The Future of Credit Union Services in Ecuador Through Hub C’s Algorithm-Based Scoring Model Pilot

2025-04-08

In Ecuador, where more than 5 million individuals and small businesses lack access to traditional financial services, algorithm-based scoring models are becoming increasingly important for credit unions. These models use advanced algorithms and a wider range of data to assess creditworthiness, offering a more accurate and inclusive approach to lending. This is crucial in a country like Ecuador, where a significant portion of the population, especially youth, women, migrants and rural families, is underbanked or lacks an established credit history.

In this featured blog, we’ll explore why algorithm-based scoring models are a strategic move for credit unions in Ecuador and how they can help improve financial inclusion, reduce risk, and enhance member satisfaction.

What is an Algorithm-Based Scoring Model?

An algorithm-based scoring model is a credit assessment system that relies on data-driven algorithms to evaluate a borrower’s creditworthiness. Unlike traditional credit scoring models, which primarily use credit history (like FICO scores in the U.S.), these models analyze a wider variety of data sources, such as transaction history, payment behavior, utility bills, income patterns, and even social media activity.

Why is It Important for Credit Unions in Ecuador?

1. Improving Financial Inclusion

   In Ecuador, a large percentage of the population is either unbanked or underbanked, meaning they lack access to traditional banking services or have limited credit histories. This creates significant barriers for individuals and small businesses that may need loans for personal or professional growth but struggle to qualify under traditional credit scoring systems. This enables credit unions in Ecuador to offer loans to a larger and more diverse segment of the population, empowering individuals and small businesses to access the capital they need to thrive. It also opens the door to a growth segment for credit unions, rather than going head to head with banks, and helps to bridge the gap for those excluded from the traditional financial system.

2. Better Risk Management

   Credit unions face the challenge of assessing credit risk accurately, especially in regions where many potential borrowers have little or no credit history. Traditional credit scoring models may not fully capture the financial behavior of these individuals, leading to higher default rates or missed opportunities for responsible lending. By leveraging algorithm-based scoring models, credit unions can analyze a wider range of data and gain a more accurate picture of a borrower’s ability to repay a loan. For example, transaction data, income history, and even patterns in mobile phone usage can provide valuable insights into an individual's financial reliability. On the other hand, credit unions in Ecuador value and need this type of credit solutions as they are expecting to issue credit cards through Coonecta, the transactional and payments network of credit unions.

3. Faster Loan Approvals and Enhanced Member Experience

   In a fast-paced world, individuals and small businesses in Ecuador need access to credit quickly, whether it’s for personal reasons or to take advantage of business opportunities. Traditional loan approval processes can be slow and bureaucratic, especially for members without established credit histories.

   Algorithm-based scoring models offer faster loan processing times by automating much of the credit assessment process. Credit unions can quickly analyze a range of data points and provide instant or near-instant credit decisions. This is particularly valuable for businesses needing short-term capital or individuals with urgent financial needs, especially for small loans, microcredit and nano-credits.

   This quick, streamlined approach leads to higher member satisfaction, as borrowers appreciate the ease and speed of obtaining loans. Additionally, credit unions can foster stronger relationships with members by offering a more responsive and member-focused service.

Ecuador Scoring Pilot Results

Made possible through a small innovation WFCU grant, WOCCU’s Hub C has partnered with CB Cooperativa and a third-party developer to pilot a lending algorithm and a software tool to streamline the credit granting process in the credit unions in Ecuador, expanding services to underserved populations, including women and youth. CB Cooperativa started testing the model in June 2024. As of January 2025, CB Cooperativa scored at least 1,500 loans it went on to issue based on its existing lending criteria. Below we highlight a set of recommendations meant to guide further efforts to scale the scorecard’s use to other credit unions in Ecuador and to potentially increase outreach to underserved populations such as women, youth and migrant workers.

Table 1 summarizes the scorecard results when applied to 1,000 of the loans approved and issued by Cooperative CB. Based on approval and delinquency rates, it appears the scorecard decision thresholds used in the pilot were set too conservatively to deliver meaningful efficiency gains for Cooperativa CB.

Table 1: Summary of Pilot Lending Decisions and Performance through February 2025

Scorecard Decision	All Loans	Share of Total	Problem Loans	% Non-Performing
Approve	113	11%	15	13%
Deny	317	33%	90	28%
Review	570	57%	78	14%
Total	1,000	100%	183	18%

The scorecard approved only 11% of the loans that CB Cooperativa approved and disbursed, but even that relatively small share of loans had a relatively high non-performing loan rate of 13%. A similarly high 14% non-performing loan rate was observed in the 57% of loans recommended for “review”. The 33% of loans with “deny” decisions had a non-performing rate of 28%, meaning that if Cooperativa CB declined one-third of the borrowers it approved using its existing underwriting standards, it would have rejected roughly three good borrowers for every bad borrower it avoided (see text box).

Goods Lost Per Bads Avoided at 28.3% Bad Rate for “Deny”

Further analysis of 195 loan applications rejected by Cooperativa CB showed closer agreement with the scorecard, which also rejected 87% of them.

Pilot Data Insights on Loans Underserved Populations such as Women and Youth

The pilot loan cohort had a slight majority (52%) of women borrowers. Table 2 shows that the approval rate for women was 4 percentage points higher and the rejection rate was 4 percentage points lower than for men.

Table 2: Recommended Scorecard Decisions by Gender

Decision	Female	%	Male	%
Approve	70	13%	42	9%
Deny	155	30%	160	34%
Review	299	57%	274	58%
Total	524	52%	476	48%

Table 3 indicates that the scorecard’s current pilot decision thresholds favored older applicants and would have considerably reduced lending to younger borrowers. It denied 52 of the 106 loans Cooperativa CB issued to borrowers under the age of 25. This points to the need for a modified scoring strategy that will enable a greater number of youth borrowers.

Table 3: Recommended Scorecard Decisions by Age

Decision	< 25		26-45		> 45
	Count	%	Count	%	Count	%
Approve	1	1%	64	9%	47	27%
Deny	52	49%	218	30%	45	26%
Review	53	50%	439	61%	81	47%
Total	106		721		173

For further testing and implementation of the scorecard with other credit unions in Ecuador, this initial pilot reveals the opportunity for a lender to set separate decision thresholds for particular borrower populations it hopes to increase its lending to. For example, if women have a better repayment rate than men (on average), the main scorecard strategies for more accurately targeting women at any given risk-appetite are:

1. Use gender as a characteristic in the model. This will increase credit scores for women or reduce the associated likelihood of the loan becoming “bad” (delinquent).
2. Using one scoring model, set separate decision thresholds for women and men. The lower passing score for women ensures the maximum number of loans can be issued to women for a give risk appetite level.
3. Develop separate scorecards for women and men. This strategy can result in the most accurate predictions, as different characteristics may have different risk thresholds for men and women and some characteristics could work better in predicting repayment for men or women.

Early Learnings and Key Considerations for the Potential Scaling of the Scorecard

Cooperativa CB’s management team has expressed that its main interest in the WOCCU scorecard was to decrease portfolio risk (non-performing loans) through more accurate decisioning. Its credit process, which uses a less sophisticated “expert” scorecard in the decision process, is efficient.

In its own review of the scorecard performance, Cooperativa CB noted that the loans approved by the algorithm-based scorecard had a par 30 delinquency rate that was 3% lower than Cooperativa CB’s overall loan portfolio. However, as noted earlier, the scorecard approved a very small share (11%) of the loans Cooperativa CB currently approves, and that 57% of loans requiring “Review” would still rely on the Cooperativa’s existing decisioning tools.

Cooperativa CB’s management team indicated that the pilot taught them a great deal about how to implement an automated credit scoring model for loan decisioning. At the same time, they desire more insight on how the model scores are calculated and in addition to visibility into the database of scoring results and dashboard with which they can view the decision results for specific loans.

The main challenge in implementing scorecards for organizations new to algorithm-based scoring is that any scorecard needs to be regularly monitored and managed with reference to its performance. Scorecards are most commonly managed using a set of scorecard management reports, as described in Siddiqi.1 The relative complexity of the piloted credit scorecard’s design will make use of traditional scorecard management reports challenging and will likely require recourse to the scorecard’s developer to make any changes to anything other than decision thresholds related to the model scores. To have a “global” scorecard for Ecuador credit bureaus, it is essential that an appropriate organization is designated to maintain the scorecard, monitor it regularly, and ensure it is working as intended.

Broad Conclusions

Algorithm-based scoring models can be an essential tool for credit unions, offering a more inclusive, accurate, and efficient way to assess creditworthiness. These models can improve financial inclusion by expanding access to credit for individuals and businesses that are typically excluded from traditional lending systems.

To merit further investment by CB Cooperativa, tracking turn-around time, minimizing manual loan approval, and maximizing segmentation analysis will provide the net benefits for further investment in this model. It also demonstrates the potential for further scaling for Ecuador credit unions as they expand their services to the members.

The following actions are recommended prior to using the scorecard with another credit union:

1. Adjust the decision profiles to increase the share of “approve” borrowers and reduce the share of “review”. During this initial pilot, the observed delinquency performance suggested the risk level of the “approve” and “review” score ranges are similar.
2. Prior to its use in decision making in any credit union, conduct a similar parallel test or back-test of the scorecard on a cohort of loans issued by that credit union to study the degree of agreement between the scorecard decisions and the lender’s current decisions. This will help to set (or re-set) decision thresholds that align with the lender’s current risk appetite.
3. Be sure to collect and store data needed to identify any populations targeted for outreach—such as women, youth, or migrants.
4. Measure the current baseline levels of lending to the target populations mentioned in point 3 in order to be able to track the impact of the scorecard and any lending strategies it informs on lending outcomes in the target population.
5. If and when there is enough scoring data and repayment history with a given lender, explore the intentional credit scoring strategy to maximize lending to groups of interest such as women, youth or migrants. Particularly, setting separate decision thresholds or cut-off scores for the target group can ensure the maximum number of loans are approved for any given risk appetite.
6. Explore different models or same model with different profiles, thresholds or businesses rules for microcredit, nano-credit and lines of credit or credit even for rural environments.