What actually makes mainland China different
Five structural facts that change the research, not five facts about the economy.
Ownership type is the biggest pay variable in the country
The National Bureau of Statistics reports average annual wages separately for urban non-private and private units, and the gap is not marginal. For 2025 the figures were ¥129,441 and ¥71,590 respectively, with nominal growth of 4.3 per cent and 3.0 per cent.
No other market we cover has a single structural variable that moves pay this far. A China benchmark that does not state whether its comparator set is state-linked, domestic private or foreign-invested is not telling you anything usable.
Three sectors sit far above the rest
On the same 2025 release, information transmission, software and information technology services led all sectors at ¥248,752, finance followed at ¥211,164, and scientific research and technical services came third at ¥182,064. Manufacturing pay grew faster than the national average, at 5.2 per cent in non-private units and 6.4 per cent in private ones.
That ordering is stable enough to design around, and it explains most cross-sector moves we are asked to research.
Cities are separate labour markets with separate policy
Beijing, Shanghai, Shenzhen, Guangzhou, Hangzhou, Suzhou, Chengdu, Hefei and Wuhan each hold different industrial capability, and municipal talent subsidy programmes actively move senior specialists between them.
The national wage release reports by province and region — Guangdong at ¥143,284 for non-private units, the eastern region at ¥149,902 — but not by city. The city-level numbers have to come from municipal yearbooks, and we say so instead of presenting a provincial figure as a city one.
Intermediary activity is licensed; information work is not
Under the Interim Regulation on the Human Resources Market, a human resources service licence is required for intermediary activity — recommending candidates to employers, running job fairs and similar. Collecting and releasing information about labour supply and demand, and providing guidance, is treated differently, and pure consulting, assessment and training generally sits under a filing rather than a licence.
That line is exactly where Audentia operates. We produce research and analysis. We do not introduce candidates to employers for a fee and we do not charge anything contingent on a hire.
Personal information has a regulated route out of the country
Under the Personal Information Protection Law and the Cyberspace Administration of China’s provisions, personal information leaves the mainland through a security assessment, a standard contract with filing, or certification, unless an exemption applies. The certification measures took effect on 1 January 2026, and the regulator published further question-and-answer guidance on administrative policy in July 2026.
Reported volume thresholds differ between secondary summaries, and data classified as important triggers assessment regardless of volume. We agree the transfer route with you in writing before fieldwork, and where the right answer is an aggregated or in-country deliverable we say so rather than designing around it.
What we deliver in China
Scoped to the decision rather than sold as a package.
Population construction against a written definition
The definition is agreed, in writing, before fieldwork: capability, employer set, level band, geography, language requirement and explicit exclusions for mainland China.
It is reproduced on the front of the deliverable, so anyone reading the dataset later can see what was counted without asking.
Field-level verification with a stated rule
Each material field — employer, function, scope, level, location, tenure — is corroborated against a second independent source wherever one exists. Where it does not, the field carries a confidence grade rather than a silent guess.
Our rule is simple and unglamorous: we would rather hand over a smaller dataset we can stand behind than a larger one we cannot.
Local-language source work
Research is conducted against Chinese sources in the language they are published in, and names are reconciled on the local form rather than on a transliteration.
The original form is retained in the dataset alongside the romanisation, so any record can be audited back to the source it came from.
Decay modelling and refresh planning
Every dataset has a half-life. We state what we think it is for this population in this market, and what the leading indicators of staleness are.
A refresh of an existing dataset costs a fraction of the original build, which is only true because the structure and the sourcing were recorded the first time.
White-label delivery
Your templates, your branding, your confidentiality terms, your file formats. We do not contact your client market in our own name and we do not retain the right to re-use research done for you.
Where you need our analysts on a call with your client, they attend as part of your team and are briefed accordingly.
Research administration and dataset hygiene
De-duplication, entity resolution across corporate name variants, structural tagging and ingest-ready formatting for your own systems.
In mainland China entity resolution is a substantial part of the work rather than a clean-up step, because corporate and personal name forms vary between sources far more than they do in English-language markets.
Most engagements use three or four of these rather than all of them. The scoping call establishes which questions you actually need answered, and the written scope says what is in and what is deliberately out.
Pay and labour cost in China
A benchmark that does not state its basis is not a benchmark.
| Component | What it is | Where the evidence comes from |
|---|---|---|
| Base salary and the payment pattern | Usually a monthly figure, frequently paid across thirteen or more instalments. The number of payments is a real variable, not a rounding detail. | Primary research. The national statistics report annual totals, which conceal the instalment pattern entirely. |
| Ownership type of the employer | Urban non-private versus private is the structural divide the official series is built around, and foreign-invested enterprises behave differently again. | National Bureau of Statistics annual wage release, which reports non-private and private separately for the nation, by sector and by region. |
| Sector | Information technology, finance and scientific research sit well above the national average; the ordering has been stable across recent releases. | National Bureau of Statistics annual wage release, by industry category. |
| City and region | The eastern seaboard sits above the national average and inland centres below it, with municipal subsidy programmes layered on top for senior specialists. | Provincial and regional figures from the national release; city-level figures from municipal statistical yearbooks, cited individually. |
| Social insurance and housing fund | Employer contribution rates and contribution bases are set locally, so the fully-loaded cost of the same salary differs by city. | Published municipal rates and bases, quoted with the city and the effective date. |
| Equity and long-term incentive | Material in listed technology companies and in the China arms of listed multinationals; largely absent elsewhere. | Company filings and primary research. Not captured in the national wage series. |
China has one of the better national wage series of any market we cover and one of the largest gaps between that series and what a hiring decision needs. The series tells you, credibly and on the record, how sectors and ownership types compare. It will not tell you what a senior autonomous-driving engineer in Shanghai is paid.
So we report in two layers. The official layer is cited, dated and reproducible. The researched layer names its comparator set, states how many observations it rests on and carries a confidence grade. We do not blend them into a single number, because the moment you do, nobody can tell which part of it is evidence.
Regional and city markets in China
We report by market, not by national average.
| Market | What concentrates there | What a researcher has to watch |
|---|---|---|
| Beijing | Internet platforms, artificial intelligence research, state-owned group headquarters, regulators, universities and the policy ecosystem. | The densest concentration of research-grade AI capability and of state-linked employers in the country. Ownership type matters more here than anywhere, and so does the difference between an academic affiliation and an operating role. |
| Shanghai | Financial services, automotive and electric vehicles, semiconductors, pharmaceuticals, and the China headquarters of most multinationals. | The deepest multinational population on the mainland, which means a large share of candidates are legible to foreign employers — and competed for by all of them at once. |
| Shenzhen | Hardware, communications equipment, consumer electronics, electric vehicles, batteries and a very large software engineering base. | The fastest-moving market in the country and the most closely linked to Hong Kong. Engineering populations here are young, mobile and poorly captured by any dataset more than a year old. |
| Guangzhou and the Pearl River Delta | Automotive, trade and distribution, chemicals, consumer goods and a dense manufacturing supplier base. | Functions as one labour market with Shenzhen and Dongguan for many engineering roles, and as a separate one for commercial roles. The boundary has to be set in the scope. |
| Hangzhou, Suzhou, Wuxi and Nanjing | E-commerce and platform technology in Hangzhou; semiconductors, precision manufacturing and biopharma across the Jiangsu cluster. | Municipal subsidy programmes compete openly for senior specialists here, so individual mobility is partly a function of policy rather than of employer action. |
| Chengdu, Xi’an, Wuhan, Hefei and the inland centres | Aerospace, optoelectronics, displays, batteries, automotive and a growing software base, supported by large university populations. | Where cost-driven relocation of engineering capability has been most pronounced. Pay sits materially below the eastern seaboard, and retention behaves differently. |
Sectors we cover in China
Deepest where the research is hardest and the stakes are highest.
- Technology, software and artificial intelligence — the highest-paid sector in the official wage series and the deepest engineering population in the market.
- Automotive, electric vehicles and batteries — vehicle makers, battery and cell manufacturers, autonomous driving software and the supplier tiers.
- Semiconductors, electronics and displays — design, equipment, materials and fabrication, concentrated in Shanghai, Shenzhen and the Jiangsu cluster.
- Industrial manufacturing and engineering — where wage growth ran ahead of the national average in both ownership categories in 2025.
- Financial services — the second highest-paid sector nationally, spanning banking, insurance, asset management and the foreign-invested joint ventures.
- Pharmaceuticals and life sciences — domestic innovators, contract research and the China operations of global pharma.
- Supply chain, logistics and trade — a large and specialised population, increasingly focused on cross-border and nearshoring capability.
- Energy, renewables and materials — solar, wind, grid and the chemicals and materials base underneath them.
- Consumer and retail — domestic brands and the China commercial functions of international groups.
What may lawfully be recorded, and moved, in mainland China
Chinese research sits under the Personal Information Protection Law, alongside the Cybersecurity Law and the Data Security Law. For talent research the binding question is usually not whether a professional fact may be recorded but whether, and how, a dataset containing personal information may be transferred out of the mainland.
The Cyberspace Administration of China operates three routes: a security assessment, a standard contract accompanied by a filing, or certification. Provisions issued in March 2024 introduced exemptions, measures governing the certification route took effect on 1 January 2026, and the regulator published further guidance on administrative policy in July 2026. Reported volume thresholds vary between secondary summaries and we do not repeat them as settled; data classified as important triggers the assessment route regardless of volume.
What this means for an engagement. We agree the transfer route in writing before fieldwork begins, with your counsel in the loop, and we design the deliverable to fit it. For a competitor structure or a market sizing, an aggregated deliverable that contains no personal information often answers the question completely and removes the transfer problem rather than managing it. Where named individuals are genuinely required, we say what route that needs and what it will take before you commit.
On the service line: the Interim Regulation on the Human Resources Market makes intermediary activity — recommending candidates to employers, running job fairs — a licensed business, while collecting and publishing labour supply and demand information is treated differently and consulting, assessment and training generally sit under a filing. Audentia produces research and analysis. We do not introduce candidates for a placement fee and we charge nothing contingent on a hire.
One further constraint worth stating because it affects what a map can promise: post-employment non-competition restrictions are enforceable in China only for a limited period and only where the former employer continues to pay compensation, and the Ministry of Human Resources and Social Security issued compliance guidelines on their implementation in September 2025. A named individual being findable is not the same as that individual being free to move.
How a Chinese talent research project actually runs
A written scope before any fieldwork begins.
Agree the standard, not just the task
- What we establish
- What counts as a verified field, what corroboration means for this project, and what confidence grades will be used.
- Why it matters
- Research capacity bought without an accuracy standard is bought on price and delivered on hope.
Define the population in writing
- What we establish
- Capability, employer set, levels, geography, language requirement and exclusions for mainland China.
- Why it matters
- It is the only defence against a dispute about completeness after delivery.
Frame from official and registry sources
- What we establish
- The statistical and corporate frame — official labour data, company registers and filings for mainland China.
- Why it matters
- It gives a defensible starting boundary and shows where the public record stops.
Build by primary research in the local language
- What we establish
- The population itself, researched directly rather than assembled from an aggregator.
- Why it matters
- Aggregated sources inherit the gaps of the profiles they are built from, and in this market those gaps are not random.
Corroborate, grade and record the sourcing
- What we establish
- A second source per material field where one exists, a confidence grade where it does not, and the source held against the record.
- Why it matters
- A dataset whose provenance is not recorded cannot be refreshed, audited or defended.
Quality assurance against the agreed standard
- What we establish
- A sampled re-check by an analyst who did not build the records, reported as a measured error rate.
- Why it matters
- Claimed accuracy is marketing. Measured accuracy is a deliverable.
Deliver, document and hand over
- What we establish
- The dataset, the definition, the sourcing log, the QA result and the expected useful life.
- Why it matters
- You own it permanently and can refresh it with anyone, including your own team.
Talent Research, and what it is not
These get bought as if they were interchangeable. They are not.
| Commissioned research (this page) | In-house research team | Low-cost offshore | |
|---|---|---|---|
| Cost shape | Fixed project fee, variable volume | Fixed salary cost, fixed capacity | Low hourly rate, variable quality |
| Local-language capability | By design, per market | Usually one or two markets | Rarely, and rarely verified |
| Accuracy standard | Agreed and measured | Whatever you set | Usually unstated |
| Peak capacity | Scales with the brief | Capped by headcount | Scales, with quality risk |
| Provenance recorded | Yes, per field | Depends on your process | Usually not |
| Confidentiality | White-label, contractual | Total | Varies |
| Best for | Peak load, new markets, a standard you can audit | Continuous core coverage | High-volume, low-stakes list building |
Language is the other thing worth being explicit about. Mainland Chinese research runs in Simplified Chinese. The public record — company registry filings, patent and academic material, industry press, listed-company disclosure — is overwhelmingly Chinese-language, and the English layer over it is partial and lags.
The transliteration problem is sharper here than in Japan. Pinyin collapses distinct characters onto identical romanised strings, so a common name in pinyin can correspond to dozens of different people, and matching on it produces confident-looking nonsense. Corporate names compound it: a group’s registered Chinese name, its English trading name and its subsidiaries’ names frequently share no common element.
We research and reconcile on the Chinese form, hold the characters in the dataset alongside the romanisation, and deliver analysis in English. The market term for this work is 人才地图; the commercial market for contingent placement is 猎头, and the two are not the same purchase.
What talent research costs in China
Fixed project fees, scoped on the research rather than on the salary of a role.
| Engagement | What it covers | Typical duration |
|---|---|---|
| Scoping and feasibility | An honest read on whether the question is researchable in mainland China, what the sources will support and what the limits are. Written, before you commit to anything. | Days, at no charge |
| Single-decision project | One function or one defined population in mainland China, scoped to a named business decision. | Four to six weeks |
| Multi-function or multi-city project | Several functions, or several Chinese markets held to one definition so the comparison is like for like. | Six to ten weeks |
| Multi-market programme | China alongside our other eleven markets, with the population definition held constant across all of them. | Scoped per programme |
| Refresh of an existing dataset | Re-verification of a population we built previously, with decay reported against the original. | One to three weeks |
Every engagement is a fixed fee agreed in writing before work begins. There is no commission, no percentage of salary and no success fee on any service we sell — which in mainland China is a regulatory position as well as a commercial one, because fee-charging placement is a licensed activity there and research is not.
Fees are quoted as a fixed project fee and can be contracted in pounds sterling, US dollars or renminbi. Our published fee bands sit on the talent research cost page; the band that applies depends on the population size, the number of fields to be verified and the language coverage required, and the scoping call establishes all three before anything is quoted.
What drives the fee in mainland China is verification depth, not headcount. A population of 200 names corroborated on six fields costs more than 600 names corroborated on two, and the scope document says which you are buying. The fee comparison calculator shows what a fixed research fee comes to against a contingency or retained model.
Talent Research in China — FAQs
How do you build a China dataset without a transfer problem?
By settling the question first. If the deliverable can be aggregated, there is no personal information to transfer and the problem disappears. If named individuals are required, the route — security assessment, standard contract with filing, or certification — is agreed in writing with your counsel before fieldwork.
The certification route’s measures took effect on 1 January 2026 and the regulator published further guidance in July 2026, so this is a live area and we treat it as one.
What is the pinyin problem and how do you solve it?
Pinyin collapses distinct characters onto identical romanised strings, so a common name in pinyin can correspond to a very large number of different people. Matching on it produces confident-looking nonsense.
We research and reconcile on the Chinese characters and hold them in the dataset alongside the romanisation. Entity resolution on corporate names is a comparable problem and gets the same treatment.
How fast do Chinese records decay?
Faster than any other market we cover. Engineering populations in Shenzhen and Hangzhou move quickly, municipal subsidy programmes actively move senior specialists between cities, and the public record updates unevenly behind both.
We state the expected useful life, which for some populations is measured in months rather than years, and we say which fields go first.
Can you resolve corporate structures reliably?
To a useful level, with care. A group’s registered Chinese name, its English trading name and its subsidiaries’ names frequently share no common element, so entity resolution is substantive research rather than a clean-up pass.
We record the resolution decisions we made, so you can see where two records were merged and why.
Do you need a Chinese licence for this work?
The Interim Regulation on the Human Resources Market makes intermediary activity licensable — recommending candidates to employers, running job fairs. Information collection and release, and consulting and assessment, are treated differently.
We produce research and analysis, do not introduce candidates for a placement fee, and charge nothing contingent on a hire.
Can you work white-label for a firm operating in China?
Yes, with the data position agreed up front as part of the engagement rather than as a separate conversation later.
Your branding, your templates, your confidentiality terms, and no approach to your client market in our own name.
What we will not tell you about China
Because a research supplier that never says "we do not know" is not worth buying from.
- City-level average wages for Beijing, Shanghai and Shenzhen are not in the National Bureau of Statistics national release. Where a decision needs them we cite the relevant municipal statistical yearbook, and where we have not obtained one we say the figure is not established rather than substituting the provincial number.
- Widely circulated figures for the shortage of new-energy vehicle and artificial intelligence professionals are projections, and several originate with firms that sell recruitment services. We cite them as projections, attributed, or not at all.
- There is no reliable published measure of senior executive mobility between ownership types. It is one of the most commercially interesting questions in the market and one of the least documented, and we treat it as a research question rather than a statistic.
Primary sources and how we check
- Average Annual Wages of Employees in Urban Units in 2025National Bureau of Statistics of China
- National statistics, annual data and statistical yearbooksNational Bureau of Statistics of China
- Employment, human resources market and labour policyMinistry of Human Resources and Social Security
- Personal Information Protection Law of the People’s Republic of ChinaNational People’s Congress
- Cross-border data transfer policy and guidanceCyberspace Administration of China
- Doing business, investment and regulatory guides for ChinaDezan Shira & Associates
- Investment environment and sector guidesJETRO
Figures quoted on this page were checked against the sources above on 30 October 2026. Statutory rates, thresholds, licence conditions and quotas change, and several of these series are revised after first publication: where a number drives a budget or a legal obligation, confirm it against the primary source before you rely on it.