My first internship came about because of a sub-bullet in a UN resolution. I’m not sure that whoever wrote that the UN “Decides to initiate, in 2016, a virtual dialogue on Harmony with Nature among, inter alia, experts…” (A/RES/70/208) thought the way it would be carried out was by a college freshman writing SQL queries in her dorm room at midnight, but that is how the implementation happened in practice.

This was one of my first experiences of the peculiarity and unpredictability of the feedback loop between policy idea and policy implementation. Proposing and passing a good policy is the part of the process that often garners the most attention, but it is only the first step towards implementing actual change. I became deeply interested in the ways that policy and implementation actually interact in practice. How do implementation capabilities change what policies people decide to push for, and how does a policy that gets passed actually translate logistically into someone taking an action? How does one ensure that the original intentions don’t get lost in translation?

Through the Impact Fellowship, I’ve had the opportunity to closely examine various parts of that policy-implementation feedback cycle at Code for America. I’ve been working on the Clear My Record team, which is working to empower the government to automatically clear all eligible criminal records.

Code for America works in an interesting space — unlike other civic tech organizations, which work on existing government projects, Code for America works to show governments what’s possible with the mindful use of technology, particularly when you bring policy experts and implementation experts to the same table. For example, my team builds proofs-of-concept to show that automatic record clearance is possible. Our current work is on helping a legal aid organization in Cook County, IL, the second-largest U.S. county, automate parts of the paperwork process to clear eligible criminal records. Illinois doesn’t have data on who record clearance would affect, or what kinds of charges are on records that do get cleared. As a result, policy arguments currently being made are entirely anecdotal and often consist of overdramatizations and fear-mongering.

The data from the tool I’ve been working on paints a picture for the first time: the average age of someone seeking record relief in Cook County, IL is 41 years old. Their most recent charge is, on average, from 18 years ago. The most common items being expunged from records are dismissals and nonviolent drug possession charges. That means that the average story of record relief in Cook County is mostly about middle-aged people with old, non-violent charges, who are blocked from various life opportunities because of it. Once we tangibly gather data like this and show policymakers the impact that record clearance would have on people’s lives, they are more likely to take action to implement the solution — which in this case would be a policy of automatically clearing eligible records instead of requiring an arduous, sometimes yearslong paperwork process.

Even once legislative record relief is passed, the policy implementation loop hasn’t been closed. The loop isn’t completed until the person living with a record experiences an actual life change. If our prototypes convince the government to pass laws that grant record relief, but people don’t know how to navigate the new opportunities they now have access to, then change hasn’t really been affected.

For this reason, Code for America’s product culture is cognizant of users’ offline lives and the potential mental states that they will bring to the products. In a recent product discovery process I participated in, my team tested all our prospective copy with trauma-informed design principles. We found that using language like ‘person-to-person crime’ instead of ‘violent crime’ dramatically increases users’ feelings of comfort and information processing, and their ability to complete an application flow. Changes like these thus also dramatically increase the number of people able to actually access the life changes we are telling them about.

This comes into play across all Code for America products and is something I’ve kept in mind as I’ve also been leading product work on Code for America’s unified design system, Honeycrisp. Unlike with a leisure consumer application, people seeking government assistance through Code for America services are not looking for elegant animations or novel color combinations, or necessarily to enjoy their time using the product. Instead, the product needs to be as usable, predictable, streamlined, and goal-driven as possible. Across the organization, this means a consensus on avoiding dark patterns: the shorter the time a user spends in the application, the better, and the more likely the application will be able to do its job and produce an actual life outcome for a user.

Thinking in terms of this feedback loop inevitably leads to broader conversations about the criminal justice system that our work is situated within. Doing product work at a nonprofit organization means my team gets to have a lot of values-based conversations that might otherwise not be possible, and we can make space in our OKRs for exploring and defining our values.

For example, there is a lot of national momentum right now for automatically expunging old marijuana charges, in tandem with marijuana legalization efforts. Constraining our record clearance efforts to nonviolent marijuana charges would make my team’s jobs all easier. But my team has identified in our values conversations that we believe in pushing for a broader definition of eligibility. Some kinds of serious, violent charges definitely should stay on criminal records — but what would a world look like where we didn’t feel the need to obsessively track all of a person’s interactions with the justice system forever? During the course of my fellowship, I’ve seen a lot of criminal records. One had multiple jaywalking charges, calling to mind this 2017 NYT piece on Jaywalking While Black. On another person’s record, the majority of the charges were not for any actual crime but for “resisting arrest”. It’s not enough to work within a system that sees this as normal: we need to completely rethink it.

An important part of this work is to convince people who haven’t had interactions with courts or police that they are still tangibly part of the American criminal justice system. If you exist in America, your criminal record or lack thereof is not only a function of your personal actions but largely a function of the judgment of the police and courts surrounding you. This convincing, again, can be thought of as part of the same policy-implementation feedback loop. For a concept to get into this loop at all, there needs to be momentum from the public.

The product I have been working on is, on some level, just a complex PDF filler. The meaning-making of why that matters — to the Cook County courts, to Code for America funders, and to the American people — is arguably the most important part of my role. I came to this fellowship with a lot of normative ideas about what kinds of policies would improve American public life, and am now able to back those opinions up with much more tactical knowledge about how to effect change.

Amita Shukla is a 2020 Impact Fellow at Code for America. She graduated with a B.S. in Computer Science and a minor in Electrical Engineering from Columbia University. She has previously worked at the United Nations, the Department of State, the Department of Defense, and the American Civil Liberties Union on building better governance systems.

Clearing the Way for Record Clearance was originally published in Coding it Forward on Medium, where people are continuing the conversation by highlighting and responding to this story.

Can you really change your employees’ lives for the better if they don’t have office chairs?

As the first speaker in the Harvard Franklin Fellowship Tech x Social Impact speaker series, we heard from Wendy Gonzalez, the current interim CEO at Samasource.

Samasource is a perfect case study in incentives — how can a social impact company really work?

Samasource’s founding principle is to ‘Give Work’ rather than aid to people in developing countries. Samasource primarily connects Fortune 500 companies which need AI classification work with workers in India, Ghana, and Kenya. Most of its employees are women. By Samasource’s own estimates, it has improved the lives of 50,000 people since its inception in 2008.

Samasource is a perfect case study in incentives.

Samasource hits all the conventional talking points around creating a sustainable social enterprise. It works with big name companies, focuses on female representation at every level of the company, and brings advanced technology to impoverished Africans.

But while Samasource does make more provisions for its employees than it strictly has to, especially in that it provides hotels for its employees during COVID-19 — how much praise do they really deserve if they won’t even provide reasonable ergonomic support for their office workers? And when they have their employees do jobs that have been proven to cause PTSD for less than an hour in US minimum wage per day — is that really laudable?

So even given their relative success —I am unconvinced that the Samasource model is one that we should encourage other companies to emulate.

At first glance, they seem to be providing a path to a better future for their employees, who largely have no experience with technology whatsoever. Samasource says that they are employing workers in AI and setting them up for future careers in technology. But the assertion that they are employing Kenyans ‘to do AI’ is highly misleading. While Gonzalez asserted that workers were being positioned to take on real AI work in the future, and to become dataset consultants, workers spend most of their time labeling objects in images, which is essentially a manual labor job requiring no skill or insight.

And how much of their workers’ progress should actually be attributed to employees’ learning through working at Samasource, and how much to just a living wage? Previously, Samasource had a partner organization, Samaschool, that gave employees technical training. Since Samasource has become a for-profit, that side of the organization has been discontinued.

The rationale for this, as explained by Wendy Gonzalez, is that Samasource is intended as a bridge employer; employees are not really expected to stay past three or four years. It is intended as a stepping stone rather than as a permanent career. Personally, this reads to me as a rationalization for not providing any real upward trajectory for employees. Gonzalez cites the possibility that workers will move on to become consultants for dataset quality and be able to advise on companies’ data labeling needs, but it is not clear at all how this follows from employees’ manual labeling experience. The primary form of learning and employee growth is at a much more basic level — they obtain mousing skills, and familiarity with and access to technology at all. To be fair though, the HFF fellows asked about similar organizations looking to ‘give work’, but Gonzalez pointed out that they all require much more extensive technical background from applicants. Samasource accepts applicants without any technical experience whatsoever, in particular, people who have never used a computer at all.

This works only because these employees have so few other options. Samasource only operates in places where the average wage is less than $2 USD per day. Previously, it had attempted to operate in New York, California, and Arkansas, doing retraining programs, which were much less successful at changing workers’ living conditions — most participants in those programs remained impoverished after the program. Samasource pays greater than the average wage in all the regions it serves. While the cost-cutting aspect of outsourcing the work itself is ethical enough, I am less convinced that the consequent saving on providing chairs, let alone healthcare, is all that justifiable.

But the structure of the organization itself demands these types of tradeoffs. Samasource recently transitioned from being a nonprofit organization to being a for-profit organization. However, the for-profit’s majority owner is a sister nonprofit. The founder of Samasource, Leila Janah, originally envisioned it as a for-profit, but at its founding, the ‘Give Work’ idea was not compelling to investors. When it became possible to do so, they transitioned. Gonzalez underscored the idea of a ‘double bottom line’ — measuring Samasource’s success both in terms of social impact and fiscal profits. This is legally implemented through the nonprofit’s majority shareholder status.

The level of detail with which Wendy Gonzalez oversees Samasource is formidable. Most for-profit CEOs would not consider their employees’ day-to-day experience with this level of granularity. There are quite clearly more moving parts in running a social enterprise than there are in running a pure for-profit.

Is Samasource enabling the white-savior complex?

But I’m concerned that Samasource is too much of a generator of feel-good, photogenic moments for the Fortune 500 companies employing its workers. The companies get to say that they are uplifting the lives of Africans, when in fact they are outsourcing boring, and in the case of content moderation, dangerous, work that Americans frankly don’t want to do. It smacks of the corporatized version of white Western girls ‘picking up random African children and posting them on Instagram’, as VICE put it. This turns something that is at best ethically justifiable into something that is deeply self-congratulatory.

The fact that the business model is totally dependent around providing less for employees is truly questionable.

Samasource doesn’t need to provide healthcare when it employs workers in Kenya. It doesn’t even need to provide proper office chairs.

At the same time, it’s important to recognize that Samasource is criticized more than most other companies when it is in fact doing more than most other companies. Companies that are trying to be ethical should not be reward with impossible standards.

How can we incentivize companies to do better?

One solution is the B-Corp (benefit corporation), as my fellow cohort member Austin noted, which defines the best interest of the corporation as including some social responsibility.

Gonzalez felt that that legal form may not go far enough in holding CEOs accountable for employee wellbeing, and felt that the majority-owner nonprofit form was better. After our conversation, I am left feeling that even the majority-nonprofit status may not be creating an incentive structure that is truly worker-first.

‘Doing good while doing well’ is a fraught concept, and understanding the subtleties of Samasource’s operations provides a lot of insights into how that can both succeed and fail in various aspects of a business.

As someone attempting to learn more about what it takes to run a social-impact enterprise, Samasource is the perfect first case study.

What Makes a Social Enterprise? was originally published in Tech x Social Impact on Medium, where people are continuing the conversation by highlighting and responding to this story.