Why does the world even need analytics?

Analytics is all about crunching data and getting useful insights to inform business decisions and strategies. For instance, if 85% of your users drop off at the pricing page, you know that your pricing feels exorbitant to them. If 50% of the delivery people are idle from 3–5 pm, you know that you probably need to increase demand in that window by sending in more promotions.

To stay ahead in a customer-centric world, every company is increasingly gathering and crunching data on its customers. After all, analyzing that data can give you information on who your users are, where do they come from, their behavior, tastes, and preferences. This is really powerful in fueling the company’s growth — by retaining existing customers and converting the new ones.

Today, there are plenty of analysis tools that provide answers to your questions on your users, their behavior, tastes, and preferences. We feel it is the “where” part of that pie that is completely broken.

But, before we jump into that, let’s first understand what’s our take and philosophy on analytics, inspired by Uber.

What defines successful analytics?

At Locale.ai, what we are building is “actionable analytics.”

We think about analytics as starting with a problem statement and answering four questions about an event: who, what, where and why. Once that is done successfully, the loop should be closed by influencing a business decision.

We define “success” as gaining insights on all these four dimensions and each decision that our users take should in-turn contribute to making money or saving costs (either directly or indirectly).

For example, let’s assume the repeatability of orders for most users in Jaipur is low. Say, we debug and figure one of the reasons to be inaccurate ETAs. So, now we know what we have to correct for. And once we do that, we hopefully will start more $$$ due to an increase in repeat purchases.

The goal that we strive for is to be able to make the analysis of all the four dimensions (who, what, where, why) very simple so that a business user (who doesn’t know the database structure and a query language very well) can get blazing fast answers and make decisions.

How do we go about implementing this?

Now that we know the outcome we want to achieve, let’s understand the process we adopt to ensure we stick to those. Since, at Locale, our lives revolve around geospatial data, most of the examples that I have used are regarding the use cases that we focus on.

There are four parts to this: Accessibility, Visibility, Intelligence, and Actionability.

Part 1: Accessibility

With geospatial data, accessibility is a big, big challenge — especially, for analysts and business users. What we have learned that it was mostly limited to data engineers (or in rare cases, data scientists).

What I mean by accessibility is business users don’t have the right information the moment they need to make tactical decisions. Moreover, analysts or product managers don’t have the right knowledge to make long-term strategic decisions.

The reason accessibility is a pain because of the format of location data (GPS pings) and the high-frequency nature (every 3 seconds or so). Hence, mixing and matching datasets across different tables and formats in real-time at scale is really hard.

If you want to deep-dive into the challenges associated with location data, here is another piece I have written on this:

A Product for Operational Analytics using Geospatial Data!
What led to the birth of Locale.ai?

Part 2: Visibility

Once we have the data available, next we would want to “see” it presented in the right format and do some exploratory analysis on it. We think visibility helps us answer the who, where and what. You can imagine the “who” to be entities of a business (riders, orders, scooters), the “what” to be a property of the entity and the “where” to be the location.

For instance, let’s say we want to know:

  • Number of riders (who) who are idle (what) in Koramangala (where)
  • Number of order (who) cancellations that happen after 4 mins (what) in Noida 6th Block (where)
  • Number of six nearest (what) delivery executives (who) from this restaurant (where)
  • The demand-supply gap (what) of orders and bikes (who) across every sq. km in Mumbai(where)

Answering these three questions can help us understand the trends of metrics and the second they shoot up abnormally and need an intervention.

On a side note, if you are curious to learn about visualizing geospatial data, check this out:

Visualizing Tesla Superchargers in France Geospatially
A complete guide on visualizing points using Python and Folium, from scratch.

Part 3: Intelligence

Once we figure out the problem, we would want to understand why is that problem occurring. We think it’s the intelligence layer on top that helps answer the “why” question — the most difficult among them all. This intelligence can be derived using advanced analyses like correlations, similarity analyses or predictive models.

For example, if business users know that profitability in area x is reducing, can they pinpoint the top three reasons why that might be happening? Extrapolating it further, if they know that the demand tomorrow evening in area x is going to mitigate, can they include areas x and y in the same list and be prepared for that?

The purpose of models built by a data scientist is to automate recurring decisions and eliminate humans who are a part of that loop.

However, there is one catch. The real-world is really chaotic and fickle because of which it needs human intervention to either monitor the performance of the model or to debug it.

For example, how is your surge pricing strategy performing? What is the impact of your models for assignment, redistribution and batching?

If something is not working out at a particular location and time, can a business user (who has the maximum context on the business) quickly tweak it and run it again?

Just as there is an experimentation culture for UX and UI in the online world, we believe model creation should also have that culture of experimenting in the real world so that its impact can be measured.

To read more about this, check out our take on experimentation:

Making Location-Based Experimentation a part of our DNA
What can we learn about experiments on the ground from web-based experimentation?

Part 4: Actionability

Insights are the most useful when they actually drive a business decision — either taken by a human or a model.

For instance, in the short-term tactical decisions, it might be about:

  • Doing damage control in case of an anomaly, for example, a sudden dip in revenues
  • Sending marketing promotions at the right place and time, for example in areas where people order more breakfasts as compared to dinners
  • Sending notifications to your ops teams or drivers to move from lower demand areas to higher demand areas

In the long-term strategic decisions, it might be about taking strategic long-term decisions or debugging:

  • How is my model my performing in different areas and what is the exact impact?
  • Where do I open new stores? Which areas should I focus to establish more partnerships?
  • Historic correlation changes in time and space. For instance, how do my partner cost co-relate with the distance with my users?
To summarize, at Locale.ai, the four main pillars that form the fundamentals of every experience we design for our users are Accessibility, Visibility, Intelligence, and Actionability. These processes should help in answering the who, what, where and why and then closing the cycle by effecting a decision.

Read Similar:

From Thoughts to Visuals —How we designed our geospatial analytics product
How do we apply “design thinking” to bring our ideas to life at locale.ai?
How analyzing supply-demand gaps can optimize your unit economics!
After all, we all need our own Google Maps.

If you want to delve further, check our website out or get in touch with me on LinkedIn or Twitter.