Part 3: Models for Impact

In my last installment, I went over strategic approaches for impact innovation. This week, I’m getting more pragmatic: how does an impact innovation fit into a mission-driven organization’s business and what becomes of it over the long-term?

Business Models for Impact

Every organization, even a civil society organization, has a business model. This is the way the work is funded, what the work is, and who benefits. An organization’s business model determines the innovation path that is available to them, and vice-versa. Clarifying this relationship early is useful to plan around to avoid disappointment later on. I think it’s also useful to keep in mind to avoid personalizing criticisms–it takes all kinds to make a world because some choices foreclose some paths.

For example, Faradj Koliev published a recent paper illustrating a point I often make: substantive change is often the result of pursuing multiple approaches simultaneously, in this case international pressure and interpersonal relationship building. But, a single group can’t credibly pursue both. A business-friendly model may reduce barriers around privacy and collaboration, but it can often fail to create real momentum or urgency. An advocacy organization may consider naming-and-shaming key to their theory of change, but this forecloses the possibility of providing data or services to firms that may get named because potential clients would understandably worry about a conflict of interest. In my experience, it’s best to accept that reality, but not give up. Instead, consider partnerships and coalitions, subsidiary relationships, spin-outs, or other ways of resolving the essential conflict in the business model.

A key difference between NGOs and businesses is the separation of revenue and beneficiaries, and funders should weigh this in their strategies as well. Philanthropic funding has advantages over other revenue sources, such as the ability to work openly to demonstrate a point or seed new activity. There are some activities that may never be self-sustaining, or may be thought too risky, but still have high potential to be impactful.

Revenue Models for Impact

A revenue model is a more concrete plan for funding an organization’s work. In simple terms, it is a budget in a spreadsheet, with a forecast that extends a few years into the future. Most nonprofits and their activities begin with philanthropic funding, like grants or giving from small donors. A great technology innovation can become part of the organizational revenue model, as well.

An impact organization that wishes to generate revenue out of a technological or data innovation may do so by licensing it in a commercial arrangement, or by generating revenue from the delivery of the data or technology. Universities, for example, often have contractual agreements with researchers that assign the university a percentage of ownership for patentable intellectual property, which offsets the initial costs of filing and aligns financial incentives with practice and therefore impact. Nonprofits may also operate a cross-subsidy model, charging for-profit clients while subsidizing access for researchers, students, or other types of users. 

Organizations that prioritize openness of intellectual property sharing like OpenCorporates and Wikimedia, for example, commercialize delivery of their data. They allow nearly anyone to search their content, but for-profit organizations that wish to integrate the data with their own systems require a commercial license. Wikimedia’s enterprise page explains the value proposition quite clearly. The license fee pays for API access or an automated regular complete delivery.

Sometimes an intervention belongs with a consortium of dues-paying membership, like an industry association. This is particularly suited to pre-competitive standards and shareable frameworks that benefit entire industries through network effects.

Organizationally, an NGO may be able to build out a self-sustaining revenue model, but this is not the only legitimate goal. Some organizations and interventions may have an end date or a wind-down plan that is part of their broader strategy: building up a capacity, proving out its usefulness, and then transferring it to government or markets. Indeed, as we saw during the massive contraction in development aid in 2025, to avoid an abrupt and calamitous exit, it is a good idea for an organization to have at least a basic plan for what should happen if the organization experiences a collapse in revenue. Having an explicit model of sustaining revenue versus conditions for wind-down can help an organization navigate those tough moments with clarity.

Exit Models for Impact

An impactful innovation may begin its life one way, as a proof-of-concept funded by a foundation, for example, but it may not necessarily achieve scale or sustainability under that model. Over the long term, some philanthropic donors will tire of supporting an innovation that is mostly stable, and stability is often required for businesses or governments to invest in adopting your innovation. They don’t want to waste resources like staff time, creating a dependency on a product that undergoes major revision in every two-year grant cycle. The Robert Wood Johnson Foundation commissioned a good overview here (pdf) of how an organization can plan for a responsible exit, which is written for the organizational level, but much of the advice can be applied at the intervention level as well. For example, announcing in advance that you are ending support for a tech innovation gives others the opportunity to step in before the expertise that developed it moves on.

My mentors from the Global Development Incubator encouraged me to think of interventions as having multiple phases that may require different models to support, and ultimately, an exit path. A business exit is defined by what happens to the equity. In simplest terms, it can be self-sustaining and remain private, it can be acquired, and it can go through the process to be listed on a public securities exchange (IPO). 

An impact innovation can undergo a different kind of exit, and planning for it can make the difference between a tool that becomes a dead letter and one that makes a lasting impact. In some cases, they become self-sustaining and run by the organization that developed it.

An impactful innovation can be licensed to other entities, a path that many research organizations like universities pursue. This path requires some infrastructure and know-how, like lawyers, but planning for this early can help you signal that entering a licensing deal with you will not create undue turbulence, like the abrupt collapse of an agreement the business has depended on. It can also be a fast-track to impact because you can leverage the know-how of an organization that is already scaled instead of scaling yourself. I’ll go deeper into licensing for impact in my next installment.

The innovation may be so demonstrably useful that it can be transferred to the government, or another durable institution takes it on as a function. For software and data, this might be something like the Linux Foundation, a library, or a data trust. For example, Databricks developed the machine learning development framework MLFlow for its own platform and later transferred it to the Linux Foundation, which specializes in supporting open source software, hardware, frameworks, and data. It’s worth noting here, however, that MLFlow was always open source–an open source licensing model does not automatically translate to impact and adoption.

The innovation may be reduced to a lightweight framework that allows people to replicate it in a decentralized way, allowing it to be absorbed by the target entity or the commons. Finally, it may undergo a managed sunset; sometimes drawing attention to your organized withdrawal can encourage others to step up, and documenting it carefully can help others pick it up again in the future. 

Exit readiness is something you can build at any point, and the earlier the better. It helps clarify what constitutes success, even if it’s not what you imagined at the outset. One cultural norm of entrepreneurship is worth borrowing: know that some new ventures and ideas come to an end, and it is not a reflection of your personal worth. Personally, I found it helpful to have a dog (any pet will do), who could never understand that I had failed at something, even if someone had tried to explain it to him. You are not your work, and people will still love you.

A young wheaten terrier looks out the window with a tennis ball in its mouth.

Spyros Vierthaler, photo by Shannon Stewart. All rights reserved.


Thanks to Ben Savonen at GDI for responding to an earlier draft of this piece.

Part 2: Mind and Hand

Last week, I talked about the most common advice I give when a mission-driven organization already has a great idea. This week, I want to back up and begin at the beginning: where should we, as mission-driven organizations, apply ourselves to innovation and how do we do it?

I started thinking about this systematically and strategically when I worked at MIT. MIT’s cultural ethos is unlike any of the other universities I have studied or worked at. Their motto, “Mind and Hand” traces back to its founding as a technical university–not pursuing knowledge for its own sake (mind*), but with the primary aim of putting knowledge into practice (hands). 

We think of innovations as iconic inventions like the steam engine, the Haber-Bosch process, or the computer, but in my view there is a subtle difference between invention and innovation, mind and hands. An invention may exist as a defensible piece of new intellectual property, but it only becomes an innovation by being put into practice–often a much more difficult journey filled with frustratingly mundane obstacles. Counterintuitively, innovations are rarely completely novel. Often, the same invention happens multiple times, independently, but the superior invention doesn’t always win out in terms of impact–the superior innovation does, the one that made itself amenable to adoption. Crossing that last boundary to practice requires extraordinary persistence and luck, and innovators know this. A great idea can be thwarted by difficulties in staffing, in supply chain, in sales. Having ideas is easy; doing things is hard.

Innovation can be strategically and purposefully cultivated as a way of fundamentally changing a system’s dynamics. The first part of developing a strategy to apply deliberate innovation is to create a theory of change. If this isn’t an exercise you have contemplated yet, I recommend starting there. Others have better methods for developing a theory of change than I do, so I refer you to guides like this. The end result of this process should be a conceptual map of the stakeholders and forces in the system you are targeting.

In this systems-level map, briefly inhabit the perspective of each stakeholder and ask, what are the goals, activities, and authorities of each player in the system? Then ask, would amplifying that part of the system have the impact you wish to see? Are their goals aligned to your impact mission or not? Try to be parsimonious and charitable in your assumptions, and do not chain together more than a few unknowns. I typically sketch out a literal diagram at this step, just to structure my thinking.

When you identify a stakeholder whose empowerment would have the impact you want to have, ask whether there is an existing process that needs to be automated or scaled. When there is a stakeholder whose activities or goals are working counter to your desired change, that is a candidate for disruption.

Now that you have identified the stakeholders, their goals, and their activities that you wish to either amplify or disrupt, it is time to consider an innovation strategy. There is scholarship on the theory and practice of innovation, an engaging and interdisciplinary field at the nexus of the history of science, organizational psychology, operations research, and management. If the topic interests you, one of my many brilliant mentors, Anthony Sinskey at MIT recommended a book, Doing Capitalism in the Innovation Economy, to me as I started to think deeply about the topic. I have found Matt Ridley’s How Innovation Works another nice overview, with a more historical perspective. I also recommend having a look at a business school’s course syllabus and readings for a class on strategic innovation, many of which are available online. It’s also perfectly acceptable to make a visit to your local university’s library and access texts from their collections or subscription base–public university libraries are especially likely to encourage community members to use their resources and sometimes offer borrowing privileges.

In actual practice, here are the methods I have found fruitful ways to begin innovating strategically. 

At very large scale and moderately long timelines, like national and state-level government, undirected openness and generosity with data and intellectual property can encourage innovation. Many countries make a variety of non-sensitive data and intellectual property available, which people have turned into impactful innovations by stitching it together with other data or surfacing it at the right time or place. The time scales are long, and the economics support a more open-field approach to innovation that tolerates at least some kinds of failure. 

At a smaller scale, strategic innovation is still possible, but the mechanisms are more directed. I think of them as innovation by analogy, remix, amplification, and disruption. 

Innovation by analogy works by defining a problem or a goal and asking an expert in an unrelated field how they would solve the problem. This is more than just the luck of the naive (although that contributes to it). Experts in unrelated domains bring their own epistemic tools about how problems can be solved. I’ll give an example from my own life: when I was in graduate school, I met to cook dinner and socialize with a group of about seven other students who worked in other fields every Thursday night. There were experts in computer science, engineering, archaeology, physics, Chinese literature, and myself, a biologist. Most weeks passed with just a bit of yapping over cheap beers, but some weeks, a member of our group would come to dinner with a problem they were having in the lab. One week, I was at the end of my rope, having been doing delicate dissections of fruit flies with needle-sharp tweezers for weeks. The constant rubbing of the tweezers and insect pins had magnetized them, and now the whole process was being thwarted because I couldn’t release the pins easily. I complained about this to my friends, and the physicist perked up: he needed to degauss his equipment frequently for his experiments, and he could quickly put together a degaussing circuit for my tiny tweezers. No biologist in my lab had a solution–it just wasn’t a common problem in our field. But, an expert from another field immediately saw the analogy to a problem with a simple solution from his point of view. Nearly every member of our group had a similar experience in our years of graduate school–sometimes you just need an expert who is equipped with a different set of tools to look at your problems. 

The next strategic approach to innovation I have found useful in practice is remix. From my perspective, this mode of innovation gave birth to much of the field of data science, one of the hot job titles of the decade and one I held myself for a number of years. Remix means treating existing data or methods as modules that can be stacked or chained in new ways to produce a novel product or insight. Much of the work of data science is stitching systems together at the data level, finding records that could or should exist in both systems and knitting them together to produce new insights. You can see in my work on this module that I took two pieces of data about garment manufacturers and shrimp farms in India, the records of their incorporation and data collected by India’s Customs authority, to produce a new insight: to a first-order approximation, you could identify specific manufacturers in the global supply chain that were likely to be using forced labor.

Sometimes an innovation is as straightforward as amplification of an existing process or idea. There may be some actor in your stakeholder map from the theory of change exercise whose process should have the impact you want to see, but they are too small or disempowered to tip the scales. Get into their process, ask them about what they wish they could do or know, see what they are doing manually that can be automated or something they produce that needs to be surfaced to someone else at a crucial decision point, and make it happen. Transferring manual work to machines in itself is morally neutral–it depends on what you’re doing and why. Sometimes it’s work like supply chain mapping, a process that in human hands is slow, costly, and tedious work that often lags behind other processes so much that the full picture arrives too late, and the facts have already changed. The process needed to be amplified and scaled by machines, using data that was collected by a more large-scale process, something that was noticed by many people (myself included) in parallel in the last 15 years. 

Finally, another mode of strategic innovation is disruptive. It is foolish to direct disruptive innovation at any old thing; no one wants disruptive innovation to interfere with the functioning of a hospital or a school. There is a time and a place for disruptive innovation, and in my experience, there are some key features to look for. In some cases, one of the stakeholders in the map is rent-seeking, and their power over the system can be strategically disrupted. In 2003, I was living in Berkeley, California, and I can attest that the taxi medallion system in San Francisco had created such a shortage of taxis that the idea for distributed ride sharing was obvious to anyone who had spent hours in a rainstorm trying to hail a cab. Rent-seeking means there is a lower friction path that a process would take if it were available. Next, there are cases where industry fragmentation itself is driving exploitative behaviors. Many smart people in the field of labor rights have noted that a highly fragmented industry of employee-pays labor recruitment has created perverse incentives for the recruiters, and the recruiters themselves have low margins and no plausible exit. The result is an industry rife with exploitative recruitment businesses, many of which would leave the industry, except the business itself isn’t worth anything and does not accrue equity. This represents an opportunity for disruptive innovation. The incumbents are too fragmented to respond to a new entrant with a more scalable model in a coordinated way, and a rapid consolidation would enable a more sustainable (low-margin but large-scale) and less exploitative business model to develop. At the time of this writing, to my knowledge, the field is still looking for the right disruptive innovation. 

Now that you have assembled the picture–the system, the forces, the goals and activities, and you have identified leverage points for innovation, the work begins. Try to find ways these techniques might encourage breakthroughs that empower the stakeholders aligned with your impact.

Next week, I’ll be attending events related to UNGA (get in touch here if you’d like to say hello), so I’ll be returning in two weeks with a discussion of business models, revenue models, and exit models for impact innovations.

*We can safely leave this to Harvard.

Part 1: If people only knew!

I have spoken with many impact organizations with a mission to improve the world and change peoples’ behavior, especially corporate behavior. They have collected crucial data, developed an innovative tool, or have a new framework to approach an unsolved problem. But, they are failing to gain traction and cross an uncharted gap between an initial proof of concept and uptake. If people only knew what they know!

They have a problem that is at once both strategic and technical that I am calling the Last Mile for Impact. Here is my most important advice: You have to meet people where they already are and look like you have always been there. They are bridging a gap between knowledge and action that exists both at a social, human level, and at an incumbent technology layer that shapes the way people organize information and act on it. 

The most direct and parsimonious path to impact is to begin with the most collaborative and positive starting assumptions, and to only take an adversarial detour if those assumptions don’t hold up. 

First, I consider the social dimension of the task. Here, it is important to begin from a position of non-judgmental empathy and to adapt your communication style to the norms of the stakeholder. Here’s where I usually start: people as a group are smart, and they want to feel good about their work. They hate feeling shamed, feeling stress or uncertainty, and getting in trouble with authority. Organizations respond to financial incentives, risk, or the threat of regulations. Above all, remember that both people and organizations had plans in place before you launched your idea, which creates inertia that your idea will have to overcome. Put yourself in their shoes and think about what could motivate them to your side. Adopting your idea must be seamless and appealing enough that they will set aside their existing plans to jump on board.

So, the first thing you have to consider is “Who is it that needs to be more empowered or make a different choice for me to see the impact I want to have?” This is both an exercise in imagining your target user and discovering their literal job titles. I would ask some of the following questions: Who decides which company to invest in or which supplier to buy from? Who decides what questions to ask? What are the decisionmaker’s concerns about risk? What are their practical constraints? What do they have to do that is tedious, high-friction, or overwhelming? How do they feel about their job emotionally and where does their stress come from? 

The next thing to consider is, “What technologies serve that information or organizational purpose today? Which ones need to be built for future requirements? What is not automated that could and should be?” There is almost always an incumbent solution or process, and it sets the expectations your potential users hold about usability and professionalization. You will need to know as much as possible about it in order to appeal to your target users. 

Usually with responsible sourcing, the most charitable assumptions to begin with are that smart, mission-driven people are working on this, but they have too much work to do and too little information to reach the decision you want. A clue! You have to make it effortlessly easy to find your information in the course of what they are already doing to meet their responsibilities. You have to show up where they already are. Keeping your crucial data in a separate tool or database that they have to check manually is adding extra work for somebody who is already overwhelmed. 

Your task is to find out as much as possible about their process to help you build toward something usable and flexible enough to gain traction. In my experience, you have a few options here: 1) you ask them where they get information, for example by going to conferences or making a survey 2) you read job descriptions for the personas you’re targeting and look at the technology skills they expect people to know or learn or 3) you search for solutions that seem to be targeting those people by offering or organizing the types of information they already know they need.

The next thing to find out is when they look for information where your intervention could make a critical difference, why are they looking? The first plausible motivation is reputation. Companies do not like to be publicly embarrassed, but embarrassment doesn’t cost money if there is no threat of action behind it. Usually the person responsible for avoiding reputational damage works in brand, communications, or marketing. When the cards are down, their opinion is a nice-to-have, but it almost always gives way to more immediate concerns. The next possible reason for seeking information about their operations is that they want to see this system together because they are trying to optimize it. This can be for cost reduction, resiliency, generalized risk, or some other reason (ask them, you have friends in the biz now that you met gabbing about SaaS tools). Those people may not be looking for your data, but they do have a lot of authority to make operational decisions. Finally, the decisionmakers with the most authority are helping companies comply with the law, or they are the law. This is the best place for you to position if you can. Map your information on to any law that might touch on it–should this be reported under CSRD? Would this help a company avoid strict liability under CERCLA? Would this information be useful to regulators or law enforcement? Again, make it easy for them to see the value. Walk them through the connections and business case. Get as familiar as you can with the context, use it to inform your strategy, and speak to it confidently. Reading a regulation and looking up every unclear concept is basically a super power.

Finally, the context also includes a technical layer that you will need to plan for. Given the research you’ve done, you should be thinking about things like “How do people in this domain define a company or other key concepts? What are the critical points where data is needed to facilitate this system, like organizational handoffs, safety regulation, or tax collection? How do people recognize entities across data sets in this domain?” This is not as easy as it may seem; a multinational company may be listed as one entity in a securities exchange, but it is usually composed of dozens of legal incorporations and many locations. Meeting people where they are means modeling your information around the needs and assumptions of the technical systems that are already in use. A record should ideally correspond to the existing systems, and you should support a way of finding records that relate across systems. A system that exists for investors might have a single ticker symbol for a multinational company, for example, so your data should relate to that atomic unit rather than a single domestic legal incorporation. Do not use names for record linkage; this goes for both people and businesses.

At both the social and technical levels, my advice is the same: meet people where they are looking, and do everything you can to fit in. The internet is rife with dead tech tools for impact. Some of them were great ideas, and they simply failed to cross that gap and achieve sustainable adoption. I don’t want to see that happen to you! 

In the next part of this series, I will discuss developing or articulating a theory of change and identifying opportunities within it to do strategic innovation. 

Feel free to join me in developing these ideas, and comment/contact me here!

Women Who Code: My Unique Path to Data Science

I’m bringing together some projects and earned media from across the web here. This is a podcast interview I did in 2023 when Women Who Code awarded me an Applaud Her Award for Python and designated me one of the 100 Women Technologists to Watch.

Unfortunately, Women Who Code no longer operates, so I’ve reposted the episode here.

Career Nav #76: My Unique Path to Data Science
Women Who Code

Forced Labor Automated Risk Estimator (FLARE)

Years ago, when I was a data scientist at the Global Fund to End Modern Slavery, my keystone project was a piece of original software that estimates the likelihood of forced labor at the supplier level. It was intended as a proof of concept to show that by observing only passively-collected operations data, I could, with reasonable accuracy, prioritize due diligence efforts across a corporate supply chain, no matter how large. This formed the basis of discussions with both large corporations and governments, where we demonstrated the feasibility of using new technology tools to enable the smooth enforcement of and compliance with stronger laws against the use of forced labor in globalized corporate supply chains. The time was right, but regulators needed to create the business case to invest.

The tool was recognized in 2020 by the Society for International Development, Washington, winning their annual Innovation award. It was a finalist from the Thompson Reuters Foundation Stop Slavery Awards in the Innovation category in the same year.

The Fund has since ceased operations, and most of the work I contributed is no longer hosted in a discoverable place online, so I am bringing together the key elements in their original form here.

The code repository is here, and I would draw your attention to the ReadMe, which explains some of the rationale behind choices that I made about things like “What is a company?” https://github.com/shannonrstewart/FLARE

A short intro video that describes the project:










A tutorial series that takes a tour through the codebase:

Part 2






Part 3

After I left GFEMS, I joined a tech startup working on this exact issue and worked with dozens of talented software engineers. There, I learned a great deal about data pipelines, scalability, and parallelization that is not reflected in this proof of concept. While I believe it demonstrates an important point about the presence of reliable mathematical signal, almost none of this code can be directly ported into a real B2B SaaS platform for runtime reasons. Nevertheless, the FLARE tool remains available for commercial use under an MIT license, and I am happy to answer questions about it.

People for Profit: North Korean Forced Labour on a Global Scale

Edited by Remco E. Breuker & Imke B.L.H. van Gardingen

Contributors

Jan Blinka
Britt C.H. Blom
Marte C.H. Boonen
Klara Boonstra
Rosa Brandse
Remco E. Breuker
Imke B.L.H. van Gardingen
Larissa van den Herik
Tycho A. van der Hoog
Marieke P. Meurs
Cedric Ryngaert
Shannon R. Stewart
Anoma P. van der Veere

With most of the world focused on peace talks, nuclear programs, and hard rhetoric between leaders, it is easy to lose sight of one important aspect with regard to North Korea: the people of the Democratic People's Republic of Korea. Exploited and forced into excruciating conditions, North Korean workers are sent across the world in order to earn foreign currency for a relentless regime.

Tracing this modern form of slavery from Taiwan to Russia, and Europe, this volume lays bare a portion of the extensive financial and labour networks through which the DPRK earns its hard currency, despite the hard sanctions the country faces from the international community.

Link to full text (pdf).

MDMA can help fight PTSD—if scientists are allowed to use it

Some psychiatrists are developing a method for treating Post Traumatic Stress Disorder (PTSD) by using MDMA to suppress flashbacks and acute stress during cognitive behavioral therapy sessions. It has successfully completed Phase II clinical trials and is moving to Phase III. But, if successful, the approach faces an additional barrier--the Controlled Substances Act.

Read more.

IMAGE BY THE MARINES VIA FLICKR

Leveraging Industry-Academia Collaborations in Adaptive Biomedical Innovation.

Stewart SR, Barone PW, Bellisario A, Cooney CL, Sharp PA, Sinskey AJ, Natesan S, Springs SL.

Clin Pharmacol Ther. 2016 Dec;100(6):647-653. doi: 10.1002/cpt.504. Epub 2016 Oct 22.

Biomedical firms are increasingly looking to innovation alliances with university-based researchers to refresh their R&D pipelines. This paper explores some of the models in use and their relative advantages.

Read more.

Thumbnail image licensed under Creative Commons, thanks to Boegh on Flickr.