Unilever
Predicting the spread of COVID-19 to safeguard Unilever’s global ecosystem
Tags
In 2020, as the COVID-19 pandemic accelerated, Unilever needed to rapidly evolve their workforce strategy to help protect employees, while maintaining the supply of products to 2.5 billion consumers worldwide.
To tackle this challenge, PA and Unilever teamed up to create a world-leading predictive tool – COVID-19 Awareness and Situational Intelligence (CASI). Our teams combined expertise in consumer goods, business intelligence, data analytics, AI and machine learning, digital, operational resilience and global supply chains.
CASI is a live dashboard that monitors and accurately predicts COVID-19 trends, providing real-time and predictive intelligence from a global and regional level down to individual Unilever sites worldwide. CASI achieves 75 per cent prediction accuracy for 30-day forecasts. Unilever teams use CASI daily to manage supply chain operations. CASI redefines how data can be harnessed to unlock predictive insight.
Key successes
- Built a world-leading COVID-19 monitoring and forecasting tool with Unilever at speed, delivering global rollout within three months
- Applied artificial intelligence and machine learning to create predictive models and accelerate decision-making at Unilever facilities worldwide
- Aggregated, normalised, and validated more than 250,000 COVID-19-related data points daily, presented using an automated and dynamic dashboard
- Predicted COVID-19 trends at local level; 80 per cent accuracy for seven-day forecasts and 75 per cent for 30-day forecasts
Making sense of an unpredictable crisis
As COVID-19 spread across the world in 2020, information was suddenly everywhere, but hard to aggregate and validate. Countries began reporting on infections and deaths but used different methods and timeframes, and not all data was deemed trustworthy. In addition, community-level data was not widely available to enable companies to make critical decisions about local business operations. Finally, all reporting lagged actual events, meaning it provided little intelligence about the current situation, let alone what lay ahead.
As the pandemic accelerated, Unilever sites needed to make decisions quickly on changes to operations, impacting not just supply chain, but their workforce, their families, and local communities. Different regions were on different trajectories; some countries were able to reduce COVID-19 transmission, while in others, cases were skyrocketing out of control.
Unilever needed to protect their global workforce, maintain supply to a 25 million-strong global network of retailers, and ensure 2.5 billion people in over 190 countries could continue to use one of Unilever’s 400 household name brands such as Dove and Lifebuoy every day. By so doing, Unilever could also provide consumers with the hygiene supplies they need to reduce risks and protect themselves and their families.
Building a world-leading pandemic predictive tool
Unilever needed to forecast the future direction of the pandemic. The task of aggregating multiple data sources and data points for global and granular crisis management was too time-consuming as the pandemic accelerated, and the data available was often out-of-date by the time it was validated and made usable.
PA had successfully collaborated with Unilever’s General Counsel of Global Supply Chain, Wei Ling Lim, to accelerate the digital transformation of the company’s legal function, as part of Unilever’s Future First digital transformation programme.
As a trusted partner, Unilever asked PA to co-create a digital tool at speed, to improve visibility of the COVID-19 pandemic, identify emerging risks and guide decision-making. Since each day brought new developments, speed was of the essence. The goal was to launch a usable tool rapidly and move quickly from retrospective to predictive insights.
The joint Unilever-PA team combined expertise in fast-moving consumer goods, business intelligence, data analytics and engineering, AI and machine learning, digital, operational risk and resilience, and global supply chains. Our team would need to ask and test challenging questions, interpret medical data, access and aggregate data sources, and build data models to create the holistic and granular insights Unilever sought.
Combining pandemic insight with data expertise
Starting with the business need, we determined what insights Unilever supply chain leads, factory and site owners, incident management and global health teams would need from the data. The joint Unilever-PA team initially focused on 30 countries.
Our team took a simple template from Unilever, showing how they wanted to measure COVID-19 trends daily and make business decisions. Within a week we had collaborated to translate this into a visually dynamic and self-service tool that could be made available throughout Unilever businesses. To do so, we chose Microsoft Azure, Power BI, Power Apps, Power Automate and Power Flow as the core components of the platform.
One early hypothesis was that we could make evidenced predictions about the pandemic’s spread by combining the fluctuating R value for COVID-19 – the virus reproduction rate – with epidemiological and data modelling. Using published research and modelling, we designed an R-based model, combining daily calculations of national R numbers with projected secondary infection times, to generate near-term projections on expected national and subnational trends, globally.
Navigating an immense volume of data
Working in partnership with Unilever, we embarked on a vast data aggregation exercise to populate CASI, navigating an immense volume of data. PA experts in data analytics searched daily for new data sources to enrich global and national reports, using Google and Apple APIs to integrate this information. We tapped local health ministry reports, non-governmental data sources, medical models, and crowd-sourced data on infections and deaths, ultimately integrating 75 subnational data sources. Collecting data was challenging, as information sources started and shut down regularly.
We combined the primary indicators recognised by health bodies to monitor the spread of COVID-19 – including infection rates, testing rates, positivity rates and mortality rates – with secondary indicators – such as the fluctuating virus reproduction rate, stringency index, hospital capacity figures and vaccination rates. To layer in how people’s behaviour contributed to the spread of COVID-19, we supplemented this with tertiary, social data including government lockdowns, social restrictions, consumer mobility and social media behaviour.
Next, we tagged, normalised and validated this wealth of data – a considerable feat in data engineering – with automated quality checks, to ensure that the insights produced would remain valid and actionable.
Version one was up and running at Unilever within one week. From there, our joint Unilever-PA team established and maintained a delivery cadence of 24-hour sprints and weekly major releases, securing board approval for global rollout of the tool within three months.
Getting ahead of COVID-19 with AI and machine learning
While in the early weeks the data presented was largely retrospective, our experts in advanced data science soon began going further, developing artificial intelligence and machine learning models that could harness the data collected to generate predictive insights. Every day, CASI calculates a new R value for every country in the world, combining this with modelling to create short term projections in the bottom right corner of the screen.
Our teams initially tested CASI with a seven-day forecast for rates of COVID-19 testing, positive tests, vaccination rates and other pandemic-related indicators in key geographies. Comparing predictions to actual results, we maintained an accuracy rate of over 80 per cent.
This initial predictive capability was rapidly evolved to deliver extended 14-day and 30-day forecasts for over 30 geographies every day. By testing and validating the performance of CASI 30-day predictive models, we achieved and maintained an average accuracy rate of 75 per cent, a real benefit to support Unilever’s strategic decision-making.
Securing the global rollout of CASI at Unilever
Unilever’s central COVID incident management team supported CASI for worldwide business use as soon as CASI’s accuracy rate was established. Together, the joint PA-Unilever team supported the global deployment of CASI by facilitating training for Unilever regional incident management teams around the world.
Every single day, CASI is accessed 800 times across Unilever, with teams around the world monitoring COVID-19 trends impacting hundreds of facilities, warehouses and distribution centres. Every single week, all site and factory leads log in to CASI to receive metrics and forecasts for their region. CASI enables Unilever to manage the safety and resilience of their sites, evaluate supply chain impacts, evolve back-to-work and health and safety protocols, and make longer-term strategy decisions.
Sharing CASI with Unilever’s global ecosystem – and looking to the future
To support Unilever’s decision to share CASI throughout their global ecosystem, we collaborated with Unilever to create and shared additional versions of the tool. The multiple versions of CASI include the primary one for incident management teams, one for supply chain partners and a ‘CASI for all’ app for all Unilever employees and their families. All versions are automated from the primary platform.
Together with Unilever we have presented CASI to a range of industries, demonstrating how it can be used to manage crises and predict trends more strategically. The tool can be used, adapted or recreated to focus on incidents and crises; supply chain impacts and risks; cybersecurity; consumer transport and travel; workforce policy; improving employee health and wellbeing; predicting consumer behaviour; and infrastructure investment, among other use cases.
Together, Unilever and PA have redefined how data can be harnessed to unlock predictive insight on a significant and meaningful scale.