How PropTraka Predicts Which Tenants Are About to Leave

The most expensive month in any rental portfolio is the month a unit sits empty. There is no rent, there is usually a cleaning and repair bill, there is often an agent fee, and there is the time you spend showing the place. A single turnover can wipe out a good part of a year's margin on that unit.

And turnover almost never happens without warning. The tenant who leaves in November was usually visible in August — paying a little later each month, raising more issues than before, sitting on a lease with three months left and no conversation started about renewal.

The problem was never the absence of signals. It was that nobody has time to read them across twenty tenancies, every month, by hand.

That is the job PropTraka's turnover prediction does.

The Short Answer

PropTraka scores every active tenancy from 0 to 100 for turnover risk, using six signals it already holds. The score maps to a risk level — low, moderate, high, or critical — and a predicted vacancy window ranging from within 3 months to 12+ months. Alongside the score you get the specific factors that produced it and a recommended action.

You will find it in the Future Insights section of your dashboard.

It Is Arithmetic, Not a Guess

This matters more than anything else in this article, so it is worth being blunt about it: the score is a calculation, not a language model's opinion.

Each of the six signals contributes a fixed number of points under stated conditions. The same tenancy with the same data produces the same score today, tomorrow, and next year. You can see which factors fired and what each contributed.

That is a deliberate design choice. A prediction you cannot interrogate is a prediction you cannot act on, and a number that changes when nothing changed is not information — it is noise. When ARDO™, PropTraka's analysis layer, discusses turnover risk with you in plain language, it is reading these computed scores. It is not inventing them.

The Six Signals

Lease end proximity. The strongest single factor, worth up to 30 points. A lease ending within three months carries the full weight; within six months, less; within twelve, less again. This is not sophisticated, and it does not need to be — the majority of turnover happens at lease end, and the majority of preventable turnover is preventable precisely in the window before it.

Payment discipline. Worth up to 25 points, scaled directly from the tenant's payment discipline score. A tenant at 40 out of 100 contributes substantially more risk than one at 90. This is a measure of pattern, not of a single late month.

Income volatility. Worth up to 15 points. Where PropTraka holds an income volatility reading, unstable income raises risk — a tenant whose earnings swing is more likely to reach a month where the rent does not work, regardless of intent.

Tenure length. Worth up to 10 points. A tenancy under six months old carries elevated risk, and one under a year carries some. The first months are when a tenant discovers whether the place actually suits them.

Late payment frequency. Worth up to 20 points, scaled by the proportion of rent payments recorded as overdue. Where the discipline score measures the tenant, this measures the specific tenancy — and the two do diverge, which is often the interesting part.

TrustTraka™ level. Worth up to 10 points, where a low score raises risk.

Add them, clamp the total to 100, and that is the turnover risk score.

Reading the Output Honestly

76 and above is critical — predicted vacancy within 3 months. 51 to 75 is high — within 3 to 6 months. 26 to 50 is moderate — 6 to 12 months. Below 26 is low — 12 months or more.

Every prediction also carries a confidence figure, and this is the number most worth understanding. It is simply the share of the six signals that were actually available for that tenancy. Three usable signals out of six gives 50 per cent confidence.

Confidence is not a measure of how sure the model is. It is a measure of how much it had to work with — a deliberately unglamorous statement about the completeness of your data. A high risk score at 33 per cent confidence is a prompt to look at the tenancy yourself, not a verdict.

The practical consequence: the more complete your records, the more the predictions are worth. Confidence is the honest feedback loop on your own data hygiene.

Consent Is Respected in the Arithmetic

If a tenant has opted out of TrustTraka, that signal is skipped entirely. PropTraka does not recompute a withdrawn score, does not substitute a neutral placeholder, and does not mention it in the risk factors. The signal is dropped, and the confidence figure falls to reflect that one of the six inputs was genuinely unavailable.

This costs the prediction some accuracy, which is the correct trade. A consent choice that quietly makes no difference to the output is not a consent choice.

The same principle governs what reaches ARDO. Tenants who have not consented to that lane are filtered out before any name or risk factor is assembled into a prompt, so their personal information never reaches the model at all.

When a Unit Is Already Empty

For vacant properties, a separate estimator projects how many days it should take to fill, with a best case and a worst case.

It starts from a baseline expectation for the property's area, then adjusts for the single factor most within your control: how the asking rent compares to market rent. Ask meaningfully above market and the expected vacancy stretches considerably. Ask below it and the unit fills faster.

A word on the baselines: they are starting assumptions built into the model, refined by your property's own data. They are not a published survey of Kenyan letting times, and they should not be quoted as one. The value is in the comparison — this unit against this asking price — not in the absolute day count.

What To Do With a High Score

The recommendations come with the prediction, and they are ordinary property management, which is rather the point:

A lease ending soon with no renewal conversation started is a conversation to start this week. Deteriorating payment discipline is a reason to offer M-Pesa auto-debit or move the due date closer to payday, before the tenant concludes the arrangement is not working. Short tenure with early friction is a check-in. Rent significantly above market on a tenant with options is a renewal negotiation you would rather open than receive.

None of this is exotic. The advantage is not that the recommended action is clever — it is that you are doing it in August, for the tenancy that needed it, instead of finding out in November.

Turnover predictions are decision support, not certainty. They tell you where to look first. The conversation with your tenant is still yours to have.