Tenant Screening for Travel Nurses and Contract Workers: Why Good Renters Get Rejected
A travel nurse begins a 13-week contract at a hospital in Houston on July 1. On August 1 she applies for a furnished two-bedroom near the medical center, and her screening report comes back with four problems. She has been with her current employer for 30 days. Her income is a per-diem stipend, not a salary. Her last tenancy lasted 14 days. Her references are HR coordinators at a staffing agency.
On a standard tenant screening report, each of those is a red flag. Each is also a condition of the job she was hired to do. She is not an unstable applicant but an entirely ordinary mid-term one, judged by a system built for someone signing a 12-month lease.
That mismatch has consequences operators rarely see. Standard screening scores people on how long they have stayed put, and contract workers score badly on that measure by definition, not because they are risky. The applicants it misreads are among the most valuable in the furnished rental business, and turning them away never shows up as a loss on any report, which is why the practice can go on for years.
Accurate data, wrong reading
Tenant screening is the background check a landlord runs before approving an applicant. It usually combines a credit check, a search of eviction records, a criminal records check and verification of income and employment. Many services roll all of it into a single score, a number between roughly 300 and 850, and hand it to the landlord as a recommendation.
A mid-term renter takes a furnished property for roughly 30 days to 12 months. In practice that means travel healthcare workers, contract engineers, consultants on fixed assignments, families between homes and people relocating for a new job. Many of them are paid a per-diem stipend, a daily allowance that covers housing and living costs on assignment. It is real, contracted money, but it is not structured like a salary, and income verification tools often cannot read it.
Nothing is broken in the underlying data. The credit file is the credit file. The eviction search runs the same query it always has, and the criminal check is unchanged. What fails is the interpretation layer that sits on top, the part that decides which facts count as stable.
Why tenure is the wrong signal
Traditional screening treats length of tenure as the master signal, and for a 12-month lease there is a good reason. Staying in the same job and the same home does correlate with paying rent for a year. Long tenure is not valued for its own sake; it stands in for a prediction that the applicant will probably still be doing the same thing next year.
Mid-term renting reverses that relationship without weakening it. A travel nurse's rent is backed not by tenure but by a signed clinical assignment with a defined end date, a contracted weekly rate and a staffing agency legally obliged to pay it. That is arguably clearer evidence than a salaried applicant's employment letter, since in most American states a salaried job can end with two weeks' notice and no reason given.
No field on a standard report holds any of this. The information gets typed into a free-text box or left out entirely, and the score falls back on what it can measure: 30 days. The same thing happens down the line.
| What the applicant actually has | How a standard report reads it |
|---|---|
| Signed 13-week contract with a defined rate and end date | 30 days at current employer |
| Per-diem housing stipend from a staffing agency | No verifiable salary |
| Verified contractor income through Stripe or Deel | Self-employed, unverified |
| Relocation offer letter from a corporate mobility team | Not yet employed at stated address |
| 14-day prior stay with a five-star host rating | Rental history: 14 days |
Everything in the left column is evidence, and several items are better evidence than the fields being scored. None of them has a slot.
A 12-month assumption, everywhere
Once the pattern is visible, it turns up wherever a system quietly assumes a year. Pet screening and pet rent are written for annual tenancies and then applied unchanged to a 30-day stay, with a full annual deposit and a policy calibrated to a year of wear. Two nurses sharing a two-bedroom get recorded as one lease, because two screenings, two signatures, two payment methods and one shared maintenance thread do not fit a system built around a single named tenant; almost no third-party platform supports a lease per bedroom. Renewal reminders fire 90, 60 and 30 days before a lease ends, so on a 30-day stay the first one arrives after the tenant has already left.
Individually these look like minor settings problems. Together they describe an entire category running on assumptions imported from long-term residential renting, without anyone ever deciding to import them.
The law does not treat scores as neutral
It is comfortable to think of the score as mere infrastructure: the vendor supplies a number, the landlord decides, and responsibility sits with the landlord. A federal court has already rejected that argument.
In November 2024, a judge in the District of Massachusetts approved a settlement of roughly $2.3 million in Louis v. SafeRent, a class action alleging that an algorithmic tenant score disproportionately harmed people using housing vouchers. Earlier in the case, the court refused to dismiss the claim on the theory that the vendor could not be liable because landlords made the final decision to accept or deny.
One criticism of the product bears directly on this discussion. The score drew on data designed to predict credit repayment and used it to predict rent payment. Under the settlement, the company agreed to stop showing scores for voucher applicants and to have any replacement score independently validated.
That case concerned housing vouchers and protected classes, not travel nurses, and the distinction matters legally. The structural lesson still carries over. A scoring model built on inputs that do not measure what it is trying to predict is a liability, and the company that supplies it cannot stand outside the decision. An operator who declines qualified mid-term applicants because an old model misreads contract work is not in the same legal position. It is in the commercial version of it, turning away good tenants slowly and at scale, and filing the result under caution.
A cost that never appears
The failure survives because it is invisible by design. A declined applicant does not appear on any report as a loss. There is no line item for the unit that sat empty 11 extra days because the person who would have filled it scored 580 on a model that could not read her contract. The operator sees a decline, then a vacancy, then a later booking, and nothing connects them.
Meanwhile, the applicants being misread are the most valuable pool in the category. Travel healthcare workers, contract engineers on factory and data center projects, corporate relocations and consultants on fixed assignments pay above market rates for furnished units, stay for a defined and predictable period, rotate back through the same cities on their next contract and have an employer or agency standing behind the payment. They are also, without exception, short on tenure and rich in contracts. Screening on tenure filters hardest against precisely the demand this category exists to serve.
What would fix it
More data is not the answer; the market already collects plenty. Several vendors have introduced conversational screening tools that gather documents and produce a recommendation in a single session, though the performance figures they publish are their own and should be read that way.
The missing piece is narrower and much less exciting: a scoring model that reads a contract assignment as employment, with the end date and counterparty recorded; a per-diem stipend as income, verified against the agency; and a rated 14-day prior stay as rental history, weighted by the rating rather than the length. The inputs stay the same. The weights change.
Such a model also needs a report that shows why an applicant fits, rather than handing over a number and leaving the operator to argue with it. An operator who can see that rent is covered 3.4 times over by a contracted weekly rate running through a known end date has enough to decide. One who sees "580" does not, and will decline by default. As of today, no property management system ships this. The data is fine. The reading is wrong.
RentOS, the company that publishes this blog, builds software for stays of 30 days and longer, with applicant records that treat contract assignments, stipend income and short prior stays as evidence rather than free-text exceptions. Book a demo at rrentos.com.
Frequently asked questions
Why do travel nurses fail standard tenant screening? Standard screening uses job tenure and the length of rental history as proxies for stability, and contract work produces short values for both by design. A nurse on a 13-week assignment shows about 30 days with her current employer, a per-diem stipend rather than a salary and a previous tenancy measured in days or weeks. Each looks like a red flag on a model built for a 12-month lease, and each is a normal feature of the job.
Is the screening data itself wrong for mid-term applicants? No. The credit file, eviction search and criminal check are accurate and unchanged. The problem is the layer that interprets them and decides what counts as stability. Evidence such as a signed assignment with a contracted rate, verified contractor income or a highly rated previous stay is often stronger than the fields being scored, but has nowhere to go on a standard report.
Can a tenant screening company be held responsible for its score? A federal court has allowed such a claim to proceed. In Louis v. SafeRent, the District of Massachusetts rejected the argument that the vendor could not be liable because landlords made the final decision, and in November 2024 it approved a settlement of roughly $2.3 million. Part of the criticism was that the score used data built to predict credit repayment in order to predict rent payment.
What should a mid-term screening model measure instead of tenure? It should measure the structure of the contract rather than continuity. That means the signed assignment and its end date, the contracted rate and who is paying it, the ratio of that rate to the rent, and rated previous stays regardless of length. The aim is not to lower the standard but to score the evidence that actually predicts payment across a stay of 30 to 180 days.
How do I screen a contract worker if my software has no field for it? Until the tools catch up, verify the assignment directly. Ask for the signed contract or an agency confirmation showing the rate and end date, confirm the paying agency, calculate rent as a multiple of contracted income rather than salary, and ask previous furnished hosts for a reference instead of relying on the length of a tenancy. Record all of it against the application so the decision is documented.
Why does declining these applicants cost me money if I never see it? A decline never appears as a loss anywhere, and there is no report line for a unit that stayed empty because the applicant who would have filled it was misread. The pool being filtered out, contract healthcare workers, project engineers and corporate relocations, is also the pool that pays above market for furnished units and returns to the same markets on later contracts.