Prepare For AW21 & Beyond | COVID Prediction Model For Fashion

The retail scenario in India and across the world is going through uncertain times. Consumers want fewer things and clarity. When there is uncertainty over time, the insurance to business is dynamic learning models. In this edit we take you through the approach we at Stylumia are enabling our clients. We are enabling brands to navigate the period minimizing variance with reality. The engine that is powering this is a proprietary dynamic COVID prediction model for fashion. The cost of using judgemental static views will be very expensive during these uncertain times. The resultant business impact is optimizing revenue and margin through minimizing overstock/over discount, understock, and under discount.

We will take you through the dimensions which impact consumer sentiment, the local and global factors that are relevant to consider.

Why Are AW21 & Beyond Predictions Important Now?

As the lockdown restrictions relax in the coming weeks, all brands will try to sell out as much as possible. While doing so, the key considerations are,

1. Do not over discount during EOSS (end of season sale)
2. Keep what is relevant for autumn winter 2021
3. Estimate what would be demand for aw21 considering possible covid scenarios including a potential third wave
4. Make adjustments to SS22 through this period as situation develops
5. Right-size winter specific merchandise

These are dynamic challenges and difficult to make rule-based. The economic value unlock is very significant if done prudently.

Considerations For The COVID Prediction Model For Fashion?

The COVID prediction model for fashion or any retail business needs to take into account attributes or parameters impacting the predicted variable. In this case, our interest is demand during future months considering all COVID conditions. The various components that go into modelling demand for a brand are as follows,

Fig. 1: Stylumia's Covid Prediction Model For Fashion Retail

Brand's Own Data

We take the transaction data across channels, style masters, store masters and promotion, web data (for online). This data gets cleaned, evaluated for anomalies, validated with domain experts for any observation, imputed for missing values and then ready for consumption.

We come with the principle that any AI/ML prediction is 80% data. We sweep the floor of data thoroughly working with experts ensuring all domain intelligence is also infused in the data as attributes or features.

We also create new features (parameters) through iteration which are reflective of your brand's demand performance.

We are facing a future period that is not an exact reflection of the past. Hence just the brand's data is not sufficient for the model to understand the future condition.

This leads us to alternate data / external data that contain demand signals.

Local Country COVID Trends

It is important to understand the context of the covid scenario of the country where the brand is trading. We take all the COVID related features/parameters of the past like,

1. Number of cases / case rate to population
2. Number of deaths
3. Lockdown dates
4. Vaccinate rate
5. Fatality rate etc

We also create new parameters based on these which have a reflection and predictive signal of the COVID cases.

It is not enough that we have the past data for these signals to predict the future. We need to predict/ simulate the scenarios for how the COVID parameters for the country will look like into the AW 21 months.

This is a non-trivial task.

COVID Prediction Model For Fashion | AW21 Period

In order to estimate how will the future look like we first map global COVID trends country by country to see their journey along with their parameters.

We then identify patterns that look closer to what the scenario could be for the local country.

Here are some examples of countries that have gone through more than two waves which can give signals for countries that are still in their second wave. There is no straight line here and hence this needs to be seen in conjunction with their respective COVID parameters like,

1. Vaccination rate
2. Social Norms
3. Healthcare action intensity
4. Lockdown actions etc.

A small representation of countries with more than two waves with their case rates and vaccination rates...

COVID Cases & Vaccination Trend In France

A country with 3 waves so far and the third wave was as big as the second wave.

COVID Cases & Vaccination Trend In The UK

We can call it a country with 2 waves so far. The second wave had two phases.

COVID Cases & Vaccination Trend In The USA

In a country with 3 waves so far, the first wave had two phases.

COVID Cases & Vaccination Trend In India

A country with two waves, a very strong second wave with a declining trend.

The task now is to simulate what would India COVID Trends be into the AW21 period. This is done by our simulation engine using the international trends, local trends, and expected vaccination rates.

Like any uncertain event prediction, it is always good to have scenarios so that the brand can take a position on the scenario and start placing bets.

We are creating three scenarios from an Indian perspective (it can be modelled for a country's current status),

1. An Optimistic One | A third wave does come (a mild one) and does not impact business
2. A Moderate one | A third wave comes with an intensity, does create some disruption
3. Pessimistic one | A third wave comes which is a severe one

All of these scenarios are created with the simulation of all the local and international data as mentioned above.

Predicting The AW 21 Demand

Now that we have the simulation of COVID trends for three scenarios, our proprietary prediction models are fed with all the data elements in Fig. 1. The result will be a prediction of demand across the hierarchy of business at the higher levels to the style color level.

The brand can take the following actions based on the outcome,

1. Adjust AW21 orders where possible
2. Take the relevant carry forward decisions across seasons
3. Take time-sensitive seasonal merchandise decisions (Festive / Winter period)
4. Adjust promotions/markdown plan

Considering parameters in the external environment impacting the outcome, we will iterate through the season for ongoing adjustments.

In Conclusion,

In times of uncertainty, it is imperative to make data-informed decisions. It is all the more critical when the situation is complex, simple thumb rules will not help brands to optimize the business variables.

With a good amount of demand contraction already through the year (demand expansion for the digital native), brands are embracing a scientific approach to demand planning and getting ahead of the pack with maximum return on all the capital employed.

Brands we work with have not only grown through COVID wave 1 relative to their market peers, but they also have seen a significant lift in forecast accuracies upwards of 30%* (*- based on our Stylumia's engagements with existing clients)

If you would like to make the best of technology for the application through the upcoming months, time to start now. Stylumia's COVID prediction model for fashion is a step in that direction. For those interested, reach us through a form here.