In 2018 David Deming and Lisa Kahn published in the Journal of Labor Economics a study based on millions of US job postings for professional positions. They coded the skills demanded into ten categories and compared them across firms, holding the job title constant.
The result that matters to anyone thinking about moving into digital work: the variation in required skills within the same title, across different firms, is very wide. And that variation correlates with the salary offered: postings asking for cognitive and social skills together pay systematically more, within the same occupation and the same local market.
Put directly: “becoming a data analyst” isn’t a description of what you’ll do. Two positions with that same name, in the same city, can involve different work and pay different money.
”Digital” isn’t a sector
There’s a confusion at the outset that makes most guides unhelpful. Digital professions aren’t an economic sector: they’re a set of tasks that have spread into every sector.
People doing data analysis work in banks, hospitals, logistics, public administration. People doing online content and communication work in manufacturing as much as in agencies. The share of people with digital duties who work at technology companies is a minority of the total.
That has a practical consequence almost nobody mentions: the shortest route into a digital profession often isn’t a change of sector, but a change of role inside the sector you already know. Knowing a domain (how a pharmacy, a construction site or a credit office actually works) is part of the package that someone arriving from a purely technical path doesn’t have. The cost of switching sector, incidentally, is documented and high: I wrote about it in reskilling.
What the postings actually ask for
James Bessen studied the relationship between computerisation and employment at the level of the individual occupation: occupations that adopt computers most tend to grow in headcount, not shrink, while their content changes. It’s the same mechanism as the ATM case described in jobs of the future.
On content, Brad Hershbein and Lisa Kahn showed something important by studying postings before and after the 2008 crisis: in the hardest-hit areas firms raised their stated requirements (more education, more technical skills) and those requirements did not come back down when demand recovered. The recession permanently accelerated the shift toward non-routine tasks.
Putting the two together: more positions, with higher and more explicit requirements, spread everywhere.
| Role family | Core task | What it actually takes |
|---|---|---|
| Data analysis and reporting | Turning messy data into decisions | Basic statistics, SQL, ability to explain |
| Software and web development | Building and maintaining systems | Programming logic, reading other people’s code |
| Product and project | Deciding what gets done and in what order | Prioritisation, writing, handling conflict |
| Online content and communication | Being found and being understood | Writing, measurement, domain knowledge |
| Interface design | Reducing friction for users | User research, iteration, argumentation |
| Systems, cloud and security | Keeping things up and protected | Networks, automation, incident handling |
| Support and customer success | Translating between users and product | Diagnosis, patience, ability to summarise |
The right-hand column is the part job lists leave out, and it’s the only one that lets you work out whether an occupation suits you.
The uncomfortable bit: certificates
There’s one experiment worth more than a lot of debate. David Deming and colleagues published in the American Economic Review an audit study: they sent thousands of fictitious applications to real job openings, randomly varying the type of credential held.
The result for online credentials from for-profit institutions: they produce no callback advantage over applications with public-institution credentials or with none, and in some specific roles the callback rate comes out even lower.
That doesn’t mean studying is useless. It means the certificate itself, as a signal to an employer, is worth much less than it costs, and that the effective signal is something else: something you’ve built that can be looked at. That’s why it’s worth having an online portfolio before you have a certificate.
The job title tells you which room you’ll walk into. It doesn’t tell you what you’ll do in that room, or what you’ll be paid for doing it.
How to find your way among the roles
- Read thirty postings before choosing. Not to apply: to build a list of the tasks that keep recurring. That’s the real occupation, net of the name.
- Start from the sector you already know. The combination of domain knowledge and digital skill is rarer, and better paid, than digital skill alone.
- Choose by task, not by title. Do you want to spend your days writing, diagnosing faults, arguing about priorities, or cleaning data? These are different jobs with similar salaries.
- Build one real thing instead of collecting courses. A small but finished project, on your own data or your own problem, says more than three certificates.
- Learn programming logic even if you don’t want to code. It’s needed in almost every role in the table: start from how to learn to code.
- Don’t neglect the social half. The wage premium documented by Deming and Kahn is on the combination of cognitive and social skills, not on either one. The reason is in hard vs soft skills.
- Test early whether the work bores you. A month of real practice costs little and eliminates most bad choices. For data work, a starting point is where to begin.
My take: “digital careers” describes a tool, not a job
I think the category “digital professions” has outlived its usefulness, just as “professions that use the telephone” would have in the 1980s. It describes a tool, not a job, and that creates wrong expectations: people sign up for a course on a tool and discover afterwards that the work consisted mostly of persuading someone.
I also think the Deming et al. experiment on online credentials is the most ignored finding in the training industry, and I suspect it’s ignored precisely because the business model of many courses rests on the opposite promise.
I’ll stick my neck out, knowing I can’t prove it: over the next few years the advantage will shift further toward people who sit between the technical system and the people using it. Not because it’s a noble role, but because that’s where the necessary information isn’t written down anywhere.
Frequently asked questions
Do I need to code to work in digital? No for most roles, yes for understanding what you’re asking of others. The logic is nearly always needed, the specific language almost never.
How long does it take to get in? It depends heavily on your starting point and on whether you stay in your sector. Staying in the sector and changing role is almost always the shorter route.
Do bootcamps work? As an accelerator of practice they can; as a signal to employers they count for less than they promise. The experimental evidence on online credentials is blunt about this.
Which digital profession pays best? Badly posed question: the variance within the same profession is as wide as the variance between professions, and it depends on the mix of tasks.
Am I too old to switch? What weighs in the data isn’t age itself but the length of time spent out of work in the destination sector. Moving from inside a job is far more effective than moving from a standstill.
Sources
- Deming, D. J., & Kahn, L. B. (2018). Skill Requirements Across Firms and Labor Markets: Evidence from Job Postings for Professionals. Journal of Labor Economics, 36(S1), S337-S369. The ten skill categories and the variance within the same title.
- Deming, D. J., Yuchtman, N., Abulafi, A., Goldin, C., & Katz, L. F. (2016). The Value of Postsecondary Credentials in the Labor Market: An Experimental Study. American Economic Review, 106(3), 778-806. The audit study on online credentials.
- Hershbein, B., & Kahn, L. B. (2018). Do Recessions Accelerate Routine-Biased Technological Change? Evidence from Vacancy Postings. American Economic Review, 108(7), 1737-1772. The permanent upward shift in requirements.
- Bessen, J. E. (2016). How Computer Automation Affects Occupations: Technology, Jobs, and Skills. Boston University School of Law, Law and Economics Research Paper No. 15-49.
- Autor, D. H. (2015). Why Are There Still So Many Jobs? The History and Future of Workplace Automation. Journal of Economic Perspectives, 29(3), 3-30.
- Goos, M., Manning, A., & Salomons, A. (2014). Explaining Job Polarization: Routine-Biased Technological Change and Offshoring. American Economic Review, 104(8), 2509-2526.
- Acemoglu, D., Autor, D., Hazell, J., & Restrepo, P. (2022). Artificial Intelligence and Jobs: Evidence from Online Vacancies. Journal of Labor Economics, 40(S1), S293-S340. Using posting text as a measure of skill demand.
- Kroft, K., Lange, F., & Notowidigdo, M. J. (2013). Duration Dependence and Labor Market Conditions: Evidence from a Field Experiment. Quarterly Journal of Economics, 128(3), 1123-1167. Callback rates falling with unemployment duration.
- Deming, D. J. (2017). The Growing Importance of Social Skills in the Labor Market. Quarterly Journal of Economics, 132(4), 1593-1640.
- Bloom, N., Liang, J., Roberts, J., & Ying, Z. J. (2015). Does Working from Home Work? Evidence from a Chinese Experiment. Quarterly Journal of Economics, 130(1), 165-218. The controlled experiment on remote work.
This article was written following the process described in how we write. If you find a figure reported wrongly, write to me: I’ll correct it and note the correction at the bottom of the article.



