so and marks the option as short of evidence, with no number attached. Rare moves, such as a change of pricing model, have the least history behind them.

• Figures from testing. How often Unloq’s answers are right across a whole set of questions is measured in testing. Ask us for our testing figures.

• Actions Unloq cannot see. Unloq measures each decision against a do-nothing baseline, and it can only measure actions it can see. A discount a sales director agreed over the phone and left unlogged cannot be measured, and its effect gets mixed into the results Unloq does measure.

What this page does not cover

Why a number moved, with the drivers ranked by their share of the change, is on Contextualise. On Recommend, Unloq ranks the options for what to do next, each with its odds and the evidence attached, and Prove measures what each approved decision was worth once the results are in. The plan, query, documents and validation grade behind every answer are set out on how every answer is evidenced.

Ask us your question

If your data team wants to test these checks against a decision of its own, ask us your question. Our AI engineering team works the question through with our technical founders and replies with an answer.