Overview
If you are managing a program and want to improve outcomes, data is necessary. This course was designed to explore important themes in the world of data and analytics in the public and nonprofit sectors. Labs were designed to provide some exposure to the types of data and models discussed in the readings.
The cases covered in class demonstrate that rich data can be generated in a wide variety of ways, and often at a low cost. In the Ohio Schools example, scores on assignments were used to test hypotheses generated by teachers in the data room. In the Iceland example it was a national-level survey developed by professional social scientists and refined over a long period of time. In the Gottman example it was 15-minute interviews broken into one-second thin slices. In the Digital Humanitarian examples from Meier is was massive amounts of information from social media streams that were cleaned and verified through crowdsourcing.
Prediction can be a valuable exercise because it forces you to identify an important outcome, define what success looks like, and develop a way to systematically examine which factors impact the outcome. Even if you don’t build an analytical tool, the process of brainstorming ideas with a team, thinking about how outcomes might be operationalized, figuring out what data is available, and predicting which variables are most likely to product success will help you think through your program model. You will learn a lot, and the way you think about the problem will likely change.
The final project is an opportunity to think creatively about building a predictive application for your current organization. These tasks can include things like automating the digitization of old forms since automation is by nature a predictive exercise (did the person check the box on line 5?). Or it could be more accurate forecasts of future events like crews needed to fight forest fires during a drought. In the Moneyball example the Oakland A’s used a model to improve their process of hiring team members by predicting how much each member would contribute to the totals runs scored by the team in a season (predictive HR).
Memo overview:
For your final project write a short memo (2-page, single-spaced) addressed to your boss or your board to request support for your idea about a new predictive system. The memo is short because you are focusing on the value the project could bring to the organization, not on the details of implementation.
You need to convince the decision-maker that the project is feasible. Provide enough detail to show that the technology exists, and that the data required can be collected in a cost-effective manner. The final ask on the memo should be permission to create a full proposal, timeline, and a budget for the project for further review. This strategy allows you to focus on getting support for the idea without having all of the logistics fleshed out up-front (I don’t expect you to have that level of detail).
I welcome creativity, and am open to variations on the project that would make the assignment more relevant to your current job or future aspirations (you are free to write the memo for a hypothetical future employer if that is more useful).
Effective memos will do three things:
- Be clear about the predictive task (1-2 paragraphs).
- Explain how the system will bring value to the organization (1-2 paragraphs).
- Demonstrate that it is feasible (1 page).
Your new system can generate value if it targets an important outcome for the organization, such as annual fundraising for a nonprofit (Which grant applications are most likely to succeed? Or which messaging on the annual appeal will generate higher rates of giving?). Or it can generate value if it provides better inputs into common organizational tasks (predicting decision fatigue by judges to schedule onerous cases when they have the most mental acuity).
Note that new predictive systems are often characterized by high fixed costs and low marginal costs. They are typically expensive to develop, but cheap to operate. You might try to find a task within your organization that has a high marginal cost or high volume so that the costs of developing the system are off-set. For example, new windows are expensive to install but you might save enough in energy costs each month that the windows will pay for themselves in 5 years but will last 20 years. You do NOT have to estimate the cost of implementing the new system. However, picking a repetitive or expensive task to automate will help strengthen your value proposition.
Example:
To give a concrete example of a reasonable proposal, in Lab 04 we discussed the value that trees bring to cities by making buildings more energy efficient, cleaning air and water, and preventing floods. For each $1 a city spends on planting new trees, it gets $2.25 in returns.
In the memo you could make the case that a predictive model to count trees would be better than humans. A city would gain from better understanding what sorts of public programs and incentives lead to canopy growth in each census tract, but this sort of data-driven approach requires accurate counts of trees at regular intervals. A tree census is expensive: it took New York City 2 years and 12,000 person hours. Machine learning models can use LIDAR data to accurately predict the number of trees in each census tract. For example, Descartes Labs has an elegant model to do this and there are free open-source solutions available. Most large cities collect LIDAR data on regular intervals. This model will require some resources to develop, but it would cost less than a 2-year census requiring 12,000 person hours of city time and could then be replicated on a regular basis when new LIDAR data is available. Accurate measures over time help to better understand factors leading to canopy growth or decline across neighborhoods in the valley.
Deliverables:
- Submit a 2-page single-spaced memo addressed to the boss or board in your hypothetical scenario.
- Conclude the memo with an ask to develop a full proposal based upon the potential for the application to add value to the organization and some demonstration that the technology is feasible.
- Use footnotes to cite relevant material.
Grading:
Your memo will be graded on how well it achieves the three criteria listed above. Primarily, though, I am looking for you to demonstrate that you can apply the readings and labs from the semester to your own current or future professional context. You should need only a couple of paragraphs explaining the task.
The value proposition is a chance to articulate generally how data, analytics, machine learning and AI can benefit organizations. Demonstrating feasibility requires some knowledge about existing technology, so you can draw upon examples from course readings, or you can do some background research on applications relevant to your organization. The type of technology will vary greatly depending upon your organizational context. I don’t want specific details on hardware specs, brands of software, storage requirements for data, or performance times. Rather, I am looking for examples of similar technologies used in similar organizational contexts to demonstrate that the technology to accomplish a specific task exists, and the data needed for the technology is available.