The Wrong Question: Which Tech-Transition Program Fits?
Susan Athey and Emil Palikot’s study of Dare IT asks a deceptively practical question: how can women who already have technical ability move into technology jobs? Its answer is not that one program is simply better than another. The study compares two interventions in Poland—Mentoring and Challenges—that address different obstacles between having useful skills and being hired. The important judgment is therefore not “Which program wins?” but “Which constraint is stopping this person’s transition, and which program is designed to relieve it?”
That distinction matters well beyond this particular study. Technology work can offer strong demand and attractive wages, yet the route into it is not determined by technical ability alone. A candidate may know how to do the work but lack access to people who can explain where opportunities are, recommend them, or help them navigate hiring. Another candidate may reach the application stage but struggle to convince an employer that a degree, a résumé, or a self-description represents practical competence. Those are different failures of the labor-market match. Treating both as a generic “skills gap” leads to programs that teach more without fixing what is actually blocking the transition.
A choice between two different failures
Dare IT’s comparison focuses on women in Poland who had baseline technical skills but had not yet secured technology employment. That population makes the central problem unusually clear. The participants were not starting from no preparation; they were trying to convert preparation into a job in a sector where women remain underrepresented and where employers may have incomplete or biased information about technical ability.
One obstacle is a network and search problem. Hiring is not conducted only through public job postings. Information about vacancies, the expectations of hiring managers, and the unwritten rules of a sector often travels through professional relationships. Someone outside a major technology center may have fewer local contacts. Someone returning after a career interruption may have lost access to former colleagues. Someone entering a male-dominated field may find it harder to enter the informal circles through which opportunities are discussed. Even when a candidate is capable, a thin network can reduce the number of credible opportunities they hear about and the number of people willing to vouch for them.
The other obstacle is an information and credibility problem. Employers cannot directly observe a candidate’s ability before hiring. They use degrees, previous job titles, references, tests, and work samples as signals. These signals are imperfect. A woman with limited work history may know the relevant tools but have little conventional evidence that she can apply them to a real project. A formal credential may show that she studied, but not whether she can produce a usable interface, work through feedback, or complete a technical task under realistic constraints. Stereotypes can make that uncertainty more costly: when an employer already doubts a candidate’s technical ability, a vague claim of competence may carry less weight than concrete evidence.
The two problems can coexist, but they point in different directions. A new connection can open a door that a portfolio never reaches. A strong portfolio can change an employer’s assessment even when the candidate has no influential contact to make an introduction. The first is primarily about access to opportunity and the process of search. The second is primarily about making ability legible once an opportunity is found.
That is why the programs should not be treated as two versions of the same training product. Mentoring pairs participants with experienced women in technology for weekly one-to-one sessions over four months. The mentors emphasize expanding professional connections, making introductions, and teaching participants how to use those connections; the program also includes tacit recruitment knowledge, confidence reinforcement, and technical feedback. Challenges is an online program built around six bi-weekly, industry-designed assignments in user-experience design or React front-end development. The assignments culminate in a complete project that participants can show in applications and interviews, with expert review intended to make the work more credible than a generic online certificate.
A useful first approximation is therefore simple:
The diagram is a decision frame, not a sorting machine. It does not say that every person in a small town needs a mentor or that every early-career applicant needs a portfolio. It says that a program has a plausible route to impact only when its mechanism meets the obstacle. The rest of the judgment depends on evidence about the person, the local labor market, the employer, and the program’s capacity.
What the study changes about “effective”
The randomized evaluation gives the comparison weight because it separates program access from the participants’ pre-existing prospects. Both interventions increased technology employment after twelve months. Mentoring raised technology employment by about 15 percentage points, from a control level near 46% to about 61%. Challenges raised it by about 11 percentage points, from a control level near 27% to about 38%.
Those numbers are important, but they do not establish a universal ranking. The two effects were estimated in different experimental populations and against different control levels. Mentoring’s randomization occurred within pairs of applicants whom mentors had selected as preferred candidates. Challenges randomly assigned eligible applicants to access or control. The estimates therefore answer slightly different questions about people who entered each program’s study design. They show that both forms of access can matter; they do not show that an applicant would receive exactly the same benefit from either option.
The study also measured an intermediate outcome that helps explain the contrast. Mentoring participants gained an average of about 52.7 LinkedIn connections relative to the control group, roughly a 23% increase. Challenges did not produce a detectable expansion of professional networks. That difference is consistent with Mentoring operating through introductions and network-building rather than merely through encouragement or additional technical practice. The timing of the employment gains fits the same story: Mentoring participants began finding technology jobs during the program, while Challenges gains were more concentrated after completion, when participants could deploy their finished projects in later applications.
Challenges’ value is therefore not that it is “online” or that it contains six assignments. A portable signal works only if someone on the hiring side can interpret and trust it. The assignments were designed with industry practitioners, modeled on realistic business tasks, reviewed by experts, and assembled into a project that could be discussed in an interview. A self-selected collection of tutorials or an unrecognized badge would not necessarily create the same information for an employer. The relevant comparison is not personal support versus digital content. It is a relationship that changes access versus evidence that changes what an employer can observe.
This also explains why the question cannot be answered from outcome averages alone. A program can be effective in aggregate while being poorly matched to a particular participant. Mentoring depends on a scarce supply of experienced volunteers and was substantially oversubscribed. Challenges has a reported participant cost of about $15 and is easier to deliver to large online cohorts, but its payoff depends on sustained participation, credible review, and employers recognizing the resulting work. A lower reported unit cost does not make its mechanism universally better; a more intensive program does not make its mechanism universally more valuable.
The decision is about fit under constraints
The broader lesson from the World Bank’s guidebook on employment solutions is that labor programs succeed when they are tailored to both the context and the worker. Local opportunity, readiness, matching, regulation, mobility, care responsibilities, and income security can all shape whether a program has a chance to work. A person may be ready for technology work but live far from the relevant network. Another may have a strong network but lack a credible way to demonstrate practical ability. A third may fit either program on paper but be unable to attend consistently without childcare, transport, or income support.
This is why the study’s most useful contribution is not a new universal recipe for women in technology. It is a way to think about fit. Before asking which intervention to fund, admit, or recommend, a decision-maker should ask what is missing from the current match between worker and opportunity. Is the problem that the candidate cannot reach the right people? Is it that employers cannot see what the candidate can do? Is the local market too thin for either intervention to matter? Is the person’s apparent lack of progress actually a participation barrier outside the program’s scope?
The answers can change the proper response. A participant with several years of experience but few technology contacts may be able to turn a mentor’s introductions into interviews quickly. An early-career participant with limited work history may gain more from a reviewed project that gives employers something concrete to evaluate. Someone facing both obstacles may need a portfolio first and a network pathway afterward, or the reverse. Someone facing weak local demand, a severe skills gap, or a care constraint may need a broader package rather than either Dare IT option alone.
The evidence also warns against confusing selection with impact. Mentors were good at identifying applicants who had stronger baseline prospects, but the person most likely to find a job without help is not necessarily the person who benefits most from help. A fair and effective allocation process must distinguish likely success from likely improvement. It must also decide how much weight to give efficiency, access for people with the weakest starting position, speed of employment, and long-term job quality. Those are policy choices, not outputs that a program name can settle.
The right question, then, is narrower and more demanding than “Mentoring or portfolio signals?” It is: what is the binding constraint in this transition, which mechanism addresses it, and what evidence would tell us whether the match is working? Once that question is asked, Mentoring and Challenges become easier to compare without pretending they are interchangeable. The next step is to examine the labor-market frictions that sit underneath a stalled transition, because a program can only repair a problem that has first been diagnosed accurately.