Dynamic Rent Pricing at Lease-Up

A lease-up asking rent is the published price for a vacant unit or floorplate, not the rent that eventually appears on a signed lease. In a conventional setup, that price comes from a fixed schedule: one rate per unit type, maybe a handful of view premiums, updated when someone has time to pull comps. A dynamic pricing model replaces that schedule with a proposed asking rent for each unit or floor, using current market comps, vacancy, and demand signals. The model does not publish. Leasing still posts the quote to the listing, ILS feed, and offer letter.

This page is for a leasing manager who owns lease-up velocity and the rent roll that results from it. It covers what the model should output, which inputs it needs, when it must return nothing, and how a human review loop keeps concessions and floors aligned with the asset plan. Adjacent work, such as AI Tenant Screening & Scoring after an application lands, sits downstream of the asking rent. Pricing happens first.

The useful output is a proposed asking rent for a specific unit, or for a floor when units on that floor share the same product and view class. Pair it with an effective date, the unit or floor identifier, the unit type, and a short reason code that leasing can scan: comp move, vacancy stretch, demand lift, or floor bind. Optional fields that help the same review are a recommended concession package (none, weeks free, or a one-time credit) and a band around the proposal so the desk knows how far it can negotiate without a second look.

Fixed schedules fail at lease-up because absorption is uneven. Early demand may support a premium on corner two-bedrooms while stacked studios sit. A single type-level number either leaves money on occupied-ready inventory or stalls the units that need to move. Per-unit or per-floor proposals let the desk raise the asking rent where traffic converts and hold or cut where days-on-market stretch, without rewriting the whole rent matrix.

Commercial tools in this category include Yardi RENTmaximizer, RealPage revenue-management products, and Enodo-style valuation models. Treat them as examples of the same job: a scored asking rent from comps and occupancy, not as a requirement to buy any of them. Whatever the stack, the contract with operations stays the same. The model proposes. The leasing manager, or a designated pricing lead, publishes.

Do not ask the model for a signed lease rent, a renewal offer, or a mid-term bump. Those are different decisions with different comparables and different approval paths. Keep lease-up asking rents in their own queue so a renewal floor does not silently cap a vacant unit that should price to the street.

Three input families belong in every run: market comps, vacancy, and demand. Comps are recent asking and closed rents for comparable units, inside the asset and at competing properties, with unit type, size, finish, floor, view, and lease start date attached. Vacancy is unit-level status plus community occupancy, notice-to-vacate, and days vacant for the subject unit. Demand is whatever the property already records without inventing a new feed: showing volume, application starts, ILS inquiries, and conversion from showing to application over a short window.

Map those fields from the property-management system and the listing stack before you train or tune. If the PMS already stores last asking rent, last concession, and lease execution rent, include them as lagged features so the model can see how far prior quotes moved. If ILS bid prices or paid-placement flags exist, keep them as context for demand, not as the rent itself.

Concessions are part of the asking-rent decision, not a separate afterthought. A $50 higher advertised rent with two weeks free can be cheaper on a net-effective basis than a lower headline with no credit. State the output as advertised rent plus concession so leasing can post a consistent story across the website, ILS, and tour script. Net effective rent can sit in the review UI for the pricing lead; it should not replace the advertised number on public listings unless that is already house policy.

Refresh cadence should match how fast the market and the stack move. Daily proposals are enough for most conventional lease-ups. In-day updates make sense only when vacancy or showing counts change enough to justify another human review. Flooding the desk with hourly changes that leasing cannot publish is noise.

Empty output is a first-class result. If comparable rents are missing, stale beyond the window you trust, or too thin after filters for unit type and finish, return no proposed rent. If vacancy status is unknown, conflicted between PMS and the door inventory, or missing for the subject unit, return no proposed rent. Do not fill gaps with the last published schedule, a community average, or a type-level fallback. Those substitutes recreate the fixed matrix the model is supposed to replace, and they hide the data problem from the people who can fix feeds.

Define “missing” in operational terms. A comp set with fewer than the minimum count you set for that unit type is missing. A vacancy flag that is blank, “unknown,” or disagrees with a work-order that says the unit is not rent-ready is missing. Demand can be sparse on a quiet weekday; sparse demand is not a reason to withhold if comps and vacancy are present. Withhold on demand only when you have chosen demand as a required input and the source system is down.

When output is empty, send a structured skip, not a silent omission. Include the unit ID, the missing input family (comps, vacancy, or both), and a timestamp. Route that skip to the same queue as proposals so leasing sees a hole in coverage instead of assuming yesterday’s rent still applies. Units that skip should keep the last published asking rent until a human either fixes the input or manually sets a new quote. The model does not extend itself.

Lease-up adds a rent-ready constraint. A unit that is vacant but not market-ready (make-ready incomplete, CO pending, photos not live) should not receive a public asking rent from the model. Treat “not rent-ready” as a vacancy-input failure for pricing purposes, even if the occupancy code says vacant.

Leasing publishes. That is the control. A pricing lead or leasing manager reviews the proposal, the reason code, the concession, and the band, then accepts, edits, or rejects. Acceptance writes the asking rent to the listing system, the PMS, and any offer-letter template. Edit writes the human number and stores both the model proposal and the override. Reject leaves the last published rent in place and records why: floor, owner directive, or bad comps.

Floors and ceilings belong in the review layer, not only in the model. Asset-level floors protect debt-yield and comparable-sale assumptions. Unit-type ceilings stop a hot weekend of traffic from pricing a stack off the street. If the model lands on a floor or ceiling, show that bind in the reason code so the desk is not surprised. Do not auto-publish a bind. The same person who owns the rent roll should confirm it.

Concessions need the same gate. If the model recommends weeks free, leasing checks remaining concession budget, brand rules, and what competing properties are advertising this week. A proposal that is right on net effective rent can still be wrong on advertised rent if the owner wants a clean headline.

Document the loop so night and weekend coverage works. Who can publish. Who can override a floor. How long a proposal may sit unpublished before it expires and must be scored again. Expired unpublished proposals should not linger as if they were live quotes.

Work that happens after a prospect applies is a different workflow. Screening and scoring belong on AI Tenant Screening & Scoring. Form of the lease, restricted clauses, and required disclosures belong on Lease Compliance Checker. Term sheets and heads of terms belong on Lease Heads of Terms Generator. Keep asking-rent proposals out of those documents until a human has published the rent.