Key Takeaways
Pair human-centered design, live data, and iterative delivery to serve the public.
Three tools, one practice. The authors call this approach public interest technology: applying design, data, and delivery to advance the public good in the digital age. Design starts with the needs and routines of the people affected. Data helps teams understand what is happening and whether a change works. Delivery means testing in practice, learning, and improving. The three are connected, not a menu of optional techniques.
Technology is not the point. A new app cannot repair a harmful policy or a tangled process by itself. Governments matter because companies and nonprofits cannot take over core public responsibilities, such as delivering benefits or processing citizenship applications. The aim is not to make government look more like a technology company. It is to make public institutions more capable of helping people.
What's striking is that this approach treats service quality as a property of an entire system, not just its software. That aligns with socio-technical thinking, which studies how tools, work routines, authority, and human judgment interact. The framework also resists a familiar trap: measuring modernization by the number of services moved online. A digital form can be faster to submit and still leave people ineligible, confused, or waiting. One useful test is therefore not whether a tool is new, but whether the whole route from need to help has improved.
Do not digitize a broken process and call fivefold delay progress.
ELIS shows the cost. In 2005, U.S. Citizenship and Immigration Services began digitizing its paper-based immigration system. Eleven years and $1 billion later, just two of 94 form types could be processed through the new system. The first version slowed processing to five times longer than paper. Some officers had to undo what the system automated, while others relied on paper files and workarounds to finish their jobs.
The mismatch was structural. The agency lacked experienced internal technology leadership, goals were vague, and a contractor had too much influence over design choices. The team digitized forms without understanding the filing, note-taking, and case review that surrounded them. Later, observing officers at work helped developers build more useful features. The lesson is not that paper is always better, but that technology must fit the real work and solve a clearly stated problem.
This case is a vivid example of a socio-technical failure: the system treated the form as the work, when the form was only one part of a larger workflow. It also reveals a procurement problem. When public agencies lack technical capacity to question a vendor's choices, the contract can reward complexity or continued dependence rather than public outcomes. Still, replacing a contractor is not enough. Agencies need staff who can translate policy goals into product decisions and evaluate whether a system works for both employees and applicants.
Put applicants and frontline workers inside the design, not at its end.
The form hid the experience. Michigan's benefits application ran 42 pages, with more than 1,200 questions. Detroit resident Menica Harper described spending over two hours on it and still submitting an incomplete application. Civilla, a design studio, interviewed both applicants and caseworkers to learn where the process confused, burdened, or delayed them. The team also asked state officials to try the old form in a noisy waiting-room simulation. None finished within 15 minutes.
Research changed the result. Civilla and state staff developed and tested a shorter application, then trained more than 5,000 workers across 100 offices. The final form was 80 percent shorter, took workers half as long to process, and was completed correctly the first time 94 percent of the time. It reached 2.5 million residents. The work made applicants' needs and caseworkers' constraints part of the same design problem.
The redesign illustrates how administrative burden can act like a hidden eligibility rule. People who have the least time, stability, or confidence may be least able to complete a needlessly difficult form. In public administration, frontline workers are often treated as the final link in a policy chain, but their knowledge can expose where the chain breaks. Simulations can make that burden visible to leaders, though they cannot substitute for sustained contact with people who face the consequences in real life.
Use live data to spot need, then ask people what numbers miss.
Crisis Text Line learned in real time. The service began after a teenager used DoSomething.org's texting system to seek help during sexual abuse. Crisis Text Line later used messages and surveys to improve how counselors responded and how urgent conversations were prioritized. Its data showed that some signals were unexpected: the pill emoji was associated with life-threatening situations more often than the word suicide, while conversations about self-harm peaked around 4 a.m. The organization used such patterns to adjust counselor coverage and training.
Useful data still carries risk. Texting made it possible to see needs that broad, delayed surveys might miss, but the data was sensitive. The book recounts staff allegations about the former CEO's management and use of race-related counselor productivity data. Data can improve a service, but it can also expose people to misuse if access, purpose, and oversight are weak.
The service demonstrates the value of data that emerges from people's own language rather than only from categories designed in advance. Yet a predictive signal is not an explanation, and association is not proof of cause. A model that flags risk should support trained human judgment, not silently determine a person's treatment. Sensitive data also raises questions about consent, retention, and who benefits from analysis. Trust is part of the service itself: people in crisis must be able to seek help without becoming unwilling research subjects.
Pilot one real slice before betting the whole service.
Vermont tested a small piece. Rather than begin with a statewide benefits portal, Nava and state staff spent 12 weeks testing an online document uploader for SNAP applicants in one district office. They first recorded a baseline by reviewing 50 cases. During the pilot, applicants could submit documents while speaking with a caseworker and confirm that the files arrived.
The results justified expansion. The document phase of the process had taken about nine days. With the uploader, it took about four and a half days; 55 percent of pilot users received a decision within 24 hours, compared with 6 percent before. Forty percent used the tool outside office hours. The pilot supported a statewide rollout, but a larger integrated benefits project later stalled after internal champions left and difficult authentication questions surfaced. A successful test did not guarantee lasting institutional support.
A small pilot is valuable because it turns a broad argument into observable evidence while limiting the cost of being wrong. It can also reveal barriers that a prototype cannot, such as who maintains the service, how rules are interpreted, and whether staff can support it. The Vermont story makes an important distinction between validating a tool and building the capacity to sustain it. Incremental delivery reduces technical risk, but it cannot by itself settle political, legal, or staffing conflicts.
Fix the process before buying the app.
Some of the best fixes were ordinary. In Rhode Island, Marina Nitze mapped the foster-family licensing process before pursuing a technical solution. A medical consent form often went missing because it was separate from the application; stapling the pages together reduced that failure from 70 percent to zero. The application also lacked a field for Social Security numbers, which delayed background checks. Adding the field solved a problem that software alone would not have addressed.
A weekend cleared a backlog. The state invited 300 families with pending applications to an event with fingerprinting, training, and other required steps available in one place. By the end, 194 families had licenses, more than the 174 families previously licensed statewide. No new code was needed. The book does not argue against technology; it argues that technology should support a well-understood service, not distract from its underlying obstacles.
This example reflects a principle of process improvement: remove unnecessary steps before automating them. Automation can make a good process more consistent, but it can also make a bad process more efficient at frustrating people. The staple is memorable because it redirects attention from novelty toward the actual point of failure. At the same time, the licensing event should not be mistaken for a universal model. Foster-care decisions require careful judgment and time, so the goal is a responsive process, not speed without safeguards.
Design policy around the path benefits must travel to reach people.
Rockford made the problem concrete. In its effort to end veteran homelessness, the city and its partners built a shared list of the people they were trying to house. That changed the focus from counting beds or services to resolving each person's barriers. Interviews revealed that bus fare could prevent veterans from reaching medical appointments, renewing medication, or maintaining stability. After a pilot with 60 temporary bus passes, the city and county made discounted transportation available to veterans.
A policy can fail in delivery. The U.S. CARES Act provided emergency relief, but many people struggled to receive it because they lacked bank accounts, had not filed tax returns, or waited for mailed checks. By contrast, Berlin's relief effort used existing payment routes and a simpler application; more than $1.4 billion reached freelancers within five days. The lesson is to consider the last mile before a law is passed.
Policy design is often evaluated through eligibility rules and budget estimates, but a benefit has no practical value until someone can claim it. This is where implementation knowledge becomes political knowledge: the route a program takes determines who is included and who is missed. Rockford also shows how a service problem can have an unexpected cause, so familiar interventions such as more shelter beds may not address the bottleneck. Faster delivery is not automatically fairer, however. Choices about who qualifies and what evidence is required remain matters of public values.
Treat every dataset and algorithm as a decision with consequences.
Missing data can hide people. New York City analyst Amen Ra Mashariki noticed that official 311 reports showed few rats in a neighborhood he knew well. A friend living in public housing explained that many residents did not know what 311 was. The data therefore recorded who reported a problem, not simply where the problem existed. Without context, leaders could mistake low reporting for low need.
Automation can magnify harm. Michigan's unemployment fraud system wrongly denied benefits to more than 50,000 people between 2013 and 2015 and was found to be wrong 93 percent of the time. A data discrepancy was treated as proof against the claimant, and people could be penalized if they missed a notice. The deeper warning is that a faulty tool can compound inequity already built into a policy.
Data is not a neutral window onto the world. It is produced by rules, access, incentives, and unequal willingness or ability to report. This connects to measurement theory: a metric can be precise about the people it observes while missing those outside its field of view. Algorithmic review should therefore examine not only accuracy, but also the consequences of errors and the assumptions encoded in the underlying policy. Diverse teams can catch blind spots, but responsibility cannot rest on representation alone. Institutions need procedures for challenging data and correcting decisions.
Build reform into teams, budgets, and routines, not one champion.
Change needs support at every level. The projects in the book depended on leaders who could clear obstacles and frontline workers who understood how services operated. Michigan's redesign benefited from cooperation between state officials, caseworkers, and applicants. Vermont's small pilot succeeded with committed staff, but a wider project lost momentum after key champions departed and new leaders favored a different approach.
A field needs capacity, not slogans. The authors describe public interest technology teams in governments and nonprofits, training for existing staff, university programs, fellowships, and networks that connect practitioners. They also argue that effective work is often slow: teams must earn trust, navigate rules, test changes, and keep learning after launch. Sharing successful stories helps other institutions see what is possible, but sustained hiring, skills, and incentives are needed to make the practice routine.
The dependence on individual champions is both a source of energy and a structural vulnerability. Organizational change lasts when new practices enter budgets, job descriptions, review processes, and everyday decision-making, rather than remaining attached to a charismatic sponsor. Training current staff matters because external experts cannot own every service indefinitely. Still, institutionalization can dull experimentation if it becomes another compliance ritual. The challenge is to preserve room for learning while making user research, honest measurement, and accountability ordinary parts of public work.
Analysis
The book's central argument is that public problem solving fails when policy, implementation, and lived experience are treated as separate domains. Its examples make that separation tangible. ELIS shows a technology project detached from the work it was supposed to improve. Michigan's application redesign shows that a form can function as a gate to public support. Rockford's homelessness effort shows that a policy response becomes more intelligent when agencies share responsibility for specific people rather than optimizing isolated services.
The design, data, and delivery approach is strongest as a method for reducing uncertainty. User research reveals friction that policy documents conceal. Data helps show whether a problem is widespread and whether an intervention changes outcomes. Pilots provide a safer way to learn before expanding a service. In combination, these practices resemble implementation science, which studies how interventions behave in real settings rather than assuming that a promising idea works everywhere. The book also rightly insists that government remains essential: only public institutions can guarantee broad access and maintain responsibilities that cannot be replaced by a market offering.
The same method carries a serious tension. Better information and faster systems increase public capacity, but they can also make surveillance, exclusion, or punitive policy more efficient. The unemployment algorithm example demonstrates that accuracy is not enough if the underlying rules are unfair. Nor does a successful pilot settle questions of democratic authority, privacy, or distribution. These require explicit public accountability, not merely better product design.
The most durable contribution is therefore not a recipe for digitizing government. It is a demand to treat service delivery as part of policy itself. A benefit that eligible people cannot reach is not fully delivered; a system that works only while a champion is present is not yet institutionalized. The measure of modernization is whether public institutions become more able to listen, learn, and correct harm while honoring the rights of the people they serve.
Review Summary
Power to the Public receives mixed reviews, with readers praising its focus on user-centered design and public interest technology. Many appreciate the case studies and real-world examples illustrating how technology can improve government services. Critics note that some concepts feel elementary and the book could have gone deeper. Overall, readers find it a thought-provoking exploration of applying design, data, and delivery principles to solve public sector challenges, though some question its broad applicability and depth of analysis.
People Also Read
Glossary
public interest technology
Technology practice for public goodThe application of design, data, and delivery to advance the public interest in the digital age. In the book, it names a way of solving public problems that starts with people's needs, uses evidence to guide decisions, and tests and improves services. It may involve software, but it is not defined by a particular technology.
Design, Data, and Delivery
Three connected problem-solving toolsThe book's combined approach to public problem solving. Design means understanding the people affected and shaping services around their needs. Data means measuring conditions and results with useful, timely information. Delivery means testing changes in practice, learning from results, and improving before wider adoption. The authors argue that the elements work together rather than as independent options.
agile
Incremental product developmentA way of developing technology in small, testable portions rather than building a large system for years and releasing it all at once. Teams can gather feedback, release useful pieces, and revise their work as they learn. The book contrasts agile development with long, fixed projects, while noting that organizations need the skills and evaluation practices to use it meaningfully.
waterfall
Sequential, large-scale developmentA project approach in which teams plan and build a large system over an extended period before releasing it as a whole. In the book's examples, requirements and contracts can become fixed long before technology or user needs are understood. A major risk is discovering fundamental problems only after substantial time and money have already been spent.
minimum viable product (MVP)
Small test of a serviceA limited version of a product or service built to test whether an idea works in real conditions. The Vermont benefits team used a document uploader for one program and one office before considering broader expansion. An MVP is not simply a rushed launch; it should have a defined purpose, a small enough scope to manage, and evidence that can guide the next decision.
Built for Zero
Community homelessness initiativeA network and method for helping communities reduce homelessness through shared responsibility, reliable information, and coordinated action. Teams create a current account of the people experiencing homelessness, then work across agencies and organizations to resolve individual barriers and track progress. Rockford used this approach to end veteran and chronic homelessness and to work toward eliminating homelessness more broadly.
by-name list
Named record of people needing helpA regularly updated list identifying the people a community is trying to house, rather than relying only on aggregate counts such as shelter beds or meals served. In the Built for Zero approach, the list helps partners coordinate action around each person's situation, monitor whether people remain homeless, and understand changes in the population.
Model for Improvement
Questions guiding tested changeA continuous-improvement approach associated in the book with W. Edwards Deming's work. It asks what a team is trying to accomplish, how it will know a change is an improvement, and what change might produce that improvement. Teams use the answers to test interventions, study results, and revise their approach rather than relying on assumptions or delayed evaluation.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.