Nicolas Mallison

Designed and implemented a multi-horizon (26, 39, 52 weeks) hierarchical demand forecasting solution for an e-commerce furniture retailer, handling 2000+ SKUs across multiple categories. Leveraged AutoGluon's time series capabilities to create an ensemble model incorporating promotional effects. Developed a robust data pipeline that processes weekly sales data, stock levels, and promotional calendars to generate SKU-level forecasts. Included in the solution is automatic promotional adjustment through residual modeling and hierarchical reconciliation across three levels (all -> category -> sub-category), ensuring forecast consistency. Built an inventory optimization module that integrates with the demand forecasting system to determine optimal reorder points and quantities. Created a solution designed to account for supplier-specific constraints (lead times, minimum order quantities) and target service levels, while handling SKUs with varying replenishment cycles. Implemented comprehensive model monitoring and validation frameworks, including directional accuracy metrics and correlation analysis of forecast differences, enabling early detection of forecast degradation and automated model retraining decisions.

Advised on the optimal AI and ML strategy to leverage a unique dataset of historical financial, operational, and market data, encompassing over 4000 businesses and 500 variables. Surpassed the performance of their current production multivariate linear regression model whilst adding much more flexibility/dynamism in the type and nature of model inputs. Designed and implemented a suite of masked self-supervised autoencoder models using PyTorch on Azure to effectively predict the long-term strategic performance of a business. Achieved superior performance regardless of the type of historical financial, operational, and market data provided as input, while also determining the most relevant data points to sequentially improve accuracy. Leveraged autoencoder encoding and model factor importance to perform dynamic searches on any input, obtaining the best set of look-alikes. Used model insights to recommend interventions most likely to enhance long-term strategic performance based on the model-identified operational and financial indicators that differentiate top performers from underperformers in the retrieved sample.

Performed overall consultation on specific data science projects, data, and analysis tooling. Executed projects leveraging the hex tech platform and then taught the rest of the team how to do it themselves. Advised on how to cluster customers based on their gameplay activity and analyze the correlation/relationship between gameplay, marketing promotions, and customer churn to build a predictive model.

Put safely into production a demand forecasting and labor scheduling model that predicts 80+ stores' daily entrees weekly and calculates hourly staffing by roles, rewriting the data layer for the reliability of model inputs and fit convergence. Added to the weekly production model pipeline a strong model fit and output quality assurance test module that would automatically alert in case of anything missing or incorrect forecast outputs and stop uploading them in labor schedules. Designed, built, and put into production a prediction performance monitoring Shiny app, allowing interactive visual comparison of predicted-versus-actuals store by store and used the mean absolute scaled error to inform when the model needs retuning.

Loaded, cleaned, transformed, and analyzed a large set of semi-structured engineering and construction project data to determine the correlation between the information available at the bidding stage and the projects' financial performance. Analyzed customer details, project description, contract price, cost estimates, change order cost and revenue, actual costs, and invoice value. Then, I built predictive models to support risk-based decision-making at the bidding stage. Delivered an interactive visualization Shiny app that demonstrated how model predictions were derived from project data, compared a project's risk against others, and provided online scoring of new projects to allow testing of model predictions.

Combined machine learning and time series encoding to develop a bank-wide innovative, predictive, and evidence-based model that predicts future IT system stability, enabling smarter decisions. Designed, built, and deployed into production a suite of ML products that influenced an excess of 30% risk reduction across the bank IT applications due to being incorporated in the bank's strategic balanced scorecard. Extended the initial IT stability-affected risk model to application and non-application IT assets, such as infra or platform components, and from a risk-causing perspective, making the risk impact of shared infrastructure visible to management. Built NLP machine learning models to characterize the nature of the IT work performed by an agent, allowing for much more accurate measurement of task productivity and targeted routing of tickets to the ablest available agent for speedier resolution. Created unsupervised machine learning models to identify opportunities to eliminate IT service management tasks, such as repeat tickets, and automate broken processes through NLP and process mining. Devised an IT service management task demand model to match service capacity better and reduce costs by dynamically matching capacity to forecasted demand and taking advantage of volume-based discounts. Led the predictive analytics and data science function in the technology risk department to develop models, which kept gaining appreciation from a wide range of senior technology stakeholders and leaders across the bank.

Developed models accurately predicting the remaining lifetime of electrical railways assets—such as railroad switches based on asset age, point-in-time cable conductance, and external weather data—which optimized asset replacement plans. Created models to quantify airport parking demand based on price, parking occupancy, month/day of the year, duration, customer geo-demographics, and flight details—to dynamically adjust pricing based on predictions and maximize profit. Devised models to quantify the mean time between failure of expensive oil drill equipment—leveraging IoT instrumentation and other environmental usage data—to optimize maintenance intervention and maximize equipment use time (predictive maintenance). Built healthcare demand models for maternity services elective interventions such as cesarean sections to optimize resource planning and quantify the presence of any geo/socio-demographic drivers impacting the procedure decision process. Produced predictive and prescriptive models of the air force helicopter maintenance schedule and process flow—based on site locations, resource constraints, and equipment capacity—predicting the impact of budget reduction on future availability. Established and led a newly formed solutions team within the UK and globally to harvest data assets and explore insights that could inform strategy and drive business outcomes. Doubled the practice size from 23 to 45 consultants in three years by blending financial, simulation, statistical, and operational research optimization modeling expertise and deep technical experience. Incorporated data science in consulting engagement to provide a data-driven evidential basis for improving and recommending performance across all consulting go-to-market offerings. Led effective interactions with clients on an issue-based consulting basis where specific strategic and operational challenges were resolved using data-driven approaches and analytics. Provided professional advice to clients on best practices and performance through information management advisory, digital transformation analytics, strategic analytics-led consulting, operational decision support, and data science consulting.

Led the data analytics to trace a $1.6 billion customer-segregated funds shortfall after the bankruptcy of a global financial derivatives broker. See page 11 of blogs.harvard.edu/bankruptcyroundtable/files/2017/09/JPMCC-Till-MF-Global.pdf. Built a legally defensible and statistically sound extrapolation method to quantify the probability that the total amount of claim leakage would likely exceed a threshold in a contractual dispute between the underwriter and claims handling provider. Devised statistically sound and legally defensible approaches for the use of predictive document coding to accelerate the legal discovery of digital evidence by learning models on labeled samples allowing quicker identification of relevant documents. Developed a statistical method to score digital evidence based on the intensity of their use of language related to the three dimensions of Cressey's Fraud Triangle—pressure or incentive, opportunity, and rationalization—to identify fraud hotspots. Founded, led, and developed its innovative forensic data analytics team, which grew within 60 months from 1 to 25 people with revenue for FY13 in excess of £5.5 million, and £0.5 million in FY08. Spearheaded the forensic data analytics proposition development, thought leadership, recruitment, sales, and business development, which led to the growth of the wider FTDS practice from a £1.5 million to a £15.5 million business in five years. Developed a high-quality, culturally diverse team with one of the highest retention rates in the industry, comprising much sought-after talent with deep technical expertise. Contributed significantly to promoting and coordinating analytics services cross-network as the FTDS EMEIA analytics lead by developing strategic account plans executed in collaboration with the global client service partners. Ensured that applicable analytics propositions and subject matter experts were known and ready to operate proactively and reactively. Created the global forensic data analytics methodology for the successful execution of engagements.

Developed a graph mining method to identify the layering and concealment of a sophisticated financial statement fraud involving the use of around 21,000 manual general ledger journal entries that were hidden in a dataset of 35 million records. Used data analytics to identify multiple steps in which fraud was committed by automating the tracing and displaying money flows from one side of the balance sheet. Established a new forensic technology service line, built a 12-member team, and increased company revenue from £0.308 million to over £2 million within 16 months. Created analytical techniques that transform data by extracting useful information, discovering hidden patterns, and facilitating conclusions so businesses can proactively and cost-effectively seek out to prevent and detect fraud, waste, and abuse. Packaged data collection and preparation, automated investigative linking, rules-based, and model-based analysis methods and techniques into generic and sector-specific fraud detection and investigation solutions. Delivered through a purpose-built scalable Data Lab technology facility with dedicated resources, proven quality assurance processes, pre-built systems software, hardware infrastructure, and best-of-breed data analytics applications. Used Data Lab to enable fast, predictable, and consistent delivery of analytics services incorporating investigative experience across citizen, employee, accounting, supplier, customer, product, and unstructured and transactional data.

Instrumental in landing two major fraud managed service data analytics deals worth several millions of pounds in revenue to Detica (this became known later as Detica NetReveal). Designed and developed a groundbreaking cross-industry motor, home, and personal injury insurance claim fraud detection system for the UK Insurance Anti-fraud Bureau. Designed and developed groundbreaking non-compliance in a taxation detection system for HMRC/Inland Revenue UK tax authorities (identification of ghosts, moonlighters, and under declarers). Developed a strong reputation rapidly (both internally and externally) for technical excellence through repeatable delivery and innovative thinking. Ran Detica’s Datalab that operates as a center of excellence for data analytics and incubator for delivery of data analytics managed services: This encompassed recruiting, staff training and development, tools, and methods. Supported the facility's selling through a proactive working relationship with the business units and other business development functions to identify potential proposition areas. Supported business unit sales activity through demonstrable expertise and other materials, helping with Datalab proposal writing and project estimations, identifying, and recruiting staff to meet demand. Communicated the proposition and benefits of the facility more broadly within Detica and worked with the marketing team to position these externally. Put in place the appropriate business processes to ensure reliable and repeatable delivery and approaches to capture and build re-usable know-how, tools and components, coaching, counseling, and supporting the career development of staff. Provided expert analytics input on client engagements, meeting with clients and project management for clients where services are wholly provided by Datalab.

Oversaw development, improvement, and automation of the customer recruitment tracking capabilities and program planning tools. Drove a number of complex data analysis projects and strategic analysis. Analysis, reporting, and modeling using the SAS System. Performed data manipulation and transformation on huge transactional databases. Managed the improvement and full automation of the planning decision-making tool, enabling the program manager to optimize timings and allocation of budget across the different acquisition media to maximize expected net income. Developed generic SAS and VBA code to fully automate tracking of the results of every campaign on every company’s key metrics (Net income, activation rate, response rate, ECT lending, risk profile). Modeled impact of household structure and member targeting strategy on campaign response rate. Analyzed the impact of increasing credit card applications backlog (time to decision) on activation rate and customer lifetime value. Conducted over solicitation analysis with and across acquisition channels addressing cannibalization issues between media with regards to timings in contact strategies and impact on response rates. Oversaw management, coaching, and skills development of junior analysts in the team. Wrote and presented to senior management PowerPoint presentations summarizing main findings on every modeling and analytics project.

Produced sophisticated statistical analyses and models for a big online financial retailer (complex financial product behavioral segmentation on transactional data, path analysis to understand behavioral impact of DM solicitations). Developed new business methodologies and advanced applied statistical techniques for new business offerings (in particular geo-marketing, response measurement of DM activity for FMCG companies, etc.). Led the statistics and analytics new business team. In charge of day-to-day management of a part of the Stats and Analysis Team. Wrote the analytics components of business proposals for agency and/or direct clients. Supervised basic analyses by junior analysts. Improved the scheduling and project management procedures and system by using MS Project to leverage time and better staff each analyst (team of four analysts). Developed quality control and quality assurance of analysis methods and processes. Collaborated in the development (Zalpha CRM Partner) of a new business offering: How to obtain an ROI from existing investments in CRM software analytics and modeling components of CRM architecture focus

Pioneered innovative models to dynamically determine the catchment trade area of the point of sale based on time travel, outlet size, competitor locations and size, and census zone geo demographics. Model parameters via regression best fit to annual sales. Coordinated conception and implementation of geo-statistical analysis methodologies to realize surveys and/or implement sales and marketing information systems for large key client accounts. Managed the surveys and solutions department. Managed marketing information system conception projects. Defined and developed vertical and horizontal concepts solutions. Oversaw conception and production of both general and industry specific geo-statistical indicators (clusters etc.). Contributed to technology and software enhancement (conceptual design for automated analysis) and technological watch.

Acted as a founding partner of Cognitive Relation (now Yseop), a consulting firm specializing in customer relationship management personalization solutions. Managed the development of decisional marketing capability. Developed personalization solutions using state-of-the-art technology, combining a powerful rules engine (IA) with a natural language text generator. Developed a web-based personalized sales dialogue builder and a personalized mailing generator. Created three prototypes showing the functionalities and benefits of the former solutions. Completed the business development of those offers, focusing on retail groups and banks. Wrote the business and product development plans, focusing on the decisional feedback module. Conducted extensive meetings with venture capital investors to raise product and commercial development funds.

Conducted strategic customer insight and analytical CRM project for a large French retail group. Defined the analysis framework for generating customer information to build a predictive model of purchasing behavior. Implemented this conceptual analysis in the form of a qualification questionnaire. Designed and implemented the methodological process for efficient data warehousing and exploitation of the information using SAS. Programmed under SAS of a parametric predictive model and estimation of the parameter (precision obtained: 80%, robustness: 97%). Estimated additional turnover potential associated with a relationship marketing loyalty program.