AI Strategy

Harla uses AI across three layers of the property operations workflow.

At the extraction layer, AI interprets incoming documents — lease agreements, heads of terms, compliance records, KYC packs — and converts unstructured, inconsistently formatted content into structured, queryable data. This is the foundation of everything else: without reliable extraction, neither portfolio intelligence nor transaction automation is possible. AI handles the variance in document format, layout, and language that makes manual data entry slow, expensive, and error-prone at scale.

At the intelligence layer, the system surfaces what the data means for portfolio performance. Void forecasting, re-letting strategy, counterparty benchmarking, and obligation tracking are all derived from the structured data accumulated across every transaction. AI identifies patterns and anomalies — a lease approaching a break clause, a counterparty with a history of slow execution, a compliance gap that has appeared across multiple properties — before they become problems requiring reactive management.

At the automation layer, AI drives the workflows that move transactions from first contact to cash. It orchestrates documentation, coordinates actions across multiple parties, drafts outbound communications with portfolio-specific context, and populates compliance records. This removes the administrative coordination load that currently makes commercial property operations expensive to scale and slow to execute.

Together, these three layers mean a lease or asset manager using Harla handles a larger portfolio with less manual effort, a more complete compliance record, and faster execution from heads of terms to completion.

Where a capability is commodity — language understanding, document OCR, general text generation — Harla buys via API. Foundation model providers have invested at a scale no early-stage company can replicate, and the marginal cost of API access is falling consistently. Competing on general language capability is not the strategy.

Where defensibility comes from proprietary data, Harla builds. The knowledge of how a specific portfolio's transactions execute over time — which counterparties move quickly, where compliance gaps recur, how void duration correlates with lease type and market conditions in the UK commercial sector — is not available in any foundation model and cannot be purchased. It is accumulated through operational engagement with real portfolios.

The goal is a thin, fast API layer on top of foundation models, with a proprietary data and retrieval layer underneath that compounds in value with every transaction processed. The model is the commodity; the operational graph built on top of it is the product.

Each portfolio onboarded adds structured execution data: lease timelines, void duration records, counterparty performance histories, compliance gap patterns, and HoT-to-completion benchmarks. This data trains and continuously improves Harla's forecasting, routing, and anomaly detection capabilities in ways that become increasingly specific to UK commercial and residential portfolios over time.

The flywheel is self-reinforcing. More transactions improve void forecasting accuracy. Better void forecasting improves re-letting strategy recommendations. Improved recommendations deliver measurable NAV outcomes for operators, which drives further adoption and further data accumulation.

Over time this creates a proprietary dataset that a new entrant cannot replicate without acquiring a comparable base of operational customers — and by that point, Harla's models will already be substantially more accurate for UK portfolios than any general-purpose tool. The moat is not the model; it is the operational graph built on top of it, and the time required to build that graph is the barrier to replication.