Using AI to spot tomorrow's underperforming assets today
A new approach to value-add sourcing in UK commercial real estate
Executive summary
AI in real estate is often sold as marketing hype. In reality, used properly, it's a sourcing weapon.
Consultants estimate that generative AI and advanced analytics could unlock $110–180bn of annual value for the global real estate industry, largely by improving how data is used across the investment lifecycle. Research also suggests more than a third of real-estate operations could be automated with AI, and AI-enabled PropTech firms attracted about $3.2bn of capital in 2024 alone.
Against that backdrop, UK commercial property faces a slow-moving but very real shock. The government's Minimum Energy Efficiency Standards (MEES) framework is converging on a minimum EPC B requirement for commercial lettings around 2030, with interim EPC C milestones likely to fall between 2030 and 2035. Multiple analyses suggest the sector is on track to miss these targets by up to a decade, with a large proportion of stock still rated C or below—especially in secondary offices. Industry bodies have warned that, if applied strictly, 70–80% of UK commercial buildings could become unlettable on energy grounds alone.
In that environment, "underperformance" isn't just about today's NOI; it's about future regulatory, leasing and liquidity risk that is not yet priced in.
The new definition of "underperforming"
Historically, an underperforming asset was obvious: high vacancy, weak rents, poor NOI trend, dated specification. Today, many of the real value traps are still fully let and apparently fine—but quietly accumulating risk.
For UK investors, tomorrow's underperformers typically share four traits:
1. Regulatory risk
EPC ratings of C or below, where capex to reach B may be substantial and technically complex. Assets likely to fall foul of tightened MEES rules between 2027 (C) and 2030 (B), after current leases expire.
2. Leasing & income risk
Clusters of lease events in the next 3–5 years in sub-markets with elevated availability. Tenants in sectors with elevated default risk or footprint rationalisation.
3. Functional obsolescence
Secondary offices in "fringe" locations that lack amenity, ESG credentials or flexibility, in a market increasingly dominated by modern "flight-to-quality" space.
4. Liquidity risk
Assets that will screen poorly in lender or institutional ESG scoring models, making future refinancing or exit harder and more expensive.
The challenge is that none of these show up cleanly in a single data field. They emerge only when you connect multiple datasets—the point where AI and machine learning genuinely add value.
Where AI and data add real value
There is understandable scepticism. Many AI projects in corporate settings struggle to deliver tangible ROI. But in commercial real estate, several applications are now mature and proven:
- Predictive analytics for asset performance – models that use rent, occupancy, NOI, cost data and local market indicators to flag assets likely to underperform peers, or to identify hidden outperformers.
- Rental income and cashflow forecasting – platforms reporting forecast accuracy improvements of up to 60% versus traditional spreadsheets when using AI to incorporate more granular, real-time inputs.
- Portfolio optimisation – machine-learning models that test scenarios across thousands of assets and market conditions, identifying where to dispose, refinance or reinvest to improve risk-adjusted returns.
For Fenrir and similar UK investors, the practical takeaway is: AI is most useful upstream—in sourcing and triage—not as a replacement for detailed underwriting or human judgment.
A four-layer framework
Fenrir can think of AI-enabled sourcing as four layers, each building on the last:
1. Data foundation: build the 360° view
The starting point is to combine internal and external data into a single asset-level spine. Typical inputs include:
- Internal asset data: rents, incentives, ERVs, WAULT, void history, service charges, opex, capex, EPCs
- Lease & tenant data: expiry dates, break options, sector, covenant strength, arrears
- Market data: sub-market rents & yields, vacancy, absorption, pipeline, business-rates changes
- Regulatory & ESG: EPC trajectories, MEES deadlines, local planning constraints, retrofit feasibility
- Location & amenity: travel times, PTAL scores, local amenity density, crime stats
- Alternative use signals: planning applications nearby, demographic shifts, catchment areas
2. Descriptive diagnostics: "What's really going on?"
Once data is stitched together, the first step is benchmarking, not prediction. For each asset, AI can help:
- Highlight outliers on cost per m², service charge intensity, or energy use compared with local peers
- Flag assets at greatest MEES/EPC risk based on rating, lease profile and retrofit complexity
- Identify sub-markets where supply/demand balance is quietly deteriorating
3. Predictive models: "Where will stress actually show up?"
Building on that baseline, machine-learning models can be trained to generate forward-looking risk scores at asset or micro-location level:
- Vacancy and re-letting risk – probability of an asset being above a certain vacancy threshold at a given point in time
- EPC and capex stress – likelihood that MEES-driven capex will push equity returns below target unless rents or yields re-price
- Liquidity risk – time-to-sell or refinancing risk indicators at various stages of the interest-rate cycle
4. Action engine: "What should we actually do?"
The final step is turning scores into concrete sourcing and asset-management actions:
- Sourcing focus lists – rank assets by future underperformance risk and ease of repositioning
- Business-plan templates by archetype – define playbooks for common underperformance patterns
- Portfolio triage – apply models internally to decide which assets to defend, sell, or reposition
An illustrative example
Consider an anonymised, illustrative scenario for a UK regional office block:
- • 6,500 m² multi-let office in a Tier-2 city fringe location
- • Current EPC C, gas-fired heating, traditional floorplates
- • WAULT 4.2 years; headline occupancy 95%; passing rents slightly below prime CBD
- • Surface parking, limited amenity, walk-time to rail c. 12 minutes
Traditional view
Strong income profile today, modest capex required, "hold and forget".
AI/data-driven view
- • MEES/EPC risk model predicts significant capex required to reach EPC B
- • Vacancy risk model forecasts sharp jump in void risk around 2029–2031 lease events
- • Liquidity model flags that lenders are already discounting secondary EPC C stock
Outcome: For a typical core buyer, this is tomorrow's underperformer—benign today, but likely to suffer value erosion as EPC, leasing and liquidity risks converge. For a value-add investor with a targeted repositioning plan, this becomes a prime candidate for off-market engagement, with a clear angle: solve the seller's looming MEES and refinancing problems ahead of the curve.
Building this inside a UK value-add platform
How does this turn into something executable inside an organisation like Fenrir?
Step 1 – Define the investor's "sweet spot"
Before building models, be explicit about what "attractive underperformance" looks like for the strategy: target geographies and lot sizes, acceptable EPC starting points, maximum tolerance for lease-up risk, planning risk and construction complexity.
Step 2 – Assemble and govern the data
Establish a single asset ID across all internal systems. Secure rights to use key external datasets. Put basic data-governance in place: data dictionaries, quality checks, refresh cycles, privacy safeguards.
Step 3 – Start with narrow, testable models
Rather than a monolithic "asset score," build small, focused models that answer very specific questions. Each model must be back-tested against historic outcomes and regularly recalibrated as markets move.
Step 4 – Embed in the sourcing and IC process
Use the models to generate monthly target lists of assets or landlords for the origination team. Include risk and opportunity scores in investment committee papers. Track realised performance vs. model predictions as part of post-deal reviews, feeding learning back into the models.
What this means for stakeholders
For owners of secondary assets
Silence is not safety—strong current income can mask accumulating EPC, leasing or refinancing risk. Owners should consider running their portfolios through independent AI-based diagnostics to understand where risk is building, and where early capex or disposal might preserve value.
For lenders and capital partners
AI-based early-warning systems can help identify borrowers or assets likely to struggle with MEES compliance or leasing in the next cycle, informing covenant design, pricing and workout strategies. For forward-looking lenders, partnering with value-add investors using these tools can create structured opportunities: green loans for retrofit, transition finance, or JV recaps of under-invested stock.
For Fenrir and similar value-add platforms
The combination of tightening MEES rules and uneven adoption of AI means the market will bifurcate between owners caught off-guard as assets slip into regulatory non-compliance or leasing trouble, and investors who systematically identify those issues early and offer solutions. A disciplined AI and data strategy can turn this into a repeatable sourcing advantage—not by chasing generic "AI hype", but by focusing narrowly on the signals that define tomorrow's underperformers in Fenrir's chosen niches.
How Fenrir can help
For owners, lenders and partners facing UK commercial assets that feel "fine on paper" but may be building hidden risk, Fenrir can:
- Run portfolio-level diagnostics to surface future underperformers across EPC, leasing and market-cycle dimensions
- Co-design asset-specific repositioning plans that translate model outputs into capex, leasing and ESG strategies
- Originate and structure value-add transactions—from bilateral acquisitions to recapitalisations—where data-driven insight into tomorrow's underperformance creates value for all parties
Used well, AI doesn't replace local knowledge, sector expertise or relationships. It sharpens them—turning the noise of competing deals into a focused pipeline of underperforming assets where Fenrir can create durable, bankable value.