Using Artificial Intelligence in Economic Development: Challenges and Opportunities

The role of economic development leaders is arguably more complex than ever. From supporting economic growth and sustainability to promoting job creation and labor market transformation, the work is vast. Spending on economic growth can reach trillions of dollars - OECD countries, for example, spend an average of 3.9 percent of their GDP on economic purposes. However, interest rates have risen in many countries around the world, limiting the fiscal flexibility of governments. In addition, the world economy is still reeling from the effects of the COVID-19 pandemic, which crippled global economic output.

At the same time, new forms of data and new ways of processing it are impacting both business and society, including government. For example, generative AI alone could add between $2.6 trillion and $4.4 trillion of value across industries. However, capturing this value will require new ideas and approaches. In this article, we look at some of the major opportunities that AI presents, the challenges that may stand in the way, and how some organizations are addressing them.

AI can transform many sectors and functions, but five areas of economic development can particularly benefit from AI technologies.

Globally competitive value chains. Economic development leaders are often asked to identify areas where their country or region can become globally competitive. To do so, they assess the potential contribution of sectors to sustainable, inclusive and resilient growth and can be the first to spot potential new opportunities. Analytical models that can analyze markets in real time, identify nascent economic trends and identify areas ready for growth give leaders access to powerful new opportunities.

For example, an East Asian city used analytical models to identify untapped areas of economic competitiveness in the beverage and auto parts industries-areas adjacent to those in which it already had significant potential. By focusing on these sectors and capitalizing on existing talent, infrastructure, and supply chains in the respective areas, the city increased its GDP per capita by $8,500 over six years.

Investment attraction and trade support programs. Many economic development leaders are focusing on attracting foreign investment and increasing exports. AI can help leaders select companies that can contribute to local economic development, understand where companies can target their investments, evaluate investment results, and budget for government efforts to attract businesses.

US-based REDI Cincinnati-REDI stands for regional economic development initiative-uses predictive analytics to identify growing companies that are likely to make investments in the future. The model, which combines business intelligence and real-time data analytics, uses data such as M&A activity and earnings reports to proactively target companies for attraction and expansion. To date, REDI Cincinnati has helped attract more than $6 billion in capital investment.

Future of Work programs. Employment agencies can use AI to identify important long-term changes in the labor market and enable critical transitions. Analytical models can identify occupations at risk of being displaced by forces such as automation and global macroeconomic trends. They can also recommend related, more sustainable occupations into which workers can retrain.

The UK Department for Work and Pensions uses workforce analytics to assess the demand for workers in different occupations. This allows job seekers to better navigate the changing labor landscape. It also allows government agencies to offer more training programs in fast-growing occupations such as social care and information and communications technology (ICT).

Economic "forecasting" and prediction. Finance and economic ministries and central banks will no longer have to identify a crisis months after it begins. Instead, policymakers can use artificial intelligence technologies to detect early signs of shocks, allowing them to make faster course corrections and steer the economy through different economic cycles.

For example, the OECD forecasts weekly GDP growth using data from 46 countries across different economic sectors. The model uses machine learning to identify correlations between the frequency of searches for terms such as "unemployment," "investment," "crisis," and "recession" and changes in various components of GDP. By providing real-time indicators of economic activity, the OECD Weekly Tracker makes it easy to assess rapidly changing data, such as in the event of an economic crisis.

Transforming public services with geographic information systems and spatialdata.Government organizations are increasingly using spatial data and satellite imagery to improve the efficiency of public service delivery, enhance disaster response, and promote smarter, more resilient and future-ready cities. Given the size and granularity of these datasets, sophisticated artificial intelligence models may be required to derive near real-time metrics.

In Kazakhstan, a statistical model that combines geographic, demographic and economic data with analytical methods is used to identify and develop infrastructure in rural settlements. The model analyzed more than 6,293 settlements, from which 3,500 settlements with the highest development potential and home to 90% of the rural population were selected. As a result, government leaders will be able to more efficiently and accurately deliver needed services and infrastructure to rural areas.

When implementing artificial intelligence, economic development leaders should focus on three key challenges: protecting the right data, attracting talent, and gaining public trust.

Data

Lack of data can make it difficult for economic development leaders to accurately forecast macroeconomic trends, compare the impact of investments in different regions, and prepare the workforce for a changing labor market. In addition, there are gaps between data-rich and data-poor countries in terms of data quality, availability and cost. For example, less than half of all births are registered in sub-Saharan Africa, the last census in Afghanistan was conducted in 1979, and about one billion people worldwide do not have an official identity card.

In a data-rich environment, it can be difficult to distinguish signal from noise. Analytical models are only as good as the data on which they are based, and working with unintegrated or disparate data can delay projects and increase costs. Low-quality data can be incompatible with the use of AI-driven analytics. According to one global study, 45% of developers agreed that government data was clean and accurate, meaning more than half of developers thought they were working with inconsistent or inaccurate data. Less than 35% thought the data was well documented.

Access to data is another challenge cited by experts. Analytical models typically require large data sets for accurate forecasting, but bureaucratic silos, divergent political agendas, and restrictive regulations can prevent organizations from sharing relevant data.

Talent

Public sector organizations are having difficulty attracting talent. In the UK, 51% of public sector organizations report having difficulty filling vacancies, compared to 38% of private sector companies. It's even harder to recruit young, tech-savvy employees. For example, in the U.S. government, for every employee aged 30 or younger, there are more than four IT professionals aged 60 or older.

Government leaders we spoke with emphasized that competition for talent with the private sector is further limiting their ability to hire technical talent. For example, there are now as many job openings requiring Python as there are professionals proficient in that coding language, and for every job opening requiring data skills, there are 0.9 data analysts. One result of this is an increase in salary expectations that many government organizations are having difficulty meeting.

Trust

Government organizations seeking transparency in decision-making may find it difficult to use tools that cannot be interrogated for explanations or accountability. Modern machine learning methods are sometimes said to operate like a "black box," producing results that cannot be easily explained as a linear relationship between variables. This can undermine confidence in the model's results, and also explains why many people are wary of AI. An Ipsos survey of about 22,800 adults in 31 countries found that about half of respondents said they felt nervous about products and services that utilize AI.

One of the potential benefits of using machine learning techniques to analyze large amounts of data is that they allow for counterintuitive insights that a human hypothesis-driven approach is not capable of. For example, an algorithm might determine that a country has a large unrealized opportunity to produce water pumps. But it will be difficult to justify that idea to investors and taxpayers if government leaders themselves cannot understand how the algorithm arrived at that conclusion. Some of those we spoke with also doubted that AI could understand the cultural context in which economic development decisions are made.

Several public sector officials told us that their organizations launch AI pilot projects and then see progress slow down. The lack of momentum causes frustration, leading to reduced resources and further delays. For example, official estimates suggest that in the European Union, only 38% of public sector AI use cases have reached the implementation stage, with the majority still in development or at the pilot stage.

Government organizations can take advantage of the following seven strategies to help them accelerate the adoption of AI technologies.

1. using artificial intelligence and other advanced technologies to overcome complexity

New technologies are complex, and complexity can be intimidating. But combining artificial intelligence and other tools with dashboards can help decision makers focus on the most important priorities and metrics. Models can also be used to effectively synthesize information.

2. Ensure that the roadmap of use cases is based on basic needs

Leaders of government organizations may prioritize quick results when they start working with AI. Given the importance of maintaining employee enthusiasm throughout the development of new technical capabilities, this makes sense. However, organizations still need to create value with AI tools. If you start with the end in mind, you can understand the steps executives need to take to achieve their goals, allowing organizations to avoid investing resources in unnecessary models and functions.

3. Ensure seamless access to data through strong statistical management, use of private vendor data and better management of data sharing

Many leaders noted that statistical agencies can play an important role in making data accessible. A good statistical office can use its independence and technical capacity to publish reliable data that can be used to build robust models. Operating in accordance with local privacy and security regulations, such an office can also work with government organizations to collect useful data for decision-making.

Ensure correct data sources. If government data is not available, other published data sources can be used - also in accordance with regulatory requirements. Consider the following examples:

• Geolocation and Satellite Data. The McKinsey Global Institute (MGI) report,"Pixels of Progress: A Granular View of Human Development Around the World," breaks the world into 40,000 microregions using nighttime satellite imagery and other data. MGI found that a country's GDP growth explains only about 20 percent of the growth in a given micro-region and highlighted success stories that may have gone unnoticed, such as Mapusa, India, where GDP per capita has tripled over the past 20 years. Pixels of progress: A detailed look at human development around the world
• Telecommunications data. The World Bank used data on mobile services, including call duration, network of contacts, and recharge frequency, to identify the poorest households in Afghanistan. This approach worked about as well as more expensive data collection methods, such as field visits to count electrical appliances.
• Payment data. Credit and debit card transactions are a particularly interesting combination of high-frequency and geospatial data. The UK regularly publishes changes in credit and debit card spending in various categories such as basic consumer goods and discretionary spending. This dataset is part of a wider set of indicators that began to be collected during the COVID-19 pandemic to monitor the economic and social impact of the pandemic and the pace of recovery. It includes data on weekly Pret A Manger store transactions, company formation and liquidation, layoffs, airline flights, supermarket prices, traffic, store inventories, and value-added tax transactions.

Make data available to the entire government through robust data management.By ensuring that datasets are accessible and interoperable, and that a data management system allows organizations within a country's government to share data, governments can play a critical role in building the capacity of teams developing analytical models. This is important because models may require combining multiple datasets belonging to different organizations.

For example, the U.S. government's open data site, Data.gov, offers access to nearly 300,000 datasets; the site's stated purpose is to "inform public and policymakers' decisions, spur innovation and economic activity, fulfill agency missions, and strengthen the foundations of open and transparent government."

4. Realize that trust is earned slowly and lost quickly

Managers should consider using AI models and data analytics to demonstrate that these tools are critical to improving worker skills and fostering collaboration, rather than being a means of substituting for human input.

Use artificial intelligence to improve the performance of human experts, not replace them.This can be especially important in economic development, where decisions are often not black and white. For example, deciding how much subsidy the government should offer to attract a new electric vehicle (EV) gigafactory is not necessarily about following a set of algorithmic rules. Executives need to understand what AI models can and cannot contribute to the government's decision-making process.

In the hypothetical example of a new gigafactory EV, an artificial intelligence tool can estimate the number of jobs created by the factory, its impact on GDP and its contribution to exports. However, analytical models cannot explain what the role of governments might be in determining industrial policy or how to value personal mobility over public transportation.

The Swiss Anticipate model illustrates the potential for collaboration between experts and AI. The app collects, classifies and analyzes data from a variety of sources, including field networks and news articles. It uses expert predictions to refine its predictions, giving more weight to experts with a strong track record.

Make sure the results can be explained. Decision makers may want to consider investing in educating others about the benefits and limitations of AI models. For example, the Scottish Government and Mind Foundry are helping non-technical users of AI to understand how different factors can affect the results of a model. The goal is to encourage experts to collaborate with AI tools rather than using AI to replace human judgment.

Ensure transparent data policies (e.g., data privacy) and data bias controls to address the concerns of senior government leaders.The UK has developed a standard for recording algorithm transparency that defines how public bodies can publish information about their algorithms, including the rationale for their use, human oversight mechanisms, technical specifications, potential risks and mitigation measures, and impact assessments. This can reduce the risk of misuse of algorithms by government organizations, promote best practices, and increase confidence in the use of these tools.

Develop solutions in-house or partner with domestic organizations.Because some models require access to sensitive data, governments are understandably hesitant to share it with outside organizations. Leaders sometimes fear that their data and innovations may be leaked by foreign companies that have fewer assets and people on the ground who can be held accountable. They may also be skeptical of the ability of foreign experts to create tools tailored to local values and characteristics. Our interviews with economic development officials suggest that one way to build trust may be to partner with local organizations.

Abu Dhabi-based artificial intelligence company G42 formed a coalition of Abu Dhabi firms to develop Jais, a model of the Arabic language. The group included government agencies, academic institutions, state-owned companies and local banks - all sharing the data used to create the model.

5. Establish partnerships, mentoring, rotation programs, and engage external providers to build and expand expertise.

Organizations can implement mentoring programs, connect with external service providers to access macroeconomic models, and partner with other agencies to accelerate positive outcomes.

Partnerships. Most of the organizations we studied have modest AI and analytics teams. Partnerships can play an important role in delivering tremendous value with limited resources. For example, the University of Oxford, Vivid Economics (a McKinsey company), and the UN Economic Commission for Africa collaborated to identify green economy opportunities in the Democratic Republic of Congo. They found that investments in municipal renewable energy, transmission lines, mini- and microgrids, and climate adaptation could be particularly profitable.

Rotation programs. Rotation programs can help bring talented data scientists from private companies to government agencies for a few months where they can coach junior team members. The U.S. federal government, for example, allows IT and cybersecurity workers to temporarily work for other agencies. This allows experienced professionals to build the necessary technical capabilities in less developed agencies.

Mentoring programs. Mentoring programs can also enable experienced economic development leaders to build the capacity of people in new organizations. Some UK agencies mentor analytics organizations in developing countries, helping under-resourced teams learn new coding languages and adopt best practices for project scoping, quality assurance, project management, etc.

External providers. In particular, under-resourced teams can focus on providing specialized information to decision-makers rather than reinventing the wheel. Macroeconomic models and other tools can be purchased and customized rather than built from scratch.

6. Pooling of resources through the establishment of Centers of Excellence (COEs)

Centers of excellence focused on specific areas of expertise are being used by some organizations and governments to create digital platforms and other tools to reduce duplication and simplify access to data.

CoEs can help attract talent and implement large-scale projects.Governments can create CoEs to mentor, nurture and direct talent to the most relevant programs. They can also use CoEs to create common platforms to improve e-services and help overcome silos in government. For example, the U.K. Government Digital Service, made up of 750 product managers, software engineers and others, has created digital platforms that are now used by more than 1,900 public sector organizations.

Centralizing tools and processes will help organizations save time and money.Managers should consider which AI tools can be shared rather than duplicated. For example, a government's Ministry of Economy and Ministry of Finance may need access to macroeconomic forecasts, but that doesn't mean creating two AI-based platforms. Eliminating redundancy can yield significant savings. An OECD study found that centralizing government procurement in one country can significantly reduce costs.

Managers should consider where AI tools can be shared rather than duplicated..... Eliminating redundancy can yield significant savings.

7. Create a strong value proposition to attract, retain and develop diverse talent

Compensation isn't the only lever organizations can use to attract top talent, as flexibility is becoming increasingly important to potential employees. And those that place more emphasis on skills-based hiring can expand the talent pool.

Take compensation into account, but don't limit yourself to it. Governments spend an estimated $209 billion dollars a year on IT services. In recent decades, many governments have outsourced their digital and artificial technologies, in part because of talent acquisition challenges. Compensation is a contributing factor, as public sector pay scales may not keep pace with market demand for data analysts and other professionals. However, compensation is just one of several levers that public organizations can use to attract and retain talent.

A McKinsey survey of U.S. government employees found that employees who want to stay with their organization value meaningful work and workplace flexibility. At the same time, those who want to leave say they are discouraged by limited opportunities for career development and leadership positions.

Consider alternative talent pools. AI and analytics teams can expand faster if they hire a diverse workforce, including legal migrants and qualified candidates without university degrees.

Create a team with a diverse profile. Successful teams can include economists, subject matter experts, project managers, communications specialists, and data scientists.

After years of dealing with one crisis after another, including the COVID-19 pandemic and rising inflation, economic development leaders are increasingly relying on AI and other advanced technologies to attract investment, retrain workers, and support economic growth. Recent shocks have underscored both the benefits of data-driven decision-making and the challenges leaders may face in obtaining timely data. Building an organization's AI capabilities requires new leadership skills, as well as grit and determination beyond simply responding to crises in the moment. It may not be about big, bold actions with immediate consequences, but rather sequential, small steps whose cumulative effect may not be fully understood until years later. A commitment to AI-enabled economic development may not make headlines, but it could prevent new crises.