A contractor with a Buró de Crédito score of 650 may be a better credit risk than one with a score of 720. That statement runs against the instinct of most credit analysts. It also happens to be regularly true in construction lending in Mexico, and the reason is worth understanding if you want to know why Mango's credit default rate sits at 0.4% when the industry average for new construction lenders is 2-3%.
This post explains the 14 data signals we use for credit underwriting, why the bureau score is necessary but not sufficient, and what the data has taught us about which signals actually predict whether a contractor will pay on time.
The Bureau Score Problem in Construction
Buró de Crédito scores in Mexico measure personal credit behavior: mortgage payments, auto loans, credit cards. A contractor who has never taken a personal mortgage and uses business accounts for everything may have a thin bureau file that scores lower than a salaried employee with a mortgage and two credit cards. That lower score does not reflect business credit risk — it reflects that the contractor's financial life runs through channels that bureaus do not capture.
More importantly, bureau scores do not capture the information most predictive of construction trade credit repayment: does the contractor get paid by their clients on time? Do their projects complete on schedule? Do they have a history of disputing delivery invoices in bad faith to delay payment? Is their current project pipeline sufficient to generate the cash flow to service the credit line they are requesting?
None of these questions are answered by Buró de Crédito. The bureau tells you about the person's financial behavior in consumer contexts. Construction trade credit repayment is a business behavior with different drivers.
Signal Group 1: Project Activity Data (4 Signals)
The first group of signals comes from permit records, which in Mexico City are maintained by SEDUVI (the Secretariat of Urban Development and Housing) and have been digitized since 2019. Permit records tell us: how many active building permits a contractor currently holds, the cumulative value of projects they have permitted in the past 24 months, the permit completion rate (permits that have received final sign-off versus permits that are open or abandoned), and the average time from permit issuance to project completion.
A contractor with 3 active permits totaling MXN 12,000,000 in permitted project value and a 90% historical completion rate is demonstrably a lower risk than one with 1 active permit and a 60% completion rate, regardless of bureau scores. The permit data gives us a ground truth view of how active and reliable their business operation actually is.
Completion rate is especially important. Contractors with high abandonment rates on prior projects represent the highest risk category in our portfolio, even when their bureau score is acceptable. Projects are abandoned because clients do not pay or because the contractor ran out of capital. Either way, abandoned projects predict future repayment difficulty.
Signal Group 2: Platform Payment History (3 Signals)
For contractors with any transaction history on Mango, three payment behavior signals are the most predictive features in our model: days-to-payment relative to terms (how many days late or early does this contractor typically pay), payment consistency under cash flow pressure (do they miss payments when their own projects are behind schedule), and dispute-to-payment ratio (what percentage of their invoices are disputed before payment, which can indicate either legitimate quality issues or systematic delay tactics).
These three signals alone predict repayment behavior with higher accuracy than the bureau score for contractors with 6+ months of platform history. The payment behavior data is constructed from actual transactions — not proxies or estimates. When we have it, it dominates the model.
For new contractors with no platform history, we weight this signal group at zero and increase the weight on permit data and the bank account signals described below. The model automatically adjusts signal weights based on data availability.
Signal Group 3: Cash Flow Indicators (3 Signals)
We request access to 12 months of business bank account transactions as part of the credit application. From that data, we extract three signals: average monthly cash inflow (what does the business actually collect per month), inflow-to-credit-request ratio (is the requested credit line proportional to actual revenue), and seasonal cash flow variance (does the business have reliable monthly inflows or is cash highly lumpy).
The inflow-to-credit ratio is the most actionable signal. A contractor requesting a MXN 2,000,000 credit line whose bank account shows average monthly inflows of MXN 400,000 is requesting credit at a 5x monthly revenue multiple. That is a red flag — not a disqualifier, but it triggers manual review and usually results in a lower initial credit line that increases as payment history accumulates.
Cash flow lumpiness (high variance between months) is common in construction and is not itself a risk signal. Construction is project-based; cash comes in large milestone payments separated by quiet periods. What matters is the average and trend. A contractor with MXN 600,000 average monthly inflows showing a clear upward trend over 12 months is a better risk than one with MXN 600,000 average but a declining trend over the same period.
Signal Group 4: Order Behavior on Platform (2 Signals)
How a contractor uses the Mango platform itself is informative. Two behavioral signals have emerged as unexpectedly predictive: order-to-delivery acceptance rate (what percentage of delivered orders are fully accepted versus returned or disputed), and catalogue browsing pattern (does the contractor browse and order systematically by project phase, or do they place many small unplanned orders?).
High return rates can indicate a contractor who is ordering without clear specifications and relying on returns to correct mistakes. That behavior pattern correlates with higher-than-average default risk, probably because it reflects a less organized project management approach overall. This is a weak signal on its own — a single bad delivery can inflate return rate without reflecting contractor behavior. It only becomes significant in combination with other signals.
Systematic, phase-based ordering (a contractor who places 3-4 large orders aligned with project phases) correlates with lower default risk than order patterns that look reactive — many small orders with no clear phase structure. The interpretation is that systematic ordering reflects a contractor with a clear project plan, which correlates with project completion rates and cash flow reliability.
Signal Group 5: External Context (2 Signals)
Two external signals provide context that the contractor-specific data cannot: current materials price trends in the categories the contractor is ordering, and the overall economic calendar for construction in Mexico (permit issuance rates, housing starts, commercial development approvals). These are macroeconomic signals that affect all contractors simultaneously but affect some more than others depending on their project mix.
We use these signals primarily for stress-testing: when materials prices are rising sharply (as they did in the cement and rebar categories in H1 2024), we tighten initial credit lines for new applicants and increase monitoring frequency for existing accounts. This is not because any individual contractor is a worse risk — it is because the cash flow pressure from rising material costs affects the whole portfolio, and we want to catch potential problems earlier rather than later.
How the Model Weighs the Signals
We do not use a fixed formula. The model is a gradient-boosted tree that was trained on 18 months of platform data from our beta cohort and updated monthly as new payment outcomes arrive. The feature weights shift as the training dataset grows. Currently, the permit completion rate and the inflow-to-credit-request ratio are the two highest-weight features. The bureau score contributes about 12% of the model's prediction weight — necessary but not dominant.
For applicants with no platform history and no bank account data access, the model falls back to permit data and bureau score only. In that reduced-data mode, the model is less confident, and we reflect that by offering a lower initial credit line — typically 50-60% of what a well-documented applicant with equivalent business scale would receive — with a clear path to credit line increases after the first 3-6 months of payment history accumulate.
What a 0.4% Default Rate Actually Means
Our 0.4% lifetime default rate (as of December 2024) covers all credit disbursed since platform launch. The construction industry average for new trade credit lenders in Mexico is 2-3%, based on published data from CONDUSEF (the National Commission for the Protection and Defense of Financial Services Users) for SME lending in the construction sector.
We are not claiming the 0.4% rate is permanently sustainable. Our beta cohort is self-selected — contractors who were comfortable with a new digital platform are probably more organized than the median contractor in the market. As we scale and attract less self-selected applicants, the default rate will likely increase toward 0.8-1.0%. We have modeled for that in our unit economics.
The model's job is not to get to zero defaults. Zero defaults would mean we are declining too many creditworthy contractors. The job is to price risk correctly: extend credit to contractors who will repay, charge rates that reflect the risk of those who might not, and build a portfolio that is profitable at scale. The 14-signal model is the mechanism for doing that more accurately than a bureau score alone.
If you are a contractor who has been declined or undersized by a traditional lender for materials credit, the Mango application process surfaces your project track record directly rather than relying on a bureau proxy. Apply at mangxo.org. Credit decisions arrive within 4 hours for standard applications.