The Typing-Work Index:
what German SMBs advertise for automatable work.

Correction 1, 31 July: Until 31 July the title filter did not run in the standard query, so the count included everything the search returned: apprenticeships, working students, and trade or medical roles. Measured against the same run, 85 of 624 positions were clearly not office work, that is 14 percent.

Correction 2, 2 August: Until 2 August the check against the agency's own detail records only ran on request, and therefore not in the series run. The search parameter for temporary work only covers employee leasing, not private job placement. Measured against the same data, 180 of 522 positions carried the agency's own private-placement flag, that is 34.5 percent; on top of that 13 large corporates, 21 medical billing roles and 9 duplicates or records without a title. The cleaned run of the same day yields 318 positions instead of 522.

Correction 3, 4 August: On 4 August the search endpoint used until then was retired and now returns 404. The new version can filter out private job placement itself. The filter therefore now sits before the page cut instead of after it: a result page previously returned 100 records of which roughly a third was discarded, whereas now all 100 are usable. Measured across the same eight search terms that is about 166 additional usable records. The rise from 308 to 416 positions is therefore an artefact of method and not a movement in the market. 4 August is the new baseline.

Correction 4, 4 August: The filter knew large corporates but not municipalities, universities, hospitals or non-profit providers. For a figure that explicitly claims to describe the Mittelstand, those do not belong in it. On top of that, construction, warehouse and field-service titles were missing from the list of out-of-scope roles. Measured against the same run that is 34 plus 10 of 416 positions, together 10.3 percent. The cleaned figure for the same day is 377 positions at 335 companies.

Correction 5, 4 August: Fifth correction, 4 August 2026: the carrier list from the fourth correction barely worked in German, because it was built with word boundaries while German puts the carrier inside the compound. Ten municipalities were still being counted. The cleaned run of the same day yields 363 positions instead of 377.

Correction 6, 5 August: Sixth correction, 5 August 2026: Austrian postings entered the count through the European job network, the title list knew 'Ausbildung' but not 'Auszubildende', and corporates advertise under abbreviations that do not contain their name. Together 4.4 percent of the previous day's count.

Correction 7, 5 August: Seventh correction, 5 August 2026: two previously adopted checks were never invoked by the survey tool. They sat in the shared file and on this page, but had no effect. Added to that were staffing providers without the agency's flag and corporates whose names no pattern matches. Together 20.7 percent of the previous day's count.

Correction 8, 5 August: Eighth correction, 5 August 2026: every previous cross-check ran against the sample. Large employers fill the deeper result pages and barely appear there. In the full survey 8.0 percent drop out, in the sample only 0.7. Cross-checks now run against the full survey.

Correction 9, 8 August: Ninth correction, 8 August 2026: five organisations remained in the data because their names carry no public-sector marker, among them the Jülich research centre trading as a GmbH, a statutory health insurer and a trade union federation. This removes 2.0 percent of the postings. They were found by reading the full list of entities by hand, not by a pattern search.

Correction 10, 8 August: Tenth correction, 8 August 2026: state figures move by up to 98 percent between two surveys three days apart; only the three largest states stay stable. The minimum case count rises from 8 to 20, and the measured volatility is now published beneath the table instead of sitting in our notes.

Median checked, 4 August: Since 4 August the agency's interface also returns the salary ranges the companies state themselves. Assumed: 3,820 euros gross per month, so 45,840 euros a year (BA pay atlas, the lowest of the values considered, data as of 31 Dec 2024). Measured: 42,600 euros a year, median of the ranges companies state themselves, taken from the full survey of the same day. That is 7.1 percent apart, and the assumed value is the higher one. The caution in this figure therefore lies in the sample, not in the salary. Using the measured median instead of the assumed one would lower the weekly figure from 13.2 to 12.3 million euros. Counting the same week as a full survey instead of a sample yields 23.4 million. The second gap is roughly twelve times the first. The limit: Only 12.0 percent of the postings in the full survey state a range at all (61 of 510), 60 of them usable. Whoever publishes a salary may not be the average, so this check places the assumed figure, it does not replace it.

The readings below were taken before one of these corrections. They do not all err in the same direction: the July values are too high, because neither the title filter nor the agency-flag check was running. The two August values before the 4th are correctly filtered but sit on a thinner sample, because back then the filter ran after the page cut and discarded part of every page. Neither can be recalculated, because the individual postings of those runs were not stored. They stay visible because they have been quoted, but they are not compared with the series.

• 16 July475 positions, 401 companies, 21.8 M€
• 30 July594 positions, 426 companies, 27.2 M€
• 31 July539 positions, 400 companies, 24.7 M€
• 1 August509 positions, 372 companies, 23.3 M€
• 2 August318 positions, 282 companies, 14.6 M€
• 3 August308 positions, 279 companies, 14.1 M€
• 4 August416 positions, 375 companies, 19.1 M€
• 4 August377 positions, 335 companies, 17.3 M€
• 4 August363 positions, 323 companies, 16.6 M€
• 4 August347 positions, 311 companies, 15.9 M€
• 5 August323 positions, 286 companies, 14.8 M€
• 5 August256 positions, 229 companies, 11.7 M€
• 5 August303 positions, 276 companies, 13.9 M€
• 8 August300 positions, 267 companies, 13.8 M€

Coming from a published article? Figures of 475 positions, 401 companies and 21.8 M€ (surveyed 16 July) predate all ten corrections above. They cannot be recalculated retroactively, because the individual postings were not stored. What can be stated is what each of those corrections removed when it was measured against the stock of its own day: 13.6 percent titles outside office work, then 34.5 percent private job placement, corporates and medical billing, then 10.3 percent public-sector and non-profit employers plus non-commercial titles, then 3.7 percent further public-sector employers that the word boundaries had let through, then 4.4 percent foreign postings, apprenticeships and corporates advertising under subsidiary names, then 20.7 percent staffing providers, further corporates and municipal subsidiaries, plus everything two never-invoked checks should have removed, then 0.7 percent further corporates and staffing providers, visible only in the full survey, then 2 percent large-scale research bodies, statutory health insurers, unions and two corporate subsidiaries not carrying the group name, then 0 percent no postings, but four state figures that are no longer marked reliable. Those rates were measured on different days, so they compound to an estimate rather than a measurement: roughly 36 percent of that count would survive today's chain. What is measured is the value at the top of this page. The direction of the finding is unchanged, the level was too high.

Coming from a published article? Figures of 256 positions, 229 companies and 11.7 M€ (surveyed 5 August) predate three of the corrections above. They cannot be recalculated retroactively, because the individual postings were not stored. What can be stated is what each of those corrections removed when it was measured against the stock of its own day: 0.7 percent further corporates and staffing providers, visible only in the full survey, then 2 percent large-scale research bodies, statutory health insurers, unions and two corporate subsidiaries not carrying the group name, then 0 percent no postings, but four state figures that are no longer marked reliable. Those rates were measured on different days, so they compound to an estimate rather than a measurement: roughly 97 percent of that count would survive today's chain. What is measured is the value at the top of this page. The direction of the finding is unchanged, the level was too high.