Prospective Tenant Matching Model

When a unit becomes vacant, empty days create cost through lost rent, service-charge leakage, and holding costs. The matching model shortens the search for a plausible next occupier by ranking known tenant profiles against the void specification. It does not send offers or commit the landlord.

The ranker scores each candidate on sector fit, size requirement, covenant quality, and location preference. CRM records and market datasets supply those attributes. The output is an ordered shortlist of warm leads with the reasons they scored well, so a leasing agent can decide whom to approach and in what order.

Matching is a ranking problem. The model does not book viewings, issue heads of terms, or run credit approval. Approach, negotiation, and covenant decision stay with leasing and credit teams. That split keeps relationship risk and audit responsibility on named people, not on a score.

Every useful score depends on two complete sides of a pair.

The void specification describes the empty unit: use class or sector, net internal area or a usable size band, floor or unit constraints, location (building, submarket, catchment), available date, quoting terms if known, and any hard exclusions such as no food use or no medical. Without that spec, there is nothing meaningful to match against.

The prospect side comes from CRM occupier records and from market data on occupiers who are expanding, consolidating, or known to be in the market. Each profile needs enough structure to compare: sector, space requirement or range, location appetite, and a covenant or credit proxy the firm already uses (parent strength, filed accounts, existing landlord references, or an internal score). Incomplete profiles can sit in a research queue. They should not be scored as if missing fields were zeros.

Feature design should stay close to how leasing already talks about a deal. Sector mismatch is usually a hard filter when planning rules or building mix forbid the use, not a soft penalty that still leaves the name on the list. Size is a band with a tolerance, because a requirement near the unit’s area can still fit after plant, cores, or demise quirks. Location can be a ranked list of submarkets rather than a single postcode. Covenant quality is a relative band on the firm’s own scale, not a public rating the model invents.

If either the void specification or the CRM and market profile set is missing, the model returns an empty result and a clear reason. Empty output is safer than a list built from defaults. Leasing then knows to complete the spec or refresh occupier data before expecting a shortlist.

The agent opens the void, reviews the ranked matches, and chooses whom to contact. Typical fields on each row: occupier name, score, sector and size alignment, location overlap, covenant band, last CRM activity, and a short explanation of the top contributing features.

Human-in-the-loop is mandatory at outreach. The model ranks matches. Leasing still approaches. That protects relationships (a poor fit should not receive a speculative email because a score looked high) and keeps the audit trail on a named person.

A practical working sequence:

1. Confirm the void spec is current (area, use, date, exclusions).
2. Run the ranker against the allowed CRM and market universe.
3. Discard rows the agent knows are stale (already in exclusive talks elsewhere, recently rejected this building, covenant below the landlord’s floor).
4. Approach in score order unless a relationship or timing reason overrides.
5. Log outcomes back into CRM: interested, not now, wrong size, wrong location, covenant failed later.

Those outcomes are the feedback the next retrain needs. A match that never converts because the size band was too loose should tighten that band or lower that feature’s weight, not stay buried in a spreadsheet.

Warm does not mean ready to transact. CRM recency, stated requirement date, and existing representation all affect whether a high-scoring occupier is approachable this week. Surface those fields next to the score so the agent is not guessing from the rank alone.

Return no ranked rows when:

• The void has no usable specification (missing sector, size, or location).
• There are no CRM or market profiles in scope, or every profile lacks the fields required to score.
• Hard filters eliminate the entire universe (for example, the unit is industrial-only and every remaining prospect is office).
• Data freshness rules fail (the occupier file has not been refreshed within the team’s agreed window, or the void record is still marked occupied).

The interface should say why the list is empty, in the same language leasing uses: void spec incomplete (no NIA), or no prospect profiles with size and location. A blank screen without a reason looks like an outage.

Do not invent occupiers from name-only market lists. A name without sector, size, covenant proxy, and location is not a profile. Drop it or queue it for research; do not score it. Prefer an empty list over a ranked set of placeholders.

Matching starts once the unit is known to be coming back, not only after keys are in. Parallel work on the same void still matters: condition and dilapidations, notice validity, and move-out paperwork all affect when the unit is actually lettable. Related practice notes: Dilapidations Assessment via Computer Vision, Lease Break Notice Validity Checker, and Move-Out Document Automation.

The cost outcome is void duration and the quality of the next covenant, not marketing volume. A ranked shortlist that produces a few serious conversations early is more useful than a blast to every occupier in the CRM. Track time from vacant (or from known vacate date) to first qualified viewing, and the share of approaches that were already in the ranked set. Those operational measures show whether the ranker is earning its keep without claiming a conversion rate you have not measured.

Governance should stay light but explicit. Access to covenant and CRM fields follows existing data-protection rules. Scores are decision support, not an automated credit decision. If a landlord later asks why a particular occupier was approached first, the explanation string and the agent’s override note should be enough to reconstruct the day.

Retrain on logged outcomes and on spec changes, not on a calendar alone. A new sector mix in the park, or a change in the firm’s minimum covenant, should invalidate old rankings for open voids until the model is run again.