Renewal Rent Optimization Model

An asset manager setting a renewal rent is choosing a number that changes both cash collected if the tenant stays and the chance they leave. Push rent toward current market and you raise in-place income, but you also raise the odds of a notice, a vacant period, and the cost of finding and installing the next occupier. Hold rent too far below market and you keep the tenancy while leaving money on the table for every remaining year of occupancy. A renewal rent optimization model turns that trade-off into an explicit net present value (NPV) comparison per tenancy. It does not send the offer. It returns a proposed rent (or stays silent) so the asset manager can still decide what to put in writing.

The useful question is not “what is market?” in isolation. Market rent is one input. Retention probability as a function of the offered rent is another. The full cost of a void, including lost rent, void rates or operating costs that continue, reletting fees, incentives, and downtime until a stable successor lease, is the third. When any of those three is missing, the model should output nothing rather than invent a number that looks precise.

The model exists for the moment between a known upcoming expiry or option window and the rent the landlord actually offers. Property-management and valuation systems already hold much of the raw material: current passing rent, lease dates, unit attributes, and often a market-rent estimate from a comparable set or an automated valuation. Products in this class sit next to platforms such as RealPage, Yardi, and Enodo; the point of the model is not which system stores the lease, but whether the rent recommendation is an NPV-optimal figure rather than a round-number uplift or a copy of last year’s increase.

For each tenancy in scope, the model evaluates candidate renewal rents over a bounded range, typically from a floor the landlord will not go below (for example in-place rent, or a policy minimum) up to a ceiling near or slightly above the current market estimate. At each candidate, it estimates the probability the tenant accepts and stays, the discounted cash flows if they stay on that rent, and the expected value of the path if they leave: void, relet, and the successor rent and term the asset manager would underwrite. The recommended rent is the candidate with the highest expected NPV under those paths, subject to any hard constraints the fund or owner has already set (indexation floors, affordable-housing caps, or a decision not to renew at all).

This is a cost-and-value outcome, not a workflow outcome. The model does not draft heads of terms, chase the tenant, or sequence tasks. Those sit in adjacent work: Automated Renewal Heads of Terms for the document once a rent is chosen, and Renewal Pipeline Workflow Agent for dates, owners, and status. The optimization step only answers: given retention, market, and void cost, which rent maximizes expected NPV if the tenant is offered it and then responds as the retention curve implies.

Three inputs are mandatory. If any one is absent, incomplete, or marked as stale beyond the team’s rule, the model returns empty output for that tenancy. It does not fill gaps with portfolio averages presented as facts about this unit.

Current market rent must be a named estimate for the specific unit or a tightly defined comparable set, with an as-of date. A city-wide index or a last-transacted rent from a different building class is not a substitute. If the leasing team has not refreshed market evidence for this product type and location, there is no honest “discount to market” and no honest “stretch to market.” The model should refuse.

Retention probability must be a function of offered rent (and usually of remaining term, tenant type, and how far the offer sits from passing and from market), not a single stay-or-go guess. A lease-break or non-renewal score that does not vary with the rent you would charge cannot tell you whether a £2 psf increase is worth the extra vacancy risk. That risk surface often comes from a Lease Break Probability Model or an equivalent renewal-acceptance curve trained on historic offer outcomes. If that curve has not been produced for this segment, or if historic offers never varied enough to identify sensitivity to rent, the optimizer has nothing to differentiate candidates and must stay empty.

Full void cost must be a cash-flow view of vacancy, not “one month of rent” unless that is genuinely all that is lost. Typical components are: rent and recoverable charges not received during vacancy; unavoidable holding costs; agent or in-house leasing cost; incentives and rent-free on the next deal; downtime to practical completion of any works the next tenant requires; and the probability-weighted delay until a successor is in occupation. If the asset manager cannot state those items for this unit type (or a documented template for this asset), expected NPV of the leave path is undefined.

Optional enrichments improve ranking among valid candidates but do not replace the three: remaining lease term and break dates, tenant credit or sector concentration limits, planned capex that would be triggered by a void, and owner constraints (hold vs. sell, minimum yield, reputational rules on residential affordability). Optional fields may be omitted. Mandatory fields may not be imputed.

Expected NPV at a candidate rent is, in outline: probability of stay times the discounted in-place cash flows at that rent, plus probability of leave times the discounted void-and-relet path. The stay path is usually simple: the proposed rent (or stepped rents) over the assumed remaining occupancy, net of any agreed concessions that are part of the offer. The leave path is where teams understate cost. A “small” increase that looks accretive on a rent roll can destroy NPV if the tenant is price-sensitive and the unit is slow to relet, expensive to hold empty, or likely to relet below the passing rent after works.

The optimum is rarely “always match market” or “always hold passing.” Where retention is steep just above passing rent and voids are long or expensive, the NPV-optimal offer often sits below the market estimate. Where the tenant has few alternatives, the unit relets quickly at or above passing, and void cost is low, the optimum moves toward market. The model’s job is to locate that point on the candidate grid, not to moralize about being “firm” or “tenant-friendly.”

Because the leave path uses an expected successor rent, the market-rent input appears twice: as the tenant’s outside option (which shapes retention) and as the rent the landlord might achieve after a void (which shapes the leave NPV). Those two uses can differ. A sitting tenant’s alternative may be a different building; the landlord’s relet may require a different specification. If the team only has one market figure, it should be documented as serving both roles, and the model should still refuse if that figure is missing. It should not silently use passing rent as “market.”

Do not treat the output as a forecast of what the tenant will pay. It is the rent that maximizes expected NPV if offers follow the modelled retention curve and if voids cost what was input. Changing the void template or the retention curve will move the recommendation. That is correct behaviour, not instability.

Human-in-the-loop is not a review checkbox after an email has gone out. The model proposes a rent (and typically the expected NPV at that rent versus passing and versus market, plus the implied stay probability). The asset manager chooses the figure that enters heads of terms, a letter, or a portal offer. They may match the proposal, shade it for relationship or pipeline reasons, or reject renewal entirely. Nothing in the optimization step authenticates an offer, updates the rent roll, or notifies the tenant.

What the asset manager owes the model is a recorded decision: accepted as proposed, overridden with a reason (relationship, credit, sale process, works programme), or deferred because an input was wrong. Overrides are how you later see whether the retention curve is mis-calibrated. They are not a failure of the tool.