Training Entrepreneurs: Issue 4

Business training programmes typically aim to deliver a fixed curriculum of content to a group of firms, usually in a classroom setting. This has advantages for scalability, but may limit how adaptive the content is to specific needs of individual businesses, and to changes occurring in the economy. Mentoring and peer interaction approaches have been developed to attempt to better share customised knowledge amongst firms. A second issue is that delivering training to groups in classrooms is still costly, and may not be convenient for many entrepreneurs to attend. Moreover, the COVID-19 pandemic restricted gathering groups of firm owners together physically in one room, increasing attention on alternative delivery mechanisms for teaching training content. More recently, the rise of AI has led to hopes of scalable, cost-effective, and individually tailored content being possible, although so far experience is limited.

Peer interactions and mentoring

Having firm owners learn from one another offers the promise of providing a way for sharing locally-relevant, better business and management practices, and may also offer other benefits for firms in terms of establishing new business relationships. This can involve the use of formal mentors, as well as programmes that match firms with other peers to facilitate interactions. Berelowitz et al. (2020) offers practical advice for implementing mentoring, although the evidence base underlying these recommendations is still limited.

In practice, mentors paired with individual firms to give customised advice and feedback can operate like a form of individualised consulting. Brooks et al. (2018), for example, conducted an experiment with 372 female-owned microenterprises in Kenya. In addition to a control group, a first treatment group is randomly allocated to receive classroom training, while a second treatment group is assigned to mentors. The mentors were selected from the more profitable business owners amongst their sample, and paid a nominal payment of US$ 10. The mentor-mentee pairs were required / encouraged to meet weekly at the mentor’s place of business four times over a month, though many pairs continued to meet for more than a year beyond the official treatment period. The study found a substantial short-term effect of mentoring, with profits of treated enterprises increasing by 20%, whereas the classroom training shows no significant effect. The effects, however, disappear about a year after the treatment begins. The mentor treatment appears to be effective while the mentor-mentee relationship is most active and disappears as the incentives provided to the mentors are removed.  McKenzie and Puerto (2021) fail to find any positive effects of mentoring in a sample of female microenterprise owners who also received the ILO GET Ahead business training programme. They had more successful business women meet over a five-month period both in small groups and one-on-one with participants after the in-class training. They estimate that the mentoring cost US$ 553 per firm mentored, and find that it did not lead to any additional improvements relative to the training alone. Bakhtiar et al. (2022) have women who have received business training then act as mentors for other women in their networks, at a cost of around US$ 500 per mentee firm. They find this does improve business practices in the mentee firms, but the treatment effect on profits is statistically insignificant, and small (US$ 5 per month) relative to the cost. Lang and Seither (2022) show that being assigned more intensive mentoring can actually leave poor women worse off than a light-touch opt-in mentoring. Taken together, this literature suggests mentors are not always of lasting benefit to subsistence firms.

Mentors may be more effective for more advanced firms looking to innovate or expand into new markets, where their local knowledge network may not be able to provide sufficient expertise. An example is provided by Anderson et al. (2022), who conduct an experiment with 930 small business owners in Uganda. The treatment group was linked to mentors around the world via biweekly Skype meetings for up to six months. The mentors were typically management professionals in advanced markets, who tailored the activities towards the specific context and challenges facing each firm. They find that firms assigned this remote mentor did not improve overall business practices, but were more likely to “pivot” their marketing strategy, for example, by shifting the production of sale of one product line to another. As a result, sales increased by 28% over two years. This growth was highest when the firms were linked to mentors from a marketing background, with monthly sales growing 52% and monthly profits 36% (Anderson et al. 2021). Germann et al. (2023) show that the female entrepreneurs in this study fared better when they were matched with female mentors, whereas male entrepreneurs did equally well when matched to either a male or female mentor.

Peer interactions

Firms may also be able to improve their business and management practices through learning from each other. Cai and Szeidl (2018) illustrate the promise of such an approach by conducting an experiment with 2,820 firm owners in China. The firms are all SMEs, with an average of 36 employees, established within three years of the beginning of the experiment in 2013. 1,500 firm owners were randomly selected for treatment, which involved meeting monthly for 10 months with nine other firm owners. The groups were of four types: i) small size, same sector; ii) large size, same sector; iii) mixed size, same sector; and  iv) mixed size, mixed sector. The relatively large sample size and segmentation allows for a nuanced analysis of peer effects. Cai and Szeidl found that sales of treatment firms increased by 8-10% relative to the control firms, with comparable increases in material inputs, employment, and assets. Why did the interactions lead to an increase in firm growth? Cai and Szeidl show evidence on several channels. First, there is evidence that firms shared information on trading partners, with the number of referrals to trading partners and the number of direct relationships between firms in the group both significantly higher with treatment. Second, at the end of the year, they showed that firms in the treatment group had significantly better management practice scores than those in the control group. Indeed, the improvement in management practices generated by the peer-interactions is comparable to that generated by the US$ 250,000 consulting intervention carried out in India, discussed above. Third, firms randomised into groups with higher-quality peers (measured by baseline firm size) showed larger increases in sales, profits, and management practices than those randomised into groups with weaker peers. Cai and Szeidl carried out an additional experiment by providing selected members of each group information about either an individual savings product or a business grant programme. The results of this additional experiment provide important lessons for scaling up the experiment. Specifically, information on the savings product flows through all of the groups and all of the group members. But information on the grants flows only when the members are not direct competitors.

Asiedu et al. (2023) test whether virtual networking with peers can also help firms. They conduct an experiment with 1,772 female growth-oriented entrepreneurs in Ghana, where treated firms are assigned into WhatsApp groups of eight members, and scheduled to meet virtually with another group member each week with the aim of expanding business networks. They find no significant impact on sales, but that treated firms are earning 21% higher profits after a year. Agarwal et al. (2025) formed virtual discussion groups of 4-6 microentrepreneurs in an experiment in Liberia. The groups had weekly facilitated discussions on business topics for six weeks. They find treated firms adopt more innovations and improve business practices in the next three months, with slightly negative and not statistically significant impacts on profits.

The quality of peers is likely to matter a lot for these interventions. Chatterji et al. (2019) carried out a bootcamp with 100 high-growth technology start-ups in India, where firms were randomised into pairs. They find that entrepreneurs who received advice from peers with a formal approach to managing people — instituting regular meetings, setting goals consistently, and providing frequent feedback to employees — grew 28% larger and were 10 percentage points less likely to fail than those who got advice from peers with an informal approach to managing people in the two years after the intervention. Asiedu et al. (2023) report impacts are higher when women are matched with more college-educated peers with better business practices and higher profits and sales.

Peer interactions therefore seem most effective when firms get matched with similar, but slightly better peers who are not close competitors. This raises a concern for the general effectiveness of such programmes, since by definition, every firm matched to a firm that is better managed also has a counterpart firm that is matched with a firm that is worse managed. Moreover, peer learning may not happen automatically, and training may be needed to help firms learn how to better communicate with one another. Dimitriadis and Koning (2020) conducted an experiment in Togo, in which entrepreneurs were given a two-hour communication training to help them better interact with peers, finding that this led to more information being exchanged and short-term performance gains.

Group-based networking interventions may also benefit from being bundled together with some group-consulting and financial assistance. Münch et al (2024) create consortia of firms in Tunisia that get both group-consulting and a group-level subsidy, and find that this helps the firms to increase export sales. However, since it is a package intervention, it is unclear how much is due to the networking, consulting, and financing components, and whether these components reinforce the effectiveness of each other.

Learning from other firms without peer interactions

There are a lot of organisational logistics involved in deciding which firms should be linked to one another, and getting them to meet and exchange information. An alternative is to try to help firms learn from their peers without having to physically interact. Dalton et al. (2020) provide one approach, where they ​conducted qualitative interviews with local firms in Jakarta to understand which business practices are being used, misconceptions about different practices, and implementation norms. They then used this to put together a handbook of best practices, and a documentary in which successful peers explain how they have adopted practices and their growth trajectory. This was coupled with two half-hour visits from a trained enumerator to help in implementing the practices, with the result being that both business practices and firm profits improved over the next 18 months. A second, and even more basic, approach to helping firms learn from peers is to allow them to benchmark themselves against how others are doing. Seither (2019) finds that merely providing firms in Mozambique with data on how their sales compare to other firms in the same sector, leads low-performing firm owners to work more hours, and increase profits and sales over the next year.

Alternative delivery methods

A range of technologies offer the potential to help business owners improve their business practices without having to go to in-person training. However, there is relatively little evidence on the effectiveness of these methods, and the available evidence suggests that the impacts of some forms of remote training have been quite limited.

Entrepreneurial edutainment

Television shows such as Dragons Den, Shark Tank, and The Profit illustrate how the process of pitching a new product or improving a struggling business can be entertaining to millions. Can watching such shows also teach entrepreneurial skills or inspire entrepreneurial attitudes? Two “edutainment” shows for entrepreneurship have recently been evaluated: Ruka Juu (“Jump Up”) in Tanzania (Bjorvatn et al. 2020) and El Mashroua (“The Project”) in Egypt (Barsoum et al. 2018). Both were reality show competitions, with weekly episodes over a span of 11-13 weeks, that followed the journeys of young entrepreneurs as they undertook challenges teaching and testing entrepreneurial skills. Key business concepts such as market assessments, planning, advertising, record-keeping, etc., were emphasised in each episode. 

Randomised encouragement designs, in which a treatment group gets invited and reminded to watch the show, and a control group gets invited to watch something else or does not receive a reminder, have been used to measure the impacts. Bjorvatn et al. (2020) used this approach with a sample of 2,132 secondary school students, and Barsoum et al. (2018) with a sample of 5,924 Egyptian youth. The findings indicate that these edutainment shows do seem to make viewers slightly more interested in entrepreneurship, and seeing women succeed makes viewers think it is a little easier for women to go into self-employment than they had originally thought. However, neither study finds any impact on business knowledge, or on people taking actions towards starting businesses. However, it may be that these shows have effects on a minority of viewers that cannot be measured in standard impact evaluations. For example, if only 0.25% of the 4 million Egyptian viewers start a business as a result of watching, that would still amount to 10,000 new businesses created, but would need an experiment with almost 250,000 individuals to detect an impact.

SMS messages and voice messages

SMS messages have been used to send reminders and nudges to get people to save. They could potentially be used to disseminate simple business practices and business information, as well as to offer feedback based on automated rules. Several trials have shown limited impacts of this approach. Cole et al. (2019) tested sending weekly voice messages with rule-of-thumb in the Philippines and India, finding modest improvements in business practices but no significant changes in business performance. Acimovic et al. (2022) worked with mobile money agents in Tanzania, and experimented with sending daily personalised recommendations on inventory levels, finding no impact. Mehmood (2023) tests an SMS-based business training course in Kenya, and finds short-term increases in knowledge and adoption of business practices that then fade out by 12 months, with insignificant impacts on sales and profits. Like television, SMS can be a very cheap way of helping firms, and so the magnitude of changes needed for this approach to satisfy cost-benefit calculations may be much smaller than existing studies can detect, but the benefits appear to be at best modest.

Impacts may be slightly stronger when targeted to specific actions at the right point in time for firm owners to make changes. Ronconi (2025) conducts an experiment with 14,500 online retailers in Brazil and Argentina, sending them messages reminding them that Black Friday is approaching and reminding them to plan their advertising and pricing strategies. He finds this increases sales 5.2% in the 60 days following the intervention.

Online training and consulting

Both edutainment and SMS are limited in the amount of detail they can provide compared to standard classroom training. In contrast, developing fully online training modules offers the possibility of covering at least as much content, having firm owners do interactive exercises, and teaching a wide range of business skills and mind-sets. Online delivery may also enable a much broader geographic reach of such programmes, and potentially lower costs compared to in-person training. COVID-19 made such approaches particularly attractive given the restrictions placed on in-person gathering and an emerging literature has started to test the effectiveness of these programmes.

There are several modalities emerging of the way to provide this online training. One method holds live classes or consulting sessions with an instructor via Skype, Zoom, or some other video conferencing service. Davies et al. (2023) conducted an experiment with over 2,200 female microentrepreneurs recruited from throughout Mexico and Guatemala, with the treated group offered nine two-hour sessions taught live in small groups over Zoom. They find it is now feasible to offer such training even to small-scale firms, and training attendance rates were similar or higher than for previous in-person sessions. However, the cost savings relative to in-person training is modest ($50 vs $62) due to small groups requiring considerable instructor time. Training is found to significantly improve business practices and sales (by $200 or 23%) in the first two months, with a positive but statistically insignificant impact on profits ($45 or 13%). However, by six months these impacts were close to zero and no longer statistically significant. 

Live online training and consulting can be more expensive, but has so far shown stronger results when delivered to more growth-oriented SMEs. The Anderson et al. (2022) experiment in Uganda, discussed previously in Section IVA, was implemented at distance  via Skype. Cusolito et al. (2023) conducted an experiment with 225 firms drawn from across six countries in the Western Balkans, with treated firms receiving 30 hours of live group-based training sessions and five hours of one-on-one virtual consulting from Deloitte consultants. The firms had a mean (median) of 17.6 (8) workers and were looking to expand sales into export markets. They find treated firms improve their digital presence through using tools like search engine optimisation, enabling them to attract more customers, resulting in an increase in export sales for firms that were exporting. Training and consulting using a top consulting company had a marginal cost of $2,140 per firm, with point estimates suggesting firms could earn this back within six months to one year, but with considerable uncertainty.

These live classes and one-on-one sessions have succeeded in expanding the geographic reach of training, but do not dramatically lower the costs and may be difficult to scale to many thousands of firms. An alternative is to use interactive self-paced online assignments that do not rely on a live instructor. A big concern with self-paced online training is that many massive open online courses (MOOCs) have had very high drop-out rates. Cassidy et al. (2025) compare the same training in Ethiopia randomised to be offered via a digital app versus in-person. While 80% of women entrepreneurs start the app-based training, only 22% complete it, far lower than the 71% completing in-person training. One approach that has been used with business training is to work within a supply chain and incentivise take-up with coupons or discounts. Jin and Sun (2022) conducted an experiment with over 700,000 new sellers on a Chinese e-commerce platform, in which one quarter are offered task-based training modules focused on setting up a website, marketing, and customer service. An AI algorithm assigns these training materials based on real-time operating data. Despite offering incentives in the form of additional platform services, take-up is much lower than in-person classes, with only 24% of firms starting a task, and only 12.6% completing at least one task. They find that revenue increases 1.7% for firms assigned to training, and 6.6% for firms taking up training, but since this gain comes from spending more on marketing and promotions, it is unclear how profitable this was for firms. Larger impacts are found in an experiment by Estefan et al. (2023) with 498 chicken franchise store owners in Guatemala. The franchiser offered treated firms 28 video capsules (between one and seven minutes) that provided a mix of traditional business training and rules of thumb, combined with three 1.5 hour virtual one-on-one sessions. They find a 6-12.7% increase in sales over the next year, and a 16-22% increase in profits over a six-month horizon.

Another context in which individuals may be more likely to complete self-paced entrepreneurship training is when it is offered through schools to students. Asanov et al. (2023) deliver online self-paced training to over 45,000 high schools in Ecuador during the COVID-19 pandemic, and find the average student completes over 29 hours, or 24 of 27 sessions on this platform, with centralised management by the Ministry of Education boosting take-up. La Fortune et al. (2022) offer gamified business challenges over a six-week period during COVID-19 to high school students in Rwanda, finding the treated group completed 60% of challenges. Although very high attrition (50%) suggests caution in interpreting results, a short-term follow-up survey did find this increased the likelihood students owned a business a month after training.

The research discussed above shows both the promise and some of the challenges of delivering training online. The recency of most of this evidence, along with the context of delivery during the COVID-19 pandemic for some of the studies, means that the sustainability of impacts is still largely unknown for many of these training programmes. Maintaining quality and take-up in a way that can be scaled to a large number of firms outside of supply chains and schools remains an open challenge. As well as the digital delivery of training, another open question is whether more of the training should focus on teaching firms digital skills. Digital marketing skills have offered promising returns in some studies, but it is less clear whether small firms benefit from digital accounting, digital inventory control and production tracing, etc. Another type of digital skills training aims to help entrepreneurs do online freelancing, which offers the promise of helping skilled entrepreneurs access global customers. Fazio et al. (2025) test a 12-week training programme in El Salvador. They find training completion rates are low, with only 39% completing the first phase, and 16% their full programme. Trained individuals have a 6-7 percentage point increase in getting an online contract, but there is no lasting impact over a year on online entrepreneurial freelancing, which the authors suggest may be in part due to low ratings on the initial contracts. Future work may need to both do more on screening the right types of entrepreneurs for such programmes, as well as helping them succeed once they have their first contract.

Generative AI as business trainer or consultant

The rapid advancement in generative artificial intelligence (AI) has raised hopes that business training and consulting may be able to be made interactive and customised, and delivered at low-cost and scale. A first test of this potential was undertaken by Otis et al. (2024) with a sample of 640 Kenyan small businesses in 2023. They wrote prompts for a GPT-4 ‘business assistant’ and developed a WhatsApp interface for business owners to ask questions which the AI would then provide structured responses to. They then measure impacts over the subsequent two months, finding a small and statistically insignificant average impact on firm performance. They highlight heterogeneity in responses, with firms with above median performance at baseline seeing gains of 15% in an overall performance index, relative to an 8% worsening for below-median firms. The short follow-up period means that it is not possible to know whether firms would continue to use this technology over time, and whether they would get better at learning how to use it with experience.

Further efforts are underway to better train and operationalise generative AI for small businesses in developing countries, such as MYPE Asesor IA, along with rapid improvements in the underlying AI models. One area for further exploration is the extent to which generative AI will be used as an assistant or resource, relying on business owners to ask questions, versus the extent to which it can be used like a proactive consultant, programmed to ask a set of questions, provide a diagnosis, and follow through on implementation of improvements.

In addition to using AI to provide the training or consulting, firms may also benefit from training on how to better use AI tools in their businesses. Pereira-Lopez et al. (2025) provide preliminary results from an experiment in the Czech Republic with firms that average 10-12 workers. They hold group training workshops in which firms are offered experiential learning on how to use AI tools such as Chat-GPT, and AI features in other business software like Canva Pro and Pipeline, for marketing, streamlining business workflows, product development, and financial management. A midline survey conducted nine months after the workshops finds treated firms have a 26% increase in the intensity of AI use in their firms, and monthly profits increase by around €500, meaning that the cost of the course would be earned back in one month.

Some additional cross-cutting research gaps

The above shows that there is now an accumulating body of evidence on the effectiveness of different types of skill-building support for entrepreneurs. This has been informative in demonstrating the economic returns to better business and management practices, and in helping provide some justification for direct government support to firms (McKenzie 2025). However, there remain many questions about how to best implement these programmes that are often asked by policymakers, but for which evidence is currently limited. We note some of these open areas with the hope that future editions will be able to provide more answers.

A first, general topic that cuts across all of these different types of programmes, is how to best select the firms to support. Often policymakers have multiple, sometimes conflicting, goals, and so may want to generate firm growth, jobs, and productivity improvements, but also help underrepresented groups, support the development of priority industries, and/or also try to address climate goals. There is a concern that trying to do everything with one policy instrument makes it less effective at meeting any of these goals. Nevertheless, even when we have a clear definition of what the desired outcome is, there is limited evidence on what selection mechanisms work best to achieve these outcomes.

One approach is to conduct multiple rounds of screening, to try to identify the firms that are more skilled to begin with, and more motivated to grow. For example, Anderson et al. (2018) approach approximately 10,000 businesses in South Africa, screen them on interest, education, formal status, aspirations, motivation, and attendance in a registration session. Their study has some of the largest point estimates on profits and sales of any in Figures 1 and 2, but it is unclear how much all this screening contributed to these results, since comparable impacts for less-screened firms are not available.

A contrasting view is that poorly managed firms have more scope to improve, but may not appreciate how much they stand to gain, and so may be less interested in training. Karlan and Valdivia (2011) find more improvement in business practices, and suggestive evidence for greater improvement in business results, for firms that were initially less interested in training. Price offers another way of screening firms based on their expected benefit from training, although it suffers from this concern about screening out those who underestimate their gains. Maffioli et al. (2023) randomised the price charged for training and find that charging a higher price results in firms who exert more effort in attending training sessions; however, authors lack power to detect whether they also have larger benefits.

Looking at heterogeneity in treatment outcomes within a study in principle offers the ability to learn whether certain types of firms benefit more (e.g. Campos et al. (2018) look to see whether personal initiative works better for women with higher or lower initial human capital, and find no significant variation), but this type of analysis is constrained by statistical power. Modern machine learning methods offer the potential to examine ex-post heterogeneity in impacts based on many more firm characteristics (e.g. Martínez et al. (2025) find heterogeneity in returns to individual consulting), but it is less clear how well these methods can then perform in selecting future cohorts. More work is needed that explicitly tests how treatment outcomes vary under different selection mechanisms.

A second topic that we have not covered much in this review is whether training or consulting is enough by itself, or whether it needs to be bundled with financial support for firms to truly benefit. A first research and policy question here is how the addition of financing changes the pool of firms willing to participate. It may draw in more applicants and encourage them to exert more effort in training if they feel that both their human and physical capital constraints are being alleviated. However, there is a countervailing concern that firms that are only interested in the capital may apply and see attending training as a tax they have to pay in order to get the money, with less or no benefit to them of training.

The ideal design here would not just compare training to training plus cash or a loan, but also have a treatment arm that offers just the cash or loan without training. Otherwise, as in de Mel et al. (2014), if we find that women offered training and a cash grant have higher profit growth than women just offered training, we do not know if this is because cash enables firms to make better use of the training (and vice versa), or whether it is just that cash helps firms grow irrespective of training. Likewise, Fiala (2018) tests cash against cash plus training, and loans against loans plus training, but has no ‘training alone’ treatment arm to compare to. Detecting this complementarity or interaction between treatments requires even larger sample sizes than are needed to detect the impacts of a single treatment, and the existing literature therefore does not provide rigorous evidence for whether such complementarity is important.

For full reference list see the end of the conclusion chapter.