5 steps to building scale in the on-demand economy (Part 2)

For this week’s member profile, Veyo’s EVP of Technology, Michael Singer, has written a two-part guide to growth in the on-demand world. Part One can be read here.

Increase Value

If you’ve completed the first leg of this journey, congrats! You’ve proven your value. You have identified all the players in your ecosystem and you have created a revolutionary platform that works. Now it’s time to build the various aspects of your value chain and decide what features to build. More importantly, it’s time to decide where you should say “no.” With great growth come many many ideas. You’ll have more stakeholders, and everyone in your ecosystem will have good ideas, but with limited resources there is only so much you can build.

A good way to identify features that you should build is to look for features that will increase value for all participants in the chain. By focusing your development efforts on increasing value, you set a defined criteria for adding product features and ensure that every feature added is measured based on the value and impact it generates throughout the ecosystem.

A good example: Some Veyo IDPs are more consistent in their online time than others. That means that they like to drive every day or every week at the same time, while others prefer to drive at different times and on varying schedules. Both types of IDPs drive the same amount of hours, but respond differently to trips dispatched to them. If we can identify patterns in our IDP schedules and dispatch trips accordingly, we can ensure that more trips will be accepted and more trips will be completed. As demand increases and more drivers join the platform, we’ll be able to dispatch trips in the most efficient way, ensuring that our members get to their appointments on time.

Predict and Influence

One of the cores of building a successful transportation network is being able to predict supply and demand and identifying what influences both. Data science provides multiple models to use in these predictions:

  • Time series analyses that use historical data to predict IDP participation

  • Neural networks that predict the probability of a trip being canceled

  • Survival analysis that can be used to predict driver churn.

These are just a few examples of the models we use at Veyo. By identifying the different variables in your system and proving out the solution, an abundance of data can be generated. You can then use that data to identify how to predict and influence behavior in your space.

But data needs careful analysis. Did you know that multiple studies show that an increased consumption of ice cream correlates to a serious hike in wildfires? Did you also know there is a big difference between causation and correlation?

When people eat more ice cream, there tends to be more wildfires - that is the correlation. But wildfires are not caused by people eating more ice cream. The cause of the wildfires and the cause of the increased consumption of ice cream is the hot weather - that’s causation. It’s important to know the difference and dig through your data. Identify what is truly causation, and don’t let random correlation mislead you.

Data science, as the name implies, is half data, half science. When testing models and ideas, build a platform that encourages scientific research. Approach experiments with scientific methodology by asking questions, searching for answers, building a hypothesis, experimenting with control groups, and testing different approaches.

To be able to run experiments, the platform should easily allow you to select cohorts, A/B test, and access data for research and results. And don’t just allow data scientists access to the data. For better results, it’s important to democratize data access, meaning that everyone has access to data and that the data they’re accessing is secured and cleansed. Creating a secure star schema data warehouse that contains cleansed and accessible data is usually a good place to start.

Automate and Scale

Each time Veyo adds a contract, there are more trips to dispatch and more IDPs to manage. Each time we add IDPs, we could add more support center agents. And each time we get more trips to dispatch, we could do it manually. But neither of those solutions are scalable. The incremental cost (and time) of adding additional agents and manually dispatching would quickly become overwhelming. To truly achieve scale, you need to automate your processes.

There are many opportunities to automate and it’s important to note that every process in the organization can always be improved. The most important thing is figuring out where automation will bring the most value. For Veyo, our core functionality is the dispatch and distribution engine, and it is fully automated. Everything relies on that automation, and it’s revolutionizing NEMT. But how about something like the code deployment process? There are many opportunities for automation that are not within the immediate problem being solved (i.e. getting a car to someone on time) but are still required to achieve scale:

New Market Implementations

Create a handbook on how to build out a new market. What are the statistics that are required? What tools does your platform provide? How does the onboarding flow work? Set up a process that creates proven step-by-step flows to get a new market into full scale. Don’t reinvent the wheel every time you open a new market.

Development

Use continuous integration tools to automate your software deployment flow. Invest in QA automation to easily deploy new features without risking major setbacks.

Sales

Use (don’t just buy the licenses...) a CRM platform. Identify the happy path from lead through qualification to customer and beyond, and build that process into a platform. Let the platform drive the flow and use its capabilities to automate campaigns and outreach efforts.

Look for any opportunity in your process to drive efficiency, improve, and automate. In the future we will all be replaced by AIs. It’s time to embrace that and move on.

Final Thoughts

The philosopher Jeremy Bentham talks about the distribution of happiness and asks, given a finite amount of happiness in the world, should we distribute that happiness equally and have everyone be mildly happy or should some people be happier than others. Is it possible to disrupt that equilibrium? Is it possible to make some people better of without making any one individual worse off? Wouldn’t that end up making the world a better place, and after all, wasn’t that what we set off to do?