Key findings
- The workforce is big and it turns over. The US Bureau of Labor Statistics (BLS) counts 2,666,000 customer service representative jobs in 2025 and about 289,500 openings a year, all of them from workers leaving the occupation. Eurostat counts 3,606.1 thousand "customer services clerks" in the EU-27 in 2025. Both are public and free to re-check. [S1, S2]
- Initial training is short. BLS says on-the-job training for US customer service representatives usually lasts 2 to 4 weeks, and finance and insurance roles may need several months. [S1]
- The headcount is flat to falling. BLS projects a 5% decline from 2025 to 2035. The EU-27 series moved from 3,492.7 thousand in 2011 to 3,606.1 thousand in 2025, with breaks in the series. [S1, S2]
- Simulation-based training has the strongest public evidence outside call centres. A pooled analysis of 609 studies in health professions education found large effects of simulation versus no intervention on knowledge (1.20) and skills (1.09 to 1.18). Those are medical trainees, not call centre agents. The transfer is an assumption, and we say so. [S3]
- Design features matter. A second pooled analysis of 289 studies found the largest skills effects for range of difficulty (0.68), repetitive practice (0.68), distributed practice (0.66) and interactivity (0.65). Feedback came in at 0.44. [S4]
- What we could not find. No public primary source we could retrieve gives the share of contact centre agents who get structured role-play, average ramp time to proficiency, or UK headcount for the occupation. These are marked as gaps, not filled.
1. Why this report exists
Contact centre training budgets and onboarding plans are usually justified with numbers that nobody can trace. A vendor survey here, a conference slide there. We wanted one page where every figure can be clicked back to a government statistics office or a peer-reviewed journal.
The result is narrower than a typical industry report. We left out anything we could not source to a public primary document. We did not use any customer, product or usage data. Where the public record is silent, the report says so in a box labelled GAP.
If you write about contact centre workforce or training, you can reuse any figure here with its source. The chart data tables are written to be copied, and each has a CSV download under CC BY 4.0.
2. Methodology
Scope. Agent-facing, phone-heavy customer service work. We include statistics for the occupation group, and research on how people learn conversational and procedural skills through practice.
Source rules.
- Only public primary sources: statistical offices, and peer-reviewed journals located through their DOI or a bibliographic database record.
- No vendor surveys, no press releases, no analyst estimates, no blog posts, no customer data, and no data from our own product.
- Every number carries a source ID and URL (see Sources) and was accessed on 2026-11-09.
- We copy numbers as published. Where we compute something, it is marked DERIVED and the arithmetic is shown.
How the numbers were retrieved.
- BLS figures were read from the Occupational Outlook Handbook page for customer service representatives, shown as last modified on 29 October 2026.
- Eurostat figures were pulled from the public dissemination API for dataset LFSA_EGAI2D (employed persons by detailed occupation, ISCO-08 two-digit level), filtered to occupation OC42, ages 15 to 74, unit thousand persons. Dataset update stamp: 2026-10-30.
- Research figures were read from the bibliographic records of the papers (Europe PMC and Crossref).
Limits you should know about.
- Occupation definitions differ. The BLS category is "customer service representatives" (SOC 43-4051). The Eurostat category ISCO-08 group 42 is "customer services clerks", which is broader and includes occupations such as bank tellers and contact centre information clerks. The two figures must not be added or compared as like for like.
- Series breaks. Eurostat flags a break in the EU-27 time series in 2011 and 2021. We plot the series with the breaks marked and draw no conclusion from a single year-to-year change.
- Country figures carry flags. In the 2024 data Eurostat flags Spain "definition differs" and France "break in time series; definition differs". Small samples in smaller countries also make the figures less precise. We show them as published.
- Education research is not contact centre research. The simulation meta-analyses are on health professions trainees. None of them measure call centre agents. Section 6 treats that as an open question, not as a finding.
- Effect sizes. The simulation papers report standardised mean differences. By convention, around 0.2 is small, 0.5 moderate and 0.8 large. That convention is general statistical practice, not a finding about contact centres.
- Deliberate practice figures are left out. A widely quoted 2014 meta-analysis on deliberate practice has a published corrigendum (Psychological Science, 2018, 29(7), 1202-1204). We could not read the corrigendum text on 2026-11-09, so we quote no number from that paper.
3. How many people do this work
United States
BLS reports 2,666,000 customer service representative jobs in 2025. [S1] The median annual pay was $44,770 ($21.53 an hour) in 2025. [S1]
Where they work, by share of employment: retail trade 15%, insurance carriers and related activities 12%, business support services 8%, professional, scientific and technical services 7%, healthcare and social assistance 6%. [S1] BLS gives shares, not job counts, for these industries, so we do not turn them into counts.
Retail and insurance employ more than a quarter of US customer service representatives
DERIVED: 15 + 12 = 27. The five industries BLS lists sum to 48; the other 52 is not itemised by BLS.
| Industry | Share of employment (%) |
|---|---|
| Retail trade | 15 |
| Insurance carriers and related activities | 12 |
| Business support services | 8 |
| Professional, scientific, and technical services | 7 |
| Healthcare and social assistance | 6 |
| All other industries (DERIVED, not itemised by BLS) | 52 |
Source: US Bureau of Labor Statistics, Occupational Outlook Handbook, S1, accessed 2026-11-09. Chart: DIRI AI. Table licence: CC BY 4.0. Download CSV
Call centre work is called out explicitly: BLS says jobs in call centres may require shifts early in the morning or late at night. [S1]
European Union
Eurostat's EU-27 figure for "customer services clerks" (ISCO-08 group 42, ages 15 to 74) was 3,632.1 thousand in 2024 and 3,606.1 thousand in 2025. [S2]
About 3.6 million people work as customer services clerks in the EU-27
ISCO-08 group 42 is broader than call centre agents. Eurostat flags a break in the series in 2011 and 2021.
| Year | Thousand persons | Flag |
|---|---|---|
| 2011 | 3,492.7 | break in series |
| 2012 | 3,616.3 | |
| 2013 | 3,628.6 | |
| 2014 | 3,699.1 | |
| 2015 | 3,703.8 | |
| 2016 | 3,741.1 | |
| 2017 | 3,805.6 | |
| 2018 | 3,788.0 | |
| 2019 | 3,742.0 | |
| 2020 | 3,590.8 | |
| 2021 | 3,607.6 | break in series |
| 2022 | 3,626.4 | |
| 2023 | 3,575.0 | |
| 2024 | 3,632.1 | |
| 2025 | 3,606.1 |
Source: Eurostat, LFSA_EGAI2D, S2, accessed 2026-11-09. Chart: DIRI AI. Table licence: CC BY 4.0. Download CSV
Women made up 2,542.3 thousand of the 3,632.1 thousand in 2024. DERIVED: 2,542.3 / 3,632.1 = 70.0%. Men: 1,089.8 thousand. [S2] Against all employment in the EU-27 in 2024 (206,449.6 thousand), the group is DERIVED: 3,632.1 / 206,449.6 = 1.76% of employed people. [S2]
Seven in ten are women
DERIVED: 2,542.3 / 3,632.1 = 70.0%. ISCO-08 group 42 is broader than call centre agents. Eurostat flags a break in the series in 2011 and 2021.
| Group | Thousand persons |
|---|---|
| Women | 2,542.3 |
| Men | 1,089.8 |
| Total | 3,632.1 |
Source: Eurostat, LFSA_EGAI2D, S2, accessed 2026-11-09. Chart: DIRI AI. Table licence: CC BY 4.0. Download CSV
Country values for 2024, in thousands: Spain 936.4 (flagged), Germany 598.7, Italy 508.3, France 341.7 (flagged), Netherlands 195.0, Poland 155.0, Czechia 56.0, Ireland 54.8. [S2]
Spain and Germany are the biggest employers in this group among the eight countries shown
Eurostat flags: d = definition differs (Spain; France), b = break in time series (France). Eight countries chosen for relevance to DIRI AI locales and size; not a full list. No per-capita ranking is made.
| Country | Thousand persons | Flag |
|---|---|---|
| Spain | 936.4 | d |
| Germany | 598.7 | |
| Italy | 508.3 | |
| France | 341.7 | bd |
| Netherlands | 195.0 | |
| Poland | 155.0 | |
| Czechia | 56.0 | |
| Ireland | 54.8 |
Source: Eurostat, LFSA_EGAI2D, S2, accessed 2026-11-09. Chart: DIRI AI. Table licence: CC BY 4.0. Download CSV
GAP: United Kingdom
We looked for an Office for National Statistics headcount for customer service occupations (SOC 2020 sub-major group 72). ONS says it no longer publishes occupation figures from the Labour Force Survey and points users to the Annual Population Survey on NOMIS. [S6] The Census 2021 bulletin for England and Wales names customer service occupations but gives no figure for them on the page we read. [S7] A figure may exist in the Census 2021 occupation minor group table (TS064), which we did not retrieve. We leave the UK out of the charts.
4. A workforce that leaves
The most useful single number in the BLS profile is the replacement figure. BLS expects about 289,500 openings a year on average over 2025 to 2035, and says all of them come from the need to replace workers who transfer to other occupations or leave the labour force, since the occupation itself is projected to shrink. [S1]
DERIVED: 289,500 / 2,666,000 = 10.9% of current jobs, every year, need to be filled by someone new. This is our arithmetic on two BLS numbers. BLS does not publish it as an attrition rate, and the openings include people who move to other occupations as well as those who leave work entirely.
About 1 in 9 US customer service jobs opens up every year, and all of it is replacement
DERIVED: 289,500 / 2,666,000 = 10.9%. This is our arithmetic, not a BLS attrition rate. BLS projects employment to change by -5% from 2025 to 2035.
| Measure | Value |
|---|---|
| Jobs, 2025 | 2,666,000 |
| Average annual openings, 2025-35 | about 289,500 |
| Projected employment change, 2025-35 | -5% (-141,800 jobs) |
| Openings as share of jobs (DERIVED) | 10.9% |
Source: US Bureau of Labor Statistics, Occupational Outlook Handbook, S1, accessed 2026-11-09. Chart: DIRI AI. Table licence: CC BY 4.0. Download CSV
What follows from the published numbers, and only from them:
- A team that holds its size has to recruit and onboard a large share of its headcount again and again.
- A two to four week initial training period [S1] is repeated for every one of those new starters.
- Anything that shortens or strengthens that period is multiplied by the replacement volume.
What the published numbers do not tell you: the attrition of any given centre, the cost of a replacement, or how long it takes a new agent to reach full productivity. Those figures circulate widely in vendor material. We could not find them in a public primary source, so they are not here.
The outlook
BLS projects employment of customer service representatives to decline 5% from 2025 to 2035, a change of 141,800 jobs. [S1] The EU-27 series is roughly flat across the longer run: 3,492.7 thousand (2011), a high of 3,805.6 thousand (2017), 3,590.8 thousand (2020) and 3,606.1 thousand (2025). [S2] DERIVED: 2025 is 5.2% below the 2017 peak and 3.2% above 2011, across series breaks flagged by Eurostat at 2011 and 2021. We do not read a trend into it.
5. How agents are trained today (what the public record says)
The public record here is thin, and it is one source.
BLS describes entry as a high school diploma or equivalent, with training that is usually short-term, about 2 to 4 weeks, and guidance from an experienced worker for the first few weeks. Training often covers answering questions, company products and services, and computer and telephone use. Workers in finance and insurance may need several months to learn complex financial regulations, and some roles need a state licence. [S1]
Most US customer service training is measured in weeks. Regulated roles take months.
BLS publishes no number of months for finance and insurance roles. None is drawn.
| Role group | Initial training length as published |
|---|---|
| Customer service representatives, typical | 2 to 4 weeks |
| Finance and insurance | several months (no exact figure) |
Source: US Bureau of Labor Statistics, Occupational Outlook Handbook, S1, accessed 2026-11-09. Chart: DIRI AI. Table licence: CC BY 4.0. Download CSV
GAP: structured practice
We found no public primary source on how many agents get role-play, simulation or call shadowing during onboarding, how long they spend in it, or how it is assessed. Vendor and trade-press surveys exist, but they are not primary, and their samples are rarely documented. We did not use them. If you have access to a regulator, works council or university survey that measures this, we would like to add it to the next edition.
6. What the research says about practice and simulation
This is the part of the report where the evidence is strongest and the transfer is weakest. Read the limits next to each number.
6.1 Simulation versus no training (health professions)
Cook and colleagues pooled 609 studies enrolling 35,226 trainees. Compared with no intervention, technology-enhanced simulation had the pooled effects below (95% confidence intervals in the table). [S3]
Simulation training beats no training on knowledge and skills. The gap narrows as outcomes get closer to the real job.
Health professions trainees, not contact centre agents. Most comparisons are against no intervention. 609 studies, 35,226 trainees.
| Outcome | Effect | 95% CI low | 95% CI high | Studies |
|---|---|---|---|---|
| Knowledge | 1.20 | 1.04 | 1.35 | 118 |
| Time skills | 1.14 | 1.03 | 1.25 | 210 |
| Process skills | 1.09 | 1.03 | 1.16 | 426 |
| Product skills | 1.18 | 0.98 | 1.37 | 54 |
| Time behaviours | 0.79 | 0.47 | 1.10 | 20 |
| Other behaviours | 0.81 | 0.66 | 0.96 | 50 |
| Direct effects on patients | 0.50 | 0.34 | 0.66 | 32 |
Source: Cook et al., JAMA 2011, S3, accessed 2026-11-09. Chart: DIRI AI. Table licence: CC BY 4.0. Download CSV
Read it this way. Effects are largest on what people know and on skills measured in the training setting. They fall as the outcome moves closer to real-world results: 0.79 to 0.81 for behaviours and 0.50 for patient outcomes. For a contact centre, the equivalent step is from "did well in practice" to "handled the live call better". No pooled study of that step in call centres is cited here. That is a gap.
Also note the comparison group. Most of these studies compare simulation with no intervention at all. That is a low bar. It tells you simulation beats nothing, not that it beats a good coaching programme.
6.2 What makes a simulation work
The same group later compared instructional design features across 289 studies (18,971 trainees, 208 randomised trials). The table shows pooled effects on skills outcomes. [S4]
Variety, repetition, spacing and interaction show the larger effects. Several rest on very few studies.
Solid bar: p below 0.05. Hollow bar: p of 0.05 or more. 289 studies, 18,971 trainees, 208 randomised trials. Health professions trainees.
| Design feature | Effect (skills) | Studies | p-value | Significant at 0.05 |
|---|---|---|---|---|
| Range of difficulty | 0.68 | 20 | < 0.001 | yes |
| Repetitive practice | 0.68 | 7 | 0.06 | no |
| Distributed practice | 0.66 | 6 | 0.03 | yes |
| Interactivity | 0.65 | 89 | < 0.001 | yes |
| Multiple learning strategies | 0.62 | 70 | < 0.001 | yes |
| Individualised learning | 0.52 | 59 | < 0.001 | yes |
| Mastery learning | 0.45 | 3 | 0.57 | no |
| Feedback | 0.44 | 80 | < 0.001 | yes |
| Longer time | 0.34 | 23 | 0.005 | yes |
| Clinical variation | 0.20 | 16 | 0.24 | no |
| Group training | -0.22 | 8 | 0.09 | no |
Source: Cook et al., Medical Teacher 2013, S4, accessed 2026-11-09. Chart: DIRI AI. Table licence: CC BY 4.0. Download CSV
Three cautions apply. First, several rows rest on few studies (repetitive practice 7, distributed practice 6, mastery learning 3), and some have p-values above 0.05, so they are signals, not settled results. Second, "clinical variation" is a medical term for varying the cases a trainee sees. Third, the negative figure for group training (-0.22, 8 studies, p = 0.09) is not statistically distinguishable from zero.
What we can say: in this body of work, practice that varies in difficulty, is spread over time, and is interactive shows the larger pooled effects. A single feature is never a recipe. The paper's own title says "comparative effectiveness", and the numbers are comparisons between design choices, not guarantees.
6.3 Deliberate practice: valuable, but not magic
The best-known claim in this area is that expert performance is built by deliberate practice: structured, effortful activity aimed at improving a specific aspect of performance, with feedback. Ericsson, Krampe and Tesch-Romer set out that account in Psychological Review in 1993. [S8] A 2014 meta-analysis by Macnamara, Hambrick and Oswald concluded that deliberate practice is important, but not as important as has been argued. [S5] That paper has a published 2018 corrigendum that we could not read, so we give none of its numbers. The direction of its conclusion matters for training managers: practice volume alone is not a complete answer, and other factors (selection, the quality of feedback, the task itself, the customer mix) carry weight. We make no claim about contact centres specifically.
6.4 Feedback and training design in organisations
Two widely cited reviews are worth reading in full, even though we could not extract their numbers: a meta-analysis of training design and evaluation features in organisations (Arthur and colleagues, Journal of Applied Psychology, 2003) [S9], and a meta-analysis of feedback interventions (Kluger and DeNisi, Psychological Bulletin, 1996) [S10]. We list them without quoting any figure from them. The simulation paper's feedback figure (0.44) [S4] is the only feedback number in this report.
7. What a training manager can take from this
These are reading notes, not product claims. Each one is tied to the evidence above and to its limit.
- Plan for replacement, not for growth. If your team is shrinking or flat, and roughly one in ten US jobs in the occupation needs refilling each year [S1, DERIVED], your onboarding programme is a core process, not a project. Check your own attrition. It may be far from the national average.
- Judge training by behaviour, not by test scores. In the simulation evidence, effects shrink as outcomes get closer to real work (1.20 for knowledge, 0.79 to 0.81 for behaviours, 0.50 for patient outcomes). [S3] Decide in advance which live-call measure you will watch.
- Vary the difficulty. Range of difficulty had the largest pooled skills effect in the design-feature comparison (0.68, 20 studies, p < 0.001). [S4]
- Space it out. Distributed practice showed 0.66 on 6 studies (p = 0.03). [S4] It is a signal. A short daily slot is a cheap way to test it in your own team.
- Do not compare your new simulation with nothing. The headline numbers in [S3] are against no intervention. Compare against what you do today.
- Pair practice with specific, timely feedback on named behaviours. Practice volume alone is not a complete answer. [S5]
- Keep regulated roles separate. BLS says finance and insurance roles may need several months of training and some need a licence. [S1] A 2 to 4 week average does not describe them.
8. Open questions and gaps
| # | Gap | Why it matters | What would close it |
|---|---|---|---|
| G1 | Share of agents who receive structured role-play or simulation in onboarding | The core adoption number for the topic | A regulator, works council or university survey with a documented sample |
| G2 | Time to proficiency for new agents | The cost side of every training decision | A peer-reviewed or statistical-office study of call centre ramp time |
| G3 | UK headcount for customer service occupations (SOC 72) | UK press needs a local number | ONS or NOMIS APS table, or Census 2021 TS064, retrieved and read directly |
| G4 | Simulation effects on live call outcomes | The transfer step from practice to the job | Controlled study in a contact centre setting |
| G5 | Agent attrition from a public statistics office, by country | Replacement volume outside the US | National labour statistics on job tenure for ISCO-08 group 42 |
| G6 | Industry-body publications | Context from trade associations | Sources we could verify as primary with documented methods (none retrieved in this edition) |
| G7 | Deliberate practice effect by domain (Macnamara and colleagues, 2014) | A corrigendum exists and its text could not be read, so the percentages are not used | Read the 2018 corrigendum and update the figures |
9. Sources
All accessed 2026-11-09.
- S1. US Bureau of Labor Statistics, Occupational Outlook Handbook, Customer Service Representatives (page last modified 2026-10-29). https://www.bls.gov/ooh/office-and-administrative-support/customer-service-representatives.htm
- S2. Eurostat, dataset LFSA_EGAI2D, Employed persons by detailed occupation (ISCO-08 two digit level); occupation OC42 customer services clerks, ages 15 to 74, thousand persons; data updated 2026-10-30. https://ec.europa.eu/eurostat/databrowser/view/lfsa_egai2d/default/table API query used: https://ec.europa.eu/eurostat/api/dissemination/statistics/1.0/data/lfsa_egai2d?format=JSON&lang=en&isco08=OC42&sex=T&age=Y15-74&unit=THS_PER&geo=EU27_2020 (total employment: the same query with isco08=TOTAL and time=2024).
- S3. Cook DA, Hatala R, Brydges R, Zendejas B, Szostek JH, Wang AT, Erwin PJ, Hamstra SJ. Technology-enhanced simulation for health professions education: a systematic review and meta-analysis. JAMA 2011;306(9):978-988. https://doi.org/10.1001/jama.2011.1234
- S4. Cook DA, Hamstra SJ, Brydges R, Zendejas B, Szostek JH, Wang AT, Erwin PJ, Hatala R. Comparative effectiveness of instructional design features in simulation-based education: systematic review and meta-analysis. Medical Teacher 2013;35(1):e867-e898. https://doi.org/10.3109/0142159X.2012.714886
- S5. Macnamara BN, Hambrick DZ, Oswald FL. Deliberate practice and performance in music, games, sports, education, and professions: a meta-analysis. Psychological Science 2014;25(8):1608-1618. https://doi.org/10.1177/0956797614535810 (no figures used) A corrigendum exists (Psychological Science 2018;29(7):1202-1204, https://doi.org/10.1177/0956797618769891). Its text could not be retrieved on 2026-11-09, so no figure from this paper is used.
- S6. Office for National Statistics, Dataset EMP09: Employees and self-employed by occupation (discontinued; Labour Force Survey occupation statistics are no longer published, Annual Population Survey figures are on NOMIS). https://www.ons.gov.uk/employmentandlabourmarket/peopleinwork/employmentandemployeetypes/datasets/employeesandselfemployedbyoccupationemp09 (no figures used)
- S7. Office for National Statistics, Industry and occupation, England and Wales: Census 2021. https://www.ons.gov.uk/employmentandlabourmarket/peopleinwork/employmentandemployeetypes/bulletins/industryandoccupationenglandandwales/census2021 (no figures used)
- S8. Ericsson KA, Krampe RT, Tesch-Romer C. The role of deliberate practice in the acquisition of expert performance. Psychological Review 1993;100(3):363-406. https://doi.org/10.1037/0033-295X.100.3.363 (no figures used)
- S9. Arthur W, Bennett W, Edens PS, Bell ST. Effectiveness of training in organizations: a meta-analysis of design and evaluation features. Journal of Applied Psychology 2003;88(2):234-245. https://doi.org/10.1037/0021-9010.88.2.234 (no figures used)
- S10. Kluger AN, DeNisi A. The effects of feedback interventions on performance: a historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin 1996;119(2):254-284. https://doi.org/10.1037/0033-2909.119.2.254 (no figures used)
10. How to cite and reuse
Cite as: DIRI AI (2026). Contact centre agent training: what the public data says (2026 edition). https://diriai.com/resources/research/contact-centre-agent-training-2026
Licence: the compiled tables and CSV files are licensed CC BY 4.0. Reuse them with attribution to DIRI AI and to the original source named in each file. The underlying data stay under the terms of the BLS, Eurostat and the journals.
Please link to the original source for each figure rather than to this page alone. If you find an error or a better public source, write to hello@diriai.com and we will correct the report and note the change in the version log.
About
DIRI AI is a phone-call practice tool built in Central Europe: salespeople and call centre staff rehearse realistic calls and get a written coaching report. This report contains no data from the product, and no customer data of any kind. The figures above come from the sources listed, and from the public record alone.
Want to practise a call before the real one? Start free
Version log
- 2026-11-09: first edition. Every figure re-read at its source on 2026-11-09. Eurostat country flags corrected to the API values (Spain d; France b and d; Ireland none). Deliberate practice percentages removed pending the 2018 corrigendum.