Mike Goldstein, the founder of Match Education and 1Up Career Coaching, is one of my favorite education thinkers. Over the years, he鈥檚 often been decades ahead of the pack when it comes to rethinking tutoring, teacher pay, professional development, school models, and more. He鈥檚 occasionally shared glimpses of his thinking here at RHSU. Well, he wrote recently, after his nonprofit 1Up Career Coaching was featured in best-selling author Dan Heath鈥檚 new book RESET: How to Change What鈥檚 Not Working. I was intrigued by Goldstein鈥檚 take, per usual. Here鈥檚 what he had to say:
鈥搁颈肠办
Dear Rick,
Last fall, after the election, you wrote that there鈥檚 an opportunity for the education community to engage in some much-needed reflection and to rethink some of the areas where we鈥檙e stuck. When I read that, I was reminded of Dan Heath鈥檚 new book RESET: How to Change What鈥檚 Not Working. You may know Dan, the best-selling author of Made to Stick and Switch.
In RESET, he digs into broken systems鈥攈ospitals where packages take three days to show up, animal shelters that can鈥檛 get people to adopt cats, fast food restaurants where drive-thrus take forever鈥攁nd how to fix them.
One example of a failed system Dan provides is K-16 education: He explores the problem where many first-generation college graduates struggle to find good entry-level jobs. They apply to dozens of openings and get no replies, let alone interviews. Dan found his way to my friend Geordie and me. We were curious about what was causing this and where the conventional wisdom might be wrong. When Geordie and I interviewed these 鈥渦ndermatched鈥� college grads, they expressed exasperation and sometimes shame. 鈥淲hat鈥檚 wrong with me?鈥� they鈥檇 ask.
In Dan鈥檚 parlance, our next step was to 鈥渕ap the system.鈥� Yes, their high schools never delved deeply into career exploration nor realistic entry-level job expectations; yes, they鈥檇 sort of haphazardly chosen majors; yes, they hadn鈥檛 really built the social capital in the form of 鈥渃onnections鈥� that would genuinely help them chase down jobs after graduation.
But the key leverage point was in the last part of their K-16 experience. Despite job-search support offered by university career centers, recent grads continued to struggle鈥攁 reality that makes sense, given that career-center programs have a bad track record. In Dan鈥檚 鈥渕ap the system鈥� parlance, we thought, that鈥檚 where the gold was buried鈥攖he opportunity for quick, productive change.
So, Rick, we took a page from your book Cage-Busting Leadership. Instead of creating one more career-coaching program, we decided to invert things. Usually, education interventions are designed to have a fixed dosage of some input and then see what happens. Even when I conceived of 鈥渉igh-dosage math tutoring鈥� at , all that meant was the dosage of inputs was bigger and of better quality! But we still weren鈥檛 steering toward a concrete outcome.
To help struggling post-grads, we decided to do things differently: We held the result constant and varied the dosage. Geordie and I first tried this approach with a small sample. We decided we would do whatever it took, for as long as it took, to get a concrete result: Find them a new job that would pay 20% more and raise alumni satisfaction from 4 out of 10鈥攖he average baseline job satisfaction among our 1Up clients in their current 鈥渦ndermatched鈥� job鈥攖o 8 out of 10.
Instead of giving generic advice for each step of the job search, we鈥檇 sit there patiently alongside students and recent graduates and just do it: search the internet to seek out jobs they might like, read job descriptions together, bypass the college鈥檚 inefficient Handshake click-to-apply tool鈥攁nd instead find actual humans who they could email directly鈥攁nd draft and send applications on the spot. When interviews came along, we鈥檇 drop everything to help them prepare.
We whipped up a little nonprofit, , to provide this service. We soon realized we鈥檇 need to reject for our clients to succeed. For one thing, with permission, we were honest to the point of bluntness鈥攖here was no other way to do the work. For another, there was no generic 鈥減rep鈥� that helped. If someone had an interview coming up, we worked with them 24 hours beforehand for that particular interview, for over an hour, creating key lines and rehearsing them over and over until they flowed. For in-person interviews, we plotted the Uber drop-off point and target time. For an interview on Zoom, we adjusted lighting, camera angle, background, and sound level.
It worked! Our first 30 people found new jobs. Then another 50.
Now, we鈥檝e shifted our focus from charter school alumni to a different cohort of low-social-capital college grads: middle-aged moms who attend one of the largest online colleges in the U.S. The program is working for them, too.
Rick, you鈥檝e , and I agree, that policy can be a bad tool because it can require things to be done, but it can鈥檛 require that things be done well. Your insight explains not just why many school-based interventions don鈥檛 produce meaningful outcomes but also why the few that do in small trials don鈥檛 scale well. The people running these interventions never know when the 鈥渏ob is done.鈥�
Most interventions just change an arbitrary 鈥渟tatistically significant improvement鈥濃�攁 kid at the 50th percentile makes it to the 53rd, for example. That allows the program and the evaluator to claim success. But it鈥檚 not an easy-to-grasp target. By contrast, if you are building a table, tutoring a kid until he can pass a certain test, or counseling a 22-year-old until they have actually landed a new job鈥攖hat鈥檚 a more concrete targeted outcome.
What if interventions first achieved a concrete, clear gain like 鈥渨hatever it takes to read reasonably well鈥� or 鈥渨hatever it takes to stop being clinically depressed,鈥� no matter the cost in time or resources? Then, only after we know what it takes to achieve a goal鈥攂oth the dosage of an intervention and the human skills needed to deliver that well鈥攄o we examine the resource constraints.
I know what you鈥檙e thinking: How can schools afford this?
First, sometimes the money is already there. Take Boston, where schools for each student. Over 13 years from K-12, that鈥檚 $390,000 per child. What if parents controlled that money? They could spend $20,000 per year on basic schooling and save $10,000 per year for bursts of high-dosage help when needed. If their 3rd grader struggles to read, or their 8th grader was spiraling into depression, or their 10th grader seemingly had a shot to be really good at tennis, they鈥檇 do what wealthy families do: Hire intensive help until the child is curling up with Harry Potter, stabilized, or hitting 105 mph serves. Wealthy families don鈥檛 pay for help that only moves their child up .08 of a standard deviation and results in their child still being a struggling reader. They buy the dosage and quality needed to get the job done.
Second, artificial intelligence can lower costs in the long term, but it must replicate something successful. Right now, we鈥檙e using AI to replace weak interventions, like replacing low-quality human tutoring and counseling with even weaker ones. That鈥檚 like trying to fix a wobbly table by removing another leg. Instead, we should use AI to enhance proven systems. Once we鈥檝e anchored to results, we can explore how AI might improve efficiency or scalability.
Dan Heath鈥檚 RESET is about programs that approach challenges from a new direction. In the education sector, the endless stream of half measures and weak interventions with bad incentives is not leading to the desired outcomes. We need a new direction.
It鈥檚 time to anchor R&D to clear, meaningful outcomes and do whatever it takes to achieve them. That means taking small steps鈥攁nd often failing and recalibrating鈥攖o achieve durable wins. Whether it鈥檚 a college grad landing a great job or an 8th grader mastering fractions, the principle is the same: We shouldn鈥檛 stop halfway or arbitrarily dole out 鈥渉elp鈥� that doesn鈥檛 produce a concrete result. We keep going until the job is done; then, we come up for air, look around, and see what it takes. Only then is it time to have the hard conversations about resource constraints.
