Visitor segmentation

Inspiration

As SafeEntry starts to get implemented on a wider scale across more businesses, we should start using these data points to not only support existing contact tracing efforts, but leverage on this data to perform analytics to understand how the lockdown has impacted people’s travel patterns and which group of visitors is risky of infection. A visitor is someone who has travelled.

This can potentially be used to assess how different lockdown measures has impacted people’s travel patterns, informing policy makers on the impact of different measures.

What it does

SafeEntry collects data on the visit location, type of essential service visited, time of visit and dwell time of each visit. From the form submission, we are also able to infer the age group of the visitor. We can aggregate these data points to start building a segmentation analysis of visitors, drawing key insights such as frequency of travel, who do they travel with, travel behaviours and preferences, number of unique locations visited per week, what essential services do they use, and infer their “shopping mission”. Examples of shopping missions that can be inferred include:

• Travel to purchase daily groceries (checked in at supermarkets every other day)
• Travel for work (regularly check in to 2 locations relatively far away)
• Travel for food (check in daily at the same shopping mall near meal times).

Each visitor segment could then be assessed for risk of infection. For example, a segment that travels almost everyday is more likely to be exposed to the virus. A visitor segment that over-indexes with people belonging to an older age group is also a riskier group. Riskier segments identified could then be targeted to bring down their risk profiles, and we can leverage on other apps that Govtech has built. Potential actionable insights include:

• A visitor segment is identified to be risky as they regularly travel for groceries - the government could provide incentives/vouchers via RedeemSG to encourage these people to shop online, such as providing free NTUC delivery slots, so as to reduce need for travel.
• A visitor segment is identified to be risky as they regularly travel during 12-2pm - this segment might be travelling to buy lunch (inferred shopping mission, but could be crossed checked by investigating if they check in with food businesses). It would be critical to monitor this segment. Does phase 1 and opening of more stores such as bubble tea shops cause their dwell time and/or visit frequency to go up?
• We can deep dive into community cases and see which visitor segment they belong to. Are there any drivers as to why they were infected? Do they belong to a segment that works in their office as essential personnel? This can help the government identify key drivers to focus policies on, and decide which measures to loosen during the phased circuit breaker period.

One key concern is personal data used and data privacy concerns. Personal identifiers (such as NRIC) tagged to the data should be hashed and the key should be stored separate from the data analysts. All results produced should also be aggregated, to a minimum visitor segment of 100 people. This can further prevent traceability of an individual.

How to build this

This idea is conceived with the fact that the data should be available, and no further data capture work needs to be done. Besides data cleansing, data scientists would need to perform EDA and perform segmentation analysis. It might be as quick as a week or two to deliver some initial insights.

Challenges that might be faced

Use of personal data needs to be well managed here. We are not aiming to track an individual, but to identify key visitor segments that are high risk to infection. This analysis provides a baseline data-driven understanding of visitor segments, and we can assess how different policies will and have impacted the risk profiles of certain segments.

Built With

• python
• sql