All Carrots and No Sticks: A Case Study on Social Credit Scores in Xiamen and Fuzhou

In 2018 Xiamen and Fuzhou, two cities in Fujian province, a region on the coast in the southeast most famous for its historic trader links and global diaspora population, became one of just a handful of cities in China with their own city-level personal credit scores (个人信用评分). These are algorithmically created scores, using data gathered by the local government to assess citizens’ level of “promise keeping” (守信), that can be used at places like hospitals and tourist sites, or when paying school tuition fees and borrowing books. However, a look under the hood reveals a reality far from any utopian or dystopian picture.

In April, Xiamen introduced Bailu score (白鹭分), named after the city’s native bird Bailu (egret). A few months later in June, Fuzhou went live with Moli score (茉莉分) named after its own famous Jasmine flower. Much has been written on China’s Social Credit System (SCS) in global media, including often inaccurate portrayals attributing the denial of some citizens’ ability to buy high-speed rail and flight tickets to a ‘low social credit score’. In reality the Social Credit System, as Shazeda Ahmed outlines, is a broad policy project for encouraging individuals, business, legal institutions, and government itself to be more “trustworthy” (守信, shouxin) through a mix of measures. One such measure, the Blacklist/Redlist system, responsible for the aforementioned loss of privileges, does not involve the computing of a numeric score at all. There are a host of other measures, such as traditional financial credit scores by the People’s Bank of China, and Internet companies, that are creating credit products aimed at individuals and firms. Now joining the mix are city governments in Jiangsu, Zhejiang, Shandong, and the aforementioned Fujian, through city level personal credit scores as a digital tool for integrating data into governance and delivery of government services.

Xiamen Big Data Group

Over two trips to Xiamen and Fuzhou in the spring of 2019 to study the implementation of these scores, this author, and a group of Yenching scholars, visited and interviewed officials from the Xiamen Guoxin Credit Big Data Innovation Research Institute, guided by the National Development and Reform Council (NDRC), the nodal government body responsible for operationalizing the social credit system nationwide; the Xiamen Information Group which developed and operates Xiamen’s Bailu score; as well as the Fuzhou NDRC Big Data office that developed and operates the Fuzhou Moli score. We also visited sites across the city that recognized these scores and spoke to employees and citizens to learn about their experience using the score.

Scoring 1–1000

bailufen high score-resized(top) Xiamen Bailu score interface on Wechat and (botton) Fuzhou Moli score interface on the eFuzhou app.


 The Xiamen and Fuzhou systems assign citizens a score between 0–1000, with bands that range from poor to outstanding credit, framed ambiguously as a reflection of a citizen’s “adherence to laws, promise keeping, and credit in daily life”. In Fuzhou, citizens receive a base score of 500 points on signing up, with the remaining 500 earned based on their individual record. The scores are voluntary and only citizens with a local hukou or registered long-term residents of Fuzhou’s 6.9 million and Xiamen’s 4.1 million are eligible to sign up.

The scores are computed based on “Big data” i.e. data collected by respective city’s Public Credit Platform (公共信用信息分享平台) that aggregate citizen and firm credit data from various government ministries, public bodies, and state owned enterprises in the city. For example, Fuzhou’s platform collects data from over 630 such entities. The data broadly is classified into includes 4 types, as specified in each city’s credit platform management provisions:
1. basic data: education, employment status, occupation, marriage status, professional qualifications,
2. positive credit: government conferred honors, contributions to public welfare,
3. notices: over-due loans, utility payments (water, electricity, etc),
4. bad credit: legal violations (civil, administrative, criminal).

What makes up ‘positive credit’ is the most subjective, leaving it up to each industry or entity to decide what would be a relevant criteria.

This also means that in these two cities people’s scores are not impacted at all by data from the private sector. So for example people’s online purchases or social media posts, has no impact on the score. Some of the key areas that would boost scores into the outstanding credit includes: a citizen’s timely contribution to the city social security or insurance fund; activities such as volunteering, donating blood, using public transport, separating waste; working in areas of public interest such as teachers or doctors. Using this set of public data cities have designed a model to create the scores.

The scores are computed using models not very dissimilar to existing credit scoring models that have been in use globally. In fact, Xiamen uses the FICO score model — used in the United States by mainstream credit rating agencies to assess financial credit worthiness, but remixed with a different set of variables. Fuzhou uses a slightly different six-variable model, both represented in the figure above.

There is extensive literature on credit scores and the lack of transparency and difficulty for individuals to scrutinize both the scoring and the underlying, sometimes a function of the system or the technology, or both, as documented in legal scholar Frank Pasquale’s book Black Box Society. Both Fuzhou and Xiamen score poorly here. Their models are plagued by very ambiguous terms and descriptions. For example ‘employment strength’ in the Moli score model is described as measurement of how ‘hard-working/conscientious and meticulous’ (爱岗敬业) a person is. Keeping promises’ in Xiamen is described as ‘behaviors that care about others and yield a return to society’. Some are more easier to understand, such as in ‘economic ability’ which is described as ‘delayed payment to common reserve fund defraud/cheat the healthcare fund, common reserve fund, wages, embezzlement’. Quantifying social behavior is at the core of the system and some experts we consulted noted that the algorithms are far from mature or proven. Moreover, with data limited to the public sector, meaning data sets available are limited in depth and diversity, moreover, millions of residents working in the private sector would not be represented to the same degree.

Neither of these models use machine learning based technologies such as predictive scoring. Therefore the specific question of AI blackbox decision making, a highly controversial issue associated with the application of AI across several industries, is not applicable. Both governments claim to be working on bringing in machine learning (ML) and see Alipay’s Sesame Credit, which uses ML in its scoring model, as the the industry benchmark which they are looking to emulate. That being said it does not mean these scores are not blackboxes themselves, outside of the variables and brief descriptions available on the apps there is no further details or explanations available. Officials pointed to customer service helplines they could reach out to in the event of issues with the data.

Score benefits

Jiming Library - ResizedBoth cities use a very different user interface for the scores. In Fuzhou, citizens access their Moli score via an app e-Fuzhou, a multi-purpose e-government services app where residents can access a range of services from getting a driver’s license to using the city metro. In Xiamen, the score is accessed via a dedicated Bailu score public account within WeChat. It is still early days but data so far suggests Fuzhou’s system has had more success in terms of users and engagement.

According to data shared with us, Fuzhou has 1.7 million registered Moli score users (roughly 21% of the city population) and Xiamen has just 210,059 (roughly 5% of city population).

Citizens in the ‘good credit’ region (600–650 and above), which is roughly the average score at the moment, can unlock a range of benefits that can be bucketed into three categories: deposit free access, discounted access, priority access to services.

In Fuzhou, more than 64% of registered users have used their score at least once to avail of a benefit. In Xiamen the number is slightly lower at 53% — which includes 55,562 people that have borrowed books from the public library deposit free. Borrowing books at the public library is popular in Xiamen and according to data from Jiming library, the number of readers who borrow books has grown by 370% since the introduction of the score. In Fuzhou, the Gulou district government services office offers citizens with a score higher than 683 access to three services: priority waiting line, administrative assistance, and express processing. According to the credit office manager at the office, while revealing a bundle of receipts of people that had used their moli score in the past week, the Moli score express line is often half the waiting time compared to the regular line, adding that since they began offering this service in 2018 their work had become harder trying to keep with with the added responsibilities.

How do citizens lose points? and what happens to those with low scores?

All carrots no sticks

In short, citizens can lose points but receive no direct penalty for a low score.

There are a range of things that citizens can do to lose points, all of which involve breaking a law. A document published by Fuzhou NDRC in November 2018 listed nearly 50 such offenses, from 16 ministries, that range from minor (5–10 points lost), moderate (15–20 points), serious (30–50), and extreme (100–150). The top three ministries with the most number of proposed offenses are the Traffic Police, Common Reserve Fund, and Committee for Urban management. Traffic Police and the Common Reserve Fund also have the most serious offenses with an average points deduction of 62 and 70 respectively per offense. A person would only have their Moli score docked once the relevant data about their offense reached the public credit data platform via the government body they are formally charged by.

The combination of incentives and disincentives, aka carrots and sticks, is critical to any system such as these which seeks to bring gamification to governance. The carrots in this case take up added importance as being denied access to carrots become the only punishment when there are no sticks for citizens with low scores. For example, in either city, a citizen would face the same legal penalties associated with their charge, the only difference is now the benefits unlocked by high scores are off the table. Officials we spoke emphasized that there are no plans to have any directly punishments or penalties for a low score, citing the lack of legal backing.

In its present iteration the scores seem more like a government version of a loyalty scheme — all citizens get access to the basic service however some can opt-in for fringe benefits for convenience and comfort. Initial data suggests a very low level of awareness about the score in both cities. Search data from Baidu Zhishu (see graph above) for terms ‘白鹭分’ and ’茉莉分’ between February-May 2019 revealed an average of around 120 searches per week for Moli score and Bailu score even less at 50. Archiving the entire history of posts on Sina Weibo (until May 2019), using a python based data scraping program, that mention either terms, reveals just 53 posts that mention Moli score and just 26 that mention bailu score. Most of them were by local government accounts and media platforms often sharing the same news articles, which were republished across various platforms. Just a handful of posts were made by individual users commenting on the score.

Anecdotal evidence from visits to the city reflect this. Several Fuzhou residents we spoke to that had the score claimed they only did so because they had the eFuzhou app which they used daily to ride the subway, but knew very little about the use of Moli score — despite the fact that those with a score of 686 and above get a 20% discount on each subway ride. In Xiamen, outside of the library, we did not meet anyone that even knew about the score. There was no advertising or government propaganda around the city either. When asked about this, officials in both cities emphasized a word-of-mouth strategy rather than a concerted top-down propaganda effort that tends to accompany major policy efforts, further reflecting the early-stage experimentation nature of the initiative.


The introduction of these city level scores by city governments marks the entry of the government in the business of scoring citizens however implementation so far reveals a very basic attempt with numerous gaps and question marks, but a far cry from the western media picture of a all-encompassing score enabled by mass surveillance.

As legal scholar Xin Dai writes in his seminal paper ‘The Reputation State: China’s Social Credit Project’, the SCS can be seen through a number of lenses including developmental interests, bureaucratic interests, private business interests, and authoritarian interests. The Moli and Bailu system appears to fit within the scope of all five but with varying degrees.

It fits well under China’s development and bureaucratic interests in exploring techno-centric solutions to modernize its governing capabilities and address in particular the problem of enforcement of laws. The SCS has certainly compelled city government to spend upgrading their ICT infrastructure and better incorporate digital tools, which also includes outsourcing to the benefit of a growing number of IT companies that win government tenders. The setup of Public Credit Platforms has enabled a basic level of data sharing across ministries overcoming the long standing ‘data island’ hurdle. It sets the initial infrastructure for more extensive centralization of data within the government for future data driven governance. Naturally this also further entrenches the power of the government and creates a new tool to better track citizen behavior and enforce existing laws, some of which are by design authoritarian. As Jeremy Daum, Editor of China Law Translate, has written, the concern is less the system itself, than the sometimes bad laws it is trying to better enforce. Moreover, the kind of data sharing taking place very much lags behind the sophisticated data collection of several other countries, with China is trying to catch up.

There are many questions about the future direction and viability of the project. For starters, the technology behind this is still unproven. There are no successful approaches to quantify social behavior through algorithms, credit scores in the US for instance, not without its issues, is strictly financial in nature. In China too the most successful credit application so far is Sesame Credit by Fintech firm Ant Financial drawing on its vast resource of consumer and firm financial data.With no data from private sector to call upon, where a large amount of a citizen’s digital footprint is generated, it remains to be seen how successful the models built by city governments can be a proxy of an individuals ‘trustworthiness’, and whether the broader system of rewards built around it lead to citizens becoming more law abiding in any meaningful way. As one representative from a company that build credit scoring applications in Guizhou shared at a conference in Beijing, there are no new algorithms and not useful enough data sets. Going forward quantitative studies on the rule of law comparing Fuzhou and Xiamen to other cities without scores would be a worthwhile study to assess if the cost of implementation is worth it.

So far the system has been developed through a combination of regional administrative rules and regulations, further evolution of the system depends how the central government solves difficult questions like defining what ‘social credit’ or ‘trustworthiness’ is in a legal sense. There is an evolving but nascent public discourse around what it means to be ‘trustworthy’ and how fair, necessary, or useful such a system is. From a citizen’s point of view the lack of transparency around scoring, and processes to amend or update data, are likely the most immediate areas of concern.

Other challenges include protecting citizen interests, data protection, legal liability, not to mention thorny issues such as privacy, and ethics. Tackling this will take a number of years and certainly not by 2020, the final year of the State Council 2014 planning document. While the setting up of the Social Credit System continues to pick up speed, the use of scoring systems for citizens remains very much in the early pilot stage, similar efforts have been tried and scuppered, as with the famous example of Suining in Jiangsu province in 2010. While other cities are following suit including Beijing which is rumored to have its own score too there is much to prove if these pilots are to be scale up further.

with research input from Wesley Harfield, Harel Sholovitz, José Pablo Ceballos, Gao Zihao (Yenching Scholars, Peking University). The post was originally published on the Berkman Klein Center for Internet and Society Medium Collection.

The development and use of credit scores and technology in the Social Credit System is the theme of Digital Asia Hub’s upcoming Podcast-Spoke, which will feature discussions with practitioners involved in developing the system.

Dev Lewis