The half-life of skills.

Every skill list an organization keeps — the skills matrix, the certification register, the LMS transcript — obeys one silent rule: it only ever adds. A skill shown in 2022 is a row that never expires. In 2026 the row is still read as if no time had passed. Meanwhile, two separate forces have been working on its truth. The person may have lost the skill. And the world may have moved past it.

The add-only list is not a filing error. It reflects a real belief: once a skill is shown, it stays. Every staffing decision made from the list inherits that belief — who handles the incident, who runs the audit, which certified team the client is promised. The two research fields this article reviews exist to price that belief.

Both forces are measured. The first belongs to psychology: the skill-decay literature, a branch of the memory research this library covers. It has its own meta-analysis — a study that pools many studies — and its own applied emergencies. The second belongs to labor economics, which tracks the fading value of knowledge in the one market that flatters nobody: wages. Read together, the two fields deliver what the keynote version of “the half-life of skills” never does. They give numbers with methods attached, and a strategy that follows from the causes.

Clock one: decay through disuse

The field’s backbone is one big synthesis. It pooled the skill-retention experiments, which all share a shape: train a skill to a set standard, leave it unused for a while, then test again. Decay is large, and it grows with the gap — modest after days, severe after months. As nonuse stretches toward a year, the loss reaches about a full standard deviation, which is a very large drop. The moderators — the factors that speed decay or slow it — are the useful part. Cognitive and procedural skills decay faster than physical ones, accuracy fades faster than speed, and many-step tasks shed their steps in just the order anyone who has forgotten a checklist would guess (Arthur, Bennett, Stanush & McNelly, 1998).

Look at where that pattern points. The skills that survive best without upkeep are the physical, always-in-use ones — driving, manual technique — which workforce training rarely owns. The skills that vanish fastest are the mental procedures and decision rules that compliance, safety, and technical courses are made of. The decay research aims its bad news, with unlucky precision, at exactly the content corporate training spends most on.

The applied studies supply the emergencies. Take resuscitation skills: trained, certified, then unused unless disaster arrives. They decay measurably within three to six months of training — the systematic review of retention studies found losses in both knowledge and hands-on skill well inside standard recertification windows (Yang et al., 2012). The pattern extends to every rare-critical skill our simulation review covers. The skills organizations most need to be reliable are, by structure, the ones disuse erodes fastest — because their real-world practice rate is near zero by design.

The same literature offers two protections, and both point at design. Overlearning — practice continued past first mastery — buys retention, though the pooled benefit fades over months rather than lasting forever (Driskell, Willis & Copper, 1992). Retrieval-based maintenance — the successive-relearning protocol of our memory cluster — turns decay from fate into a scheduling problem. Revisit a skill before its predicted failure point and it stays serviceable, at a fraction of retraining cost. Decay is not an argument against training. It is an argument against training budgets with no maintenance line — which describes nearly every training program now running.

Clock two: obsolescence

A skill kept perfectly fresh can still lose its value in an afternoon. A vendor retires an API, or a regulator rewrites a threshold. No memory process is involved, so no retrieval schedule helps. This second clock runs outside the skull, and measuring it takes different tools. Economics owns the best ones.

Economists have long treated human capital as an asset that loses value — knowledge fading as tools, methods, and markets move. They measure the loss where it cannot be argued away: in pay. The classic route tracks the wage profiles of scientists and engineers. In fast-moving fields, the returns to experience flatten and erode as earlier vintages of knowledge lose market relevance (Neuman & Weiss, 1995). Parallel work on academic careers found the same vintage effect in publication records, with faster-aging fields peaking earlier (McDowell, 1982).

The modern flagship study made the mechanism precise. It tracked applied-STEM careers and showed that the skill demands of technical jobs change fast — fast enough that the education premium is largest early, then erodes as graduates’ toolkits age. The erosion concentrates in the applied, tool-linked skills and is much weaker for foundational knowledge (Deming & Noray, 2020).

That split is the finding strategy should care about. Obsolescence attacks the surface layer — the framework version, the interface, the current procedure. The foundations underneath — the statistics beneath the tool, the principles beneath the regulation — hold value across vintages. So the answer to clock two is a portfolio: weight the curriculum toward foundations, and keep a thin applied layer that is refreshed often.

Where the two clocks compound

The clocks run on their own, but company habits couple them in the costliest way. Take the standard life of a technical certification program: a burst of training when a system launches, a certificate, then silence. Decay starts at once on whatever is not used weekly. Three years later the vendor ships a major version. Now the organization faces both bills together — refresh the skills that faded and teach the delta, the part that changed — usually as one expensive retraining event the next cycle will repeat.

The compound bill is a choice, not a law. Maintained skills absorb updates cheaply, because the delta lands on a live foundation. Decayed skills force every update to re-teach part of the base first.

The interaction runs the other way too: fear of obsolescence pushes organizations toward novelty-only training portfolios — always the new tool, never the upkeep. That choice maximizes decay losses on everything already taught. Read together, the two literatures give advice that is almost boringly conservative. Keep a stable foundation layer, maintained toward permastore — the near-permanent memory state covered below. Keep the applied layer thin and refresh it as deltas, and put rare-critical skills on scheduled rehearsal. Nothing in either evidence base rewards the churn that “skills are dying faster than ever” talk produces.

What the decay moderators mean for design

The meta-analysis’s moderators map one-for-one onto scheduling policy. The cognitive-versus-physical split says knowledge-heavy compliance and procedure content needs shorter maintenance cycles than motor skills — the opposite of most recertification calendars, which renew forklift operation yearly and product knowledge never. The accuracy-versus-speed split says decayed skills fail quietly: the practitioner still moves fluently while the error rate climbs. That is why self-report (“I’m still comfortable with it”) is a worthless freshness measure, and brief objective checks are not. And the steps finding — multi-step procedures shed steps — says checks should probe the sequence, not the concept. Ask for step four cold, because step four is what leaves first (Arthur et al., 1998).

Retention windows also depend on how the skill was trained — the link back to our whole memory cluster. Material learned through spaced retrieval enters the gap with a flatter curve than material crammed for the certificate. So two organizations with identical certificates can hold very different real capability three months later. A maintenance schedule tuned only to elapsed time will over-serve the well-trained and under-serve the crammed. That is one more reason acquisition and maintenance belong in one system — a system that knows each item’s history.

The permastore counterweight

If everything decays and everything goes out of date, workforce development would be a treadmill with the belt speeding up. That is roughly how the keynote version tells it. The retention literature’s deepest finding says otherwise, and it deserves more airtime than the panic gets. Bahrick’s fifty-year studies of Spanish learned in school found that whatever survived the first few years of forgetting then held steady for decades — a “permastore” of well-learned, often-revisited material nearly immune to further loss (Bahrick, 1984). The medical-education version repeats the shape: basic-science knowledge decays steeply at first, then the survivors persist through whole careers (Custers, 2010).

The implication turns panic into design. Skills are not uniformly perishable. Material learned to real depth, spaced and retrieved across years, graduates into near-permanence. So the maintenance burden is front-loaded and finite, not eternal. Organizations that spend on consolidation early are buying decades of stability later.

Reading obsolescence at the role level

The economics turns into workforce planning through a sorting step every role supports. List a role’s required skills, then sort them by exposure to the second clock. A claims adjuster’s list splits cleanly — negotiation and fraud-pattern judgment (foundational, slow to age), the current claims platform’s workflows (applied, fast), this quarter’s regulatory thresholds (applied, faster). The exercise routinely shows that a role’s “future-proofing” problem is smaller and more specific than the anxiety suggested. Perhaps a fifth of the list sits in the fast lane — and that fifth is exactly where steady micro-updates beat periodic retraining events.

The same sort prices the make-or-buy decision honestly. When a truly new capability lands in a role’s fast lane, ask one question. Do the current staff’s foundations support the delta, or does the new demand reach the foundation layer itself? In the first case, internal updating is cheap and fast; in the second, our reskilling review holds the relevant evidence. Organizations that skip the sort tend to buy outside help for what a delta would have fixed, and to under-scope what needed real retraining — the sorting step prevents both errors.

About those “five-year half-life” keynotes

So far this article has used real literatures with named methods. It owes the phrase in its own title the same scrutiny. “The half-life of skills” has drifted from metaphor to statistic without ever gaining a method. The figures in circulation — skills once lasted decades, now five years, soon two — come from consultancy estimates and executive surveys: informed opinion about how fast job demands change, pooled and rounded (World Economic Forum, 2023). They are not measurements of either clock.

No cohort was tested and retested. No depreciation was estimated from wages. “Skill” is not even defined across the claims — the tool version, or the foundation? And the radioactive-decay framing implies a smooth, universal process that both real literatures contradict: decay can be scheduled, and obsolescence moves in steps and layers. The honest use of such figures is directional urgency — change is real, and faster in some layers than others. The dishonest use is planning arithmetic, and the tell is any strategy document that computes headcounts, budgets, or retirement dates from a number nobody ever measured.

What the evidence doesn’t show

• It doesn’t yield one number. Decay rates vary by skill type, training depth, and interval; obsolescence rates vary by field and layer. Any single “half-life of skills” figure is a category error over both literatures.
• It doesn’t mean certifications are useless. It means they are timestamps: evidence about the date of demonstration, with a decay function attached that recertification windows should be derived from — not from tradition (Yang et al., 2012).
• It doesn’t justify perpetual-novelty curricula. The Deming–Noray split argues the opposite: chasing every tool vintage while under-investing in foundations maximizes exposure to both clocks (Deming & Noray, 2020).
• Permastore is earned, not automatic. Bahrick’s stable knowledge was acquired over years of spaced use; cram-certified material never reaches the stable regime and decays on the ordinary curve (Bahrick, 1984).

Where the evidence stops

1. 1It doesn’t yield one number
2. 2It doesn’t mean certifications are useless
3. 3It doesn’t justify perpetual-novelty curricula
4. 4Permastore is earned, not automatic

© 2026 FUTURE PROOF™

What this means for practice

The reforms come in three shapes — inventory, schedule, and curriculum — and all three run on data most organizations already hold. Put expiry logic into the skills inventory. Every capability record gets a decay class (physical or cognitive, used daily or rare-critical) and a matching confidence window. When the window closes, the row demands fresh evidence — a brief retrieval check, not a retraining course. Schedule maintenance where the meta-analysis says the cliffs are: rare-critical skills on short cycles, daily-used skills almost never (their use is their maintenance), and the rest in between. Fund the maintenance line openly, with the successive-relearning economics as the business case: return visits are cheap, and they get cheaper as material settles toward permastore.

Above all, set recertification windows from evidence, not tradition. The intervals in most compliance calendars — annual, biennial — were chosen for administrative rhythm. The decay studies place the danger zones months earlier, for exactly the cognitive material those programs certify. And run the obsolescence clock as a separate process, because no amount of retrieval practice preserves the value of a retired procedure.

Version the role maps. When tools, regulations, or methods change, the delta — not a whole new curriculum — flows to the affected roles. The analytics then separate “forgot it” from “never learned the new version,” which need different fixes. Weight new curricula toward the foundational layer that survives vintages, and keep the applied layer thin and constantly refreshed. Organizations that do both are not future-proofing by slogan. They are running two maintenance schedules against two measured clocks — which is all the famous phrase ever validly meant.