*** Delving into the Potential of AI to Tackle Service Charges

Improvements in service levels have been evident since the inception of the transformation project. However, the persistent challenge lies in the yearly escalation of service charges, where we have yet to see control.  A neighbour within our community recently undertook a brief course on Artificial Intelligence (AI) at Saïd Business School, University of Oxford. The course tasked students with formulating a business case utilizing AI to address real-world organizational issues. This presented an ideal opportunity to scrutinize our service charge dilemmas. The resulting business case specifically targets the gap between estimated and actual costs, along with the lack of transparency regarding the rationale behind substantial service charge increments and their distribution. Although this proposition remains in the conceptual realm, it serves as a catalyst for discussion and has been forwarded to our representatives in Guildhall for further consideration.  Special thanks to Leila Allen for graciously allowing us to share her paper. For any feedback, please reach out to her at allenleila@hotmail.com

Madhav-Malhotra-003, CC0, via Wikimedia Commons

Revolutionising Service Charge Management: A Business Case for AI

Author: Leila Allen (allenleila@hotmail.com)

Date: 10 February 2024

Information asymmetry is a key driver of uncertainty in business, fostering an environment conducive to sub-optimal decision-making and resource allocation.

 Harnessing artificial intelligence (AI) offers a potential remedy for mitigating uncertainty in business.  The following business case seeks to delineate a viable AI solution designed to benefit both business and customers, whilst addressing the existing challenge of information asymmetry presently contributing to sub-optimal decision-making in the given scenario.

 The City of London Corporation (CoL Corp) serves as the governing municipality for the City of London, overseeing and providing services to both residents and businesses within the Square Mile.  CoL Corp is responsible for the management and maintenance of numerous residential estates within its jurisdiction, including the notable Barbican Estate.  Flat owners, termed “leaseholders” (LHs), residing in residential estates owned and managed by CoL Corp, are charged an annual levy, known as a Service Charge (SC), imposed by CoL Corp to cover various costs related to services provided, which include

  • Customer care expenses 
  • Estate management, including window cleaning and concierge services 
  • Property management, including repairs to communal parts and general repairs, lift maintenance and electricity costs (incorporating underfloor heating costs)
  • Ground maintenance costs 
  • Non-annually recurring items, including internal redecorations to communal parts and water damage maintenance 

 At the start of each financial year, CoL Corp issues LHs an invoice for SCs detailing estimated costs for expected services for that year.  As actual costs materialise, estimated expenses are adjusted based on actual expenditures and unforeseen factors that emerge during the year.  At year-end, LHs receive an additional invoice to settle outstanding costs where earlier estimations have fallen short.  SCs have surged in recent years, indicating a substantial increase in actual costs and underscoring a notable disparity between estimated and realised expenses.  This situation has resulted in tensions among stakeholders, as CoL Corp has struggled to effectively showcase “value-for-money” to residents, who, in turn insist on transparency regarding expenditure details and demand to see evidence of rigorous tendering processes and equitable cost assessments.

A significant driver behind the recent marked increase in SCs is related to electricity costs.  CoL Corp reportedly purchases electricity a year ahead of its actual requirement.  Despite this, updates from CoL Corp highlight significant year-end adjustments to actual electricity expenses.  Moreover, in January 2023, CoL Corp entered into a Power Purchase Agreement to directly acquire solar power from the generating company, ostensibly offering advantages to residents.  However, the possibility of rebates depends on the productivity of the solar farm, which CoL Corp has indicated cannot be predetermined.  This arrangement necessitates thorough scrutiny to ensure it delivers “value-for-money” and genuinely serves residents’ interests. 

 Therefore, to mitigate invoicing uncertainty for LHs, enhance transparency, and optimise decision-making within CoL Corp, I propose the implementation of a multivariate time series prediction solution employing deep learning, which will enable both multivariate and probabilistic forecasting.  Long Short-Term Memory (LSTM) networks, a type of recurrent neural network (RNN), are widely used for sequence prediction problems due to their ability at capturing dependencies in persistent information.  In leveraging a LSTM model, this approach aims to forecast key variables using historical sequential data: 

  1. Temperature and electricity-requirements variables: Enable predictions of heating and electricity requirements based on predicted temperatures to facilitate anticipated cost projections based on past consumption 
  2. Sub-contractor cost forecasting: Determine budgets related to sub-contracted tasks through accurate forecasting of contractor costs based on historic demand 
  3. Internal staff cost forecasting: Enhance decision-making on personnel requirements by forecasting internal staff costs where, for example, reduction in duplication of clerical activities across estates may be achieved  
  4. Major-works-requirements forecasting: Forecast the probability of major-works requirements and associated costs based on historical expenditure, maintenance schedules and routine upgrades 
  5. Cash flow management: Utilising payment dates or identifying potential late payments based on historical invoicing data and payment behaviour patterns may be particularly valuable for financial planning, cash flow management, and risk mitigation 

 This proactive solution not only addresses invoicing uncertainties but also promotes transparency and operational efficiency, empowering CoL Corp to make informed decisions and optimise resource allocation.  The implementation of a LSTM network is suggested for its proficiency in capturing intricate patterns within time series data. 

 This ambitious set of goals represents a comprehensive agenda, and the proposed solution is likely to introduce a significant disruptive element.  Consequently, the business case for such a bold endeavour must present credible and viable benefits. 

 In addition to the anticipated advantages derived from reduced uncertainty, CoL Corp can harness additional insights expected to emerge from the model, fostering a more comprehensive understanding of: 

  • ‘Hot spots’ necessitating recurrent maintenance, shedding light on substandard repairs or equipment / structures nearing end of operational life, and thus requiring replacement. 
  • Potential instances of overcharging or fraudulent invoicing by sub-contractors. 
  • Enhanced precision in forecasting Return on Investment (ROI) when evaluating competing priorities with constrained resources. 
  • Processing efficiencies across the estate portfolio by mitigating duplication of tasks  
  • Enhanced negotiation capabilities to secure better pricing from sub-contractors by comparing similar works and associated costs across the estate portfolio. 

Exploring existing successful applications of LSTM models offers valuable insights into the power of this technology: Jailani et al. (2023) showcased promising results into the effectiveness of deep learning models in improving the precision of solar energy forecasts.  Furthermore, the remarkable efficiency of LSTM models is evident in their role in advancing next-word prediction (Ambulgekar et al., 2021) algorithms, significantly enhancing user experiences across smartphone messaging and search engine platforms.

Despite the undeniable advantages of LSTM models and their demonstrated success across various applications, the considerable magnitude of this venture presents a level of risk that requires careful consideration.  Moreover, implementation necessitates a considerable financial commitment, covering development, data, infrastructure, and implementation costs, and expenses related to staffing, training, maintenance, upgrades, regulatory compliance, and security.  Additionally, non-financial costs, including time and opportunity costs, must be considered.  

Nonetheless, it’s worth noting that an effective approach might involve starting with a scaled-down solution, tackling first LHs primary invoicing concerns, followed by gradual expansion as data accumulation increases.  

The feasibility of this initiative hinges on tapping into multiple data sources.  CoL Corp may leverage its access to a wealth of proprietary historical data, encompassing staffing, sub-contractor, maintenance, resident and consumption records across all estates under its operation. 

 An assessment of relevant proprietary data is crucial to ascertain quality.  Recognising that the efficacy of any prediction model hinges on the quality of input data, it becomes imperative to meticulously organise, clean, and partition records into randomised training, validation, and test sets.  This careful preparation is fundamental to the accuracy and reliability of the predictive system. 

  To enhance predictive accuracy, external data sources, including temperature and financial market data feeds, become essential.  These inputs are crucial for anticipating future heating needs using predicted temperatures, and refining electricity pricing based on real commodity market movements.  This holistic approach ensures a robust foundation for the proposed solution.   

  Furthermore, it is imperative to proactively assess any existing ethical or legal concerns.  CoL Corp is well-advised to establish strategies for evaluating and explaining the model’s output, thereby instilling confidence in residents that reasons for inaccurate invoicing are identified and mitigated as the model’s data becomes enriched over time. 

  Moreover, the model should refrain from incorporating sensitive identifiable resident data in its input to avert unintended instances of bias or preferential invoicing / service provision favouring one group over another. 

  In instances where issues arise, CoL Corp is recommended to meticulously document responsible and accountable parties tasked with ensuring redress. 

It’s evident that a more impactful solution is essential to yield mutual benefits for both CoL Corp and residents.  The escalating costs, often shrouded in information asymmetry, demand a transformative approach.  In an era dominated by concerns about the ‘cost of living’ crisis, embracing this solution becomes not just beneficial but imperative, offering a positive change for all stakeholders involved. 

References:

Jailani, N.L.M. et al. (2023) Investigating the power of LSTM-based models in solar energy forecasting, MDPI. Available at: https://www.mdpi.com/2227-9717/11/5/1382 (Accessed: 12 March 2024).

Ambulgekar, S. et al. (2021) (PDF) next words prediction using recurrent NeuralNetworks. Available at: https://www.researchgate.net/publication/353773522_Next_Words_Prediction_Using_Recurrent_NeuralNetworks (Accessed: 12 March 2024).