South Africa's global business services sector added 26,346 international-facing jobs in 2025, while conceding that its 2030 target of 500,000 jobs is unreachable. The limit is not demand, and it is not applicant supply — unemployment sits at 33.7%. Growth has shifted into complex, judgement-heavy roles that volume screening was never built to identify.
That distinction matters well beyond South Africa. Every offshore services market eventually runs out of the work that can be screened for cheaply, and the transition is a measurement problem before it is a hiring problem.
What the 2025 numbers actually show
Business Process Enabling South Africa (BPESA) briefed the market this month on a sector that is, by its own account, performing well. ITWeb reported the 26,346 offshore jobs added as the sector's best result since 2018, with export revenue above US$422 million (R7.55 billion) and roughly nine in ten of the new jobs going to young people.
The geography is concentrated. Cape Town's BPO industry employs more than 100,000 people, around 70,000 of them in international-facing roles, and added 44,803 net new international-facing jobs across the city's current five-year municipal term. The client base is concentrated too: about 60% of the local workforce serves UK clients, with the US second, followed by Australia and other parts of Europe.
Official figures from the Department of Trade, Industry and Competition put the longer arc in context. The dtic states that between 2010 and March 2025 the GBS industry created nearly 174,000 new jobs and generated over US$2.7 billion in export revenue, with more than R808 million disbursed to qualifying firms under the GBS Incentive Programme in 2024/25.
Against that, the target. The 2021 GBS master plan modelled four scenarios spanning roughly 250,000 to 500,000 cumulative jobs, and only the highest-growth scenario reached 500,000. BPESA has now told ITWeb that 500,000 by 2030 is not achievable, projecting roughly 150,000 additional jobs instead. The stated reason is worth reading closely: complex roles "are not going to scale."
This is not a shortage of applicants
The obvious explanation — that South Africa has run out of people — does not survive contact with the labour data. Statistics South Africa's Quarterly Labour Force Survey put unemployment at 33.7% in the third quarter of 2026. The number of unemployed people rose by 97,000 to 8.6 million between July and September, even as the economy added 101,000 jobs.
So the funnel is not starved. What has changed is the composition of the work at the top of it. AI is absorbing a meaningful share of the routine, entry-level volume work that the call-centre model was built on, and the net growth is moving toward roles that BPESA characterises as requiring empathy, judgement and complex problem-solving.
Here is the uncomfortable consequence. Screening for routine voice work is easy and cheap: language proficiency, basic numeracy, availability, and a generous selection ratio that lets the job itself finish the filtering. Screening for judgement is neither easy nor cheap. The sector is not short of candidates. It is short of a defensible, affordable way to tell which of 8.6 million people can do the harder work.
Why complex roles are harder to screen, not just harder to fill
Complexity raises the value of good measurement
The intuition that selection matters less for "soft" complex roles is backwards. Job complexity is one of the better-established moderators of predictive validity in the literature. Salgado and colleagues, in a European meta-analysis published in the Journal of Applied Psychology, reported operational validities for general mental ability against job performance of .64 for high-complexity jobs, .53 for medium and .51 for low (Salgado et al., 2003).
Two honest caveats. Those coefficients rest on range-restriction corrections that have since been shown to be systematically too generous, so the absolute levels in that paper are too high; Sackett, Zhang, Berry and Lievens (2021) revise cognitive ability's operational validity to about .31. The ordering by complexity, however, is the robust part. A better measure buys you more in a complex job than in a simple one. Moving up the value chain raises the return on assessment quality rather than lowering it.
The methods that measure judgement best scale worst
The same Sackett et al. matrix explains why this is awkward in practice. Revised operational validities for widely used predictors run roughly as follows.
Stronger predictors, higher cost per candidate
- Structured interviews: .42
- Job knowledge tests: .40
Cheaper to run at volume, weaker signal
- Work sample tests: .33
- Cognitive ability tests: .31
- Integrity tests: .31
- Situational judgement tests: .26
That is the whole tension in six numbers. The instruments that best capture judgement and applied knowledge are the ones that consume assessor time. The ones that scale to thousands of applicants a week carry less signal individually. A market trying to add 150,000 complex-role hires by 2030 cannot interview its way there, and cannot screen its way there on a single cheap test either.
A screening design built for the complex half
The workable answer is sequencing rather than a single better test.
- Redo the job analysis on the new work, not the old role profile. If the role has absorbed AI tooling, the competencies have moved. Screening against last year's profile measures a job nobody does now.
- Put the cheap, scalable measure first — and let it reject, not select. A short ability or language screen is defensible as a low hurdle. It is not defensible as the decision.
- Measure judgement with scenarios, scored consistently. Scenario-based assessment is the only judgement measure that survives volume. AssessAll's AI-graded scenarios and integrity bands exist for exactly this layer, where a human rater per candidate is not affordable.
- Score the structured interview; do not merely hold one. The .42 figure describes interviews with fixed questions and anchored rating scales. An unstructured conversation does not inherit that validity.
- Verify the person as well as the skill in remote funnels. Identity assurance and proctoring integrity signals belong in the design from the start, not bolted on after a dispute.
- Keep your own validity and fairness evidence. See the next section — in South Africa this is not optional.
For campaign-shaped hiring, where volume arrives in bursts tied to a client win, consumption-based pricing fits the shape of the demand better than seat licences; AssessAll's pay-as-you-go credits start at ₹30 (about US$0.50) per assessment.
The legal point many employers have missed
South Africa constrains assessment more tightly than most offshore markets, and the constraint changed recently. Section 8 of the Employment Equity Act prohibits psychological testing and similar assessments of an employee unless the instrument:
- has been scientifically shown to be valid and reliable;
- can be applied fairly to all employees; and
- is not biased against any employee or group.
A fourth clause, requiring certification by the Health Professions Council of South Africa, was added by Act 47 of 2013 and deleted by the Employment Equity Amendment Act 4 of 2022, which came into force on 1 January 2025 (current text of section 8).
Read that carefully, because it is easy to misread as a relaxation. The validity, reliability, fairness and bias tests are untouched. What has gone is the external certification an employer could point to instead of its own evidence. The burden has not been lifted; it has been relocated onto the employer. If you screen in South Africa, you now need to hold the technical documentation, the local subgroup-difference analysis and the job-relatedness argument yourself. The Standards for Educational and Psychological Testing remain the reference for what that evidence file should contain.
The takeaway
South Africa's GBS numbers are not a story about a market running out of room — they are a story about a market running out of work that can be screened for cheaply. Any offshore destination moving up the value chain inherits the same problem: the roles that are growing are the ones where measurement quality pays off most and costs most, and the answer is a sequenced funnel with real validity evidence behind it, not a better single test.