In 1973 Michael Spence published in the Quarterly Journal of Economics the paper that would win him a Nobel. The problem he addressed is exactly the jobseeker’s: the candidate knows their own worth, the employer doesn’t.
Spence’s formal solution is a precise condition. A signal carries information only if it costs more for those who are worth less. If producing the signal costs everyone the same, in equilibrium everyone produces it and the signal distinguishes nobody.
Applied to the CV, that explains everything. Writing “strong analytical skills” costs zero seconds for anyone, competent or not. By construction it cannot carry information. A finished, visible, attributable project costs a great deal for someone who can’t do it, and that’s exactly why it works.
Why recruiters need evidence
There’s a second, less intuitive piece, and it concerns the reader.
Mitchell Hoffman, Lisa Kahn and Danielle Li studied, using real firm data, what happens when hiring managers are allowed to override the result of a structured test. The finding: when they exercise that discretion and hire candidates the test ranked lower, the average quality of hires gets worse, measured by how long people actually stay.
That’s not an argument against people: it’s an argument that unstructured judgement is noisy. And noise gets reduced in only one way, by giving the decider something concrete to evaluate.
Which is where the key data comes in. In the major reassessment of predictive validities published in 2022 by Paul Sackett and colleagues in the Journal of Applied Psychology, many historical estimates get revised downward, including those for work samples, long considered the single best method. But the hierarchy survives: structured methods based on observable performance remain clearly above the unstructured interview and the reading of a CV.
A portfolio is, in essence, a work sample you administered to yourself.
Nobody will read it the way you hope
Now the part that ruins most portfolios, and it’s a problem of attention, not content.
Harald Weinreich and colleagues published in ACM Transactions on the Web a study of real browsing behaviour, collected from a panel of users over long periods and millions of page views. The central finding: page visit durations are heavily skewed and extremely short at the median. A large share of visits last a few seconds, and only a minority amount to actual reading.
That doesn’t make the portfolio useless. It means everything that has to land must sit in the first screen, and the order is: what you did, for whom, with what result. Before the biography, before the navigation menu, before the list of tools you know.
There’s also a documented time factor: field experiments with fictitious applications show callback rates falling as the spell out of work lengthens. Having something already published when the moment arrives is worth more than building it afterwards.
Even if your work isn’t visual
The most common objection is that portfolios only make sense for people who produce images. That’s false, but the word “project” needs translating.
| Role | Artefact that works as proof |
|---|---|
| Data analysis | An analysis of public data, with question, method and stated limits |
| Finance and control | A model or procedure reconstructed and explained step by step |
| Sales | A written case: situation, objection, what you did, numerical outcome |
| Customer support | A troubleshooting guide for the most frequent problems in your field |
| Project management | An honest post-mortem of a project, including what went wrong |
| Marketing and content | Three published pieces with their result data, not ten without |
| Technical roles | A small working tool that solves a real problem |
The criterion is always the same: something an incompetent person couldn’t produce in an afternoon. If they could, it isn’t a signal.
A portfolio isn’t there to show how good you are. It’s there to make looking good expensive for those who aren’t.
How to build it so it gets read
- Put the evidence in the first screen. Three projects, one line each with the result. Biography, photo and menu come afterwards.
- State the problem before the solution. The evaluator has to understand in ten seconds what was hard. Without that context the result is unreadable.
- Declare the limits. “This data only covers six months” increases credibility, it doesn’t reduce it: it’s a cost a bluffer won’t pay.
- Three finished projects beat ten sketched ones. The quality of the worst item is what gets remembered.
- Make everything attributable and checkable. Name, date, link, your exact role on the team. A result nobody can verify is worth as much as a line on a CV.
- Don’t use confidential material. Rebuild the example with invented but realistic data and say so openly: that’s the professional version, not the clever one.
- Build it before you need it. Callback rates worsen with time spent out of the market, and a portfolio can’t be improvised in a week. The rest of the visibility strategy is in personal branding.
My take: the graphics distinguish nobody
I think the advice to “polish your portfolio” is almost always given backwards: people talk about graphics, platform and colours, the part that distinguishes nobody, and not about the only thing that counts, namely that the content should be hard to produce.
I also think the most underrated move is stating what didn’t work. It’s probably the strongest signal there is, precisely because no weak candidate would ever do it: admitting a specific mistake requires having understood it.
This is my hypothesis, not a finding: as automated content production becomes free, the value of portfolios will shift rapidly from outputs to decisions. Why that dataset and not another, why that trade-off: it’s the only part that stays expensive to fake.
Frequently asked questions
Which platform should I use? The one that lets you publish without friction. The technical choice matters far less than it seems: what matters is content and how fast you update it.
How many projects do I need? Three done well. Beyond five the marginal return collapses, because nobody reaches the bottom of the page.
I have nothing to show. What do I do? Build a small project on a problem in your current sector. The value of a domain you already know is the point discussed in digital careers.
Does a portfolio replace the CV? No: the CV gets you past filters, the portfolio convinces the decision-maker. You need both and they do different jobs.
Is it worth it in non-technical roles? Yes, the form of the artefact changes. A well-written case works as well as a chart.
Sources
- Spence, M. (1973). Job Market Signaling. Quarterly Journal of Economics, 87(3), 355-374. The differential-cost condition for a signal to be informative.
- Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting Meta-Analytic Estimates of Validity in Personnel Selection. Journal of Applied Psychology, 107(11), 2040-2068. The reassessment of selection-method predictive validities.
- Hoffman, M., Kahn, L. B., & Li, D. (2018). Discretion in Hiring. Quarterly Journal of Economics, 133(2), 765-800. What happens when recruiters override a structured test.
- Weinreich, H., Obendorf, H., Herder, E., & Mayer, M. (2008). Not Quite the Average: An Empirical Study of Web Use. ACM Transactions on the Web, 2(1), 1-31. The real distribution of page visit durations.
- 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. Callbacks falling with unemployment duration.
- Eriksson, S., & Rooth, D.-O. (2014). Do Employers Use Unemployment as a Sorting Criterion When Hiring? Evidence from a Field Experiment. American Economic Review, 104(3), 1014-1039.
- Bertrand, M., & Mullainathan, S. (2004). Are Emily and Greg More Employable Than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination. American Economic Review, 94(4), 991-1013. The fictitious-application method and baseline callback rates.
- Highhouse, S. (2008). Stubborn Reliance on Intuition and Subjectivity in Employee Selection. Industrial and Organizational Psychology, 1(3), 333-342.
- 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. Why a certificate alone is a weak signal.
- Nisbett, R. E., & Wilson, T. D. (1977). The Halo Effect: Evidence for Unconscious Alteration of Judgments. Journal of Personality and Social Psychology, 35(4), 250-256. Why a first impression contaminates later judgements.
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.



