dLocal, Uruguay’s first unicorn, has established itself as a pioneer in cross-border funds since its founding in 2016. Right this moment, the corporate operates in over 40 rising nations, connecting greater than two billion customers with international expertise leaders.
Working at this scale requires strict and constant compliance processes. Every month, hundreds of service provider ecommerce web sites are reviewed to confirm alignment with dLocal’s insurance policies. These retailers have already onboarded dLocal as a cost service supplier or are within the means of onboarding. The compliance course of consists of verifying that retailers will not be promoting prohibited or non-compliant items or providers, and confirming that domains are legitimate and websites stay accessible. Nonetheless, the compliance problem extends past preliminary onboarding. Service provider web sites can change their choices or practices at any time, requiring ongoing monitoring to detect coverage violations.
In 2023, dLocal developed a proprietary generative AI resolution to assist streamline the web site assessment course of. Nonetheless, as the corporate expanded throughout extra nations and onboarded hundreds of latest retailers month-to-month, the rising complexity challenged the continued evolution of the answer. A good portion of instances nonetheless required guide assessment to validate compliance choices, making a bottleneck that restricted scalability as transaction volumes and service provider progress accelerated. The guide assessment of service provider web sites included analyzing their full product catalogs to determine objects violating the corporate’s prohibited product insurance policies. These lined a variety of unlawful or unethical classes with over 15 main classes and greater than 100 subcategories. This made the assessment course of extremely cumbersome.
To handle this, dLocal sought to automate the assessment course of to cut back guide effort, decrease human error, and facilitate compliance alignment. Moreover, as a result of onboarded web sites should be periodically reevaluated to substantiate continued adherence to insurance policies, automation would additionally assist streamline these recurring compliance checks for present retailers.
dLocal determined to companion with AWS to implement Amazon Fast Automate, a functionality of Amazon Fast Suite, as one of many service’s earliest adopters. Fast Automate helps enterprises construct, deploy and preserve resilient automations at scale.
Utilizing Fast Automate, dLocal automated its service provider compliance web site assessment course of, enabling large-scale, environment friendly, and constant coverage enforcement. With a sophisticated multi-agent automation structure engineered for enterprise-scale deployment, Fast Automate combines UI automation, API integrations, and automation orchestration in a single, absolutely managed resolution. Fast Automate makes use of generative AI to investigate inputs from the person and suggests an automation that may be modified and prolonged to take motion throughout enterprise methods and UIs, partaking a human when wanted. Via specialised AI brokers, Fast Automate helps organizations automate advanced processes throughout functions and departments whereas lowering operational prices by usage-based pricing.
On this put up, we share how dLocal labored carefully with the AWS group to assist form the product roadmap, reinforce its function as an trade innovator, and set new benchmarks for operational excellence within the international fintech panorama.
Answer overview
dLocal required a course of automation resolution able to navigating a number of product pages in web sites and reviewing every product towards their coverage to find out if the web site is compliance-aligned. The answer wanted to be extremely dependable, correct, and scalable to hundreds of internet sites. Given the excessive compliance danger related to permitting web sites that promote prohibited merchandise on its system, the accuracy bar for the automation resolution was excessive.
The automation was constructed utilizing Fast Automate, designed for automating advanced enterprise-scale enterprise processes. A key part of the answer is a UI Agent, configured to autonomously carry out web site navigation and information extraction. The UI Agent is a part of Fast Automate, enabling advanced browser-based duties.
The UI Agent takes pure language enter, navigates web sites, and performs duties on the web sites based mostly on the inputs. The automation was configured to navigate and consider service provider web site product pages to determine prohibited objects, serving to dLocal preserve their compliance posture. The AWS group collaborated carefully with the dLocal group to design and refine prompts, serving to the automation carry out optimally for the client’s wants. The next structure diagram illustrates the automation.
The workflow consists of the next steps:
- The case administration characteristic retrieves service provider instances from a processing queue for analysis, with every case containing important information, together with the web site URL that must be navigated and assessed. This characteristic permits parallel automation execution, permitting a number of instances from the queue to be processed concurrently, with every automation dealing with a delegated variety of instances to optimize effectivity and cut back total processing time.
- The UI Agent is designed to navigate web sites throughout varied languages. This agent makes use of specifically crafted prompts to determine prohibited merchandise and counterfeit objects in accordance with native rules and country-specific necessities. It autonomously browses web sites to gather product info and assess compliance with dLocal’s insurance policies.
- When web sites are inaccessible, the system marks them as Non Compliant and evaluates the accessible websites towards prohibited classes.
- The system classifies every case as Permitted, Rejected, or Web site Inaccessible, producing detailed reasoning for the choice and marking rejected instances for human within the loop (HILO) assessment.
- Every case is up to date with its compliance standing and reasoning, and the outcomes are saved to the case administration system for monitoring and future reference.
- The compliance outcomes are saved in Amazon Easy Storage Service (Amazon S3) to keep up compliance with regulatory requirements and facilitate correct documentation of the service provider evaluations.
Core automation elements
On this part, we focus on the core automation elements used within the resolution.
Case processing
The answer processes service provider web sites utilizing its case administration characteristic. This characteristic treats every recurring entity (on this case, each web site) as a person case, offering enhanced observability. Customers can observe metrics (equivalent to success fee) and assessment logs on the case stage slightly than throughout all the automation.
The characteristic handles a number of service provider evaluations concurrently utilizing parallel batch processing, and offers real-time standing updates and maintains complete data for the case information.
Automated web site evaluation utilizing the UI Agent
The UI Agent is a characteristic in Fast Automate to automate advanced browser-based actions based mostly on pure language directions. It handles inaccessible web sites, timeouts, and technical failures. The agent is ready to navigate a number of pages and cause over the content material of the pages to find out if an internet site is promoting prohibited merchandise. It offers a standardized information format for downstream enterprise methods.
The UI Agent causes with the web sites towards greater than 15 prohibited classes.
Human within the loop
For the few instances that stay ambiguous, Fast Automate can route exceptions to human operators, enabling guide assessment of crucial steps. Reviewers obtain structured proof web site information, flagged points, and screenshots to allow them to make fast, well-informed choices.
The UI agent offers detailed reasoning for the selections, which might be included within the human-in-the-loop assessment steps to provide reviewers further context.
Integration
The answer securely shops compliance paperwork utilizing its integration with Amazon S3. The answer permits customers to create, modify, and analyze structured information straight inside the automation.
Reliability and compliance
The answer logs actions for compliance data. Constructed-in exception dealing with mechanically detects and manages system errors and enterprise logic exceptions, facilitating automation continuity and enabling clever error restoration with out guide intervention.
Outcomes and influence
With Fast Automate, dLocal achieved a breakthrough in compliance automation, with as much as 75% of service provider web site critiques accomplished absolutely mechanically within the universe rolled out for the managed evaluations, with no human intervention required. Compliance specialists who as soon as spent hours on routine web site checks now focus completely on advanced instances and making nuanced judgment calls that require deep regulatory experience, resulting in higher-quality danger choices the place it issues most.
Past bettering onboarding effectivity, the automation helps determine situations of coverage drift the place beforehand compliant retailers introduce prohibited merchandise. This automation enhances dLocal’s compliance course of by offering steady, real-time monitoring throughout all the service provider portfolio.
dLocal is now positioned to scale service provider onboarding in step with enterprise progress, with a transparent path to push automation charges even increased. The aim is to develop automated protection to embody nearly all of compliance instances, serving to the group focus human experience on high-risk retailers, advanced regulatory situations, and instances requiring deep investigative judgment.
As Mauricio Clausen, VP of AI at dLocal, explains:”With Amazon Fast Automate, we’ve accelerated the evolution of one in all our superior AI assisted options: scaling service provider due diligence for cross-border market funds in rising markets. We built-in Amazon Fast Automate into our due diligence workflows to reinforce our compliance group’s capabilities. Our preliminary AI deployment streamlined processing throughout many instances, and we proceed to use knowledgeable assessment the place wanted. By leveraging Amazon Fast Automate, we’ve considerably decreased routine guide steps, permitting our compliance consultants to deal with probably the most advanced instances whereas sustaining accuracy and scalability.”
Conclusion
dLocal’s implementation of Fast Automate illustrates how fintech firms can preserve rigorous compliance requirements whereas scaling throughout rising nations. The outcomes dLocal achieved reveal how compliance can shift from a progress constraint to a strategic enabler.As an early adopter actively shaping the event of Fast Automate, dLocal has strengthened its place as an trade innovator. Their method to automating routine choices whereas reserving human experience for advanced, high-risk situations provides a mannequin for compliance organizations navigating comparable challenges at scale. In an trade the place regulatory calls for intensify as firms develop, this mix of AI-powered automation and strategic human oversight reveals how expertise can improve slightly than change knowledgeable judgment.
To study extra about how Fast Automate helps enterprises reimagine operations and automation, check with the next further assets:
In regards to the authors
Martín Da Rosa is a Senior Software program Architect at dLocal, bringing greater than 13 years of expertise to the software program trade. Primarily based in Uruguay and a part of the dLocal group since 2016. Martin performs a key function within the AI R&D division. His work focuses on bridging the hole between stable structure and utilized AI to ship high-impact expertise options.
Chethan Shriyan is a Principal Product Supervisor – Technical at AWS, based mostly in Seattle, WA. He brings almost 12 years of expertise in product and enterprise administration, together with over 6 years at Amazon. He’s obsessed with constructing and delivering expertise merchandise that create significant influence in buyer’s lives.
Guillermo Spallina is a Sr. Buyer Options Supervisor at AWS based mostly in Buenos Aires, Argentina, with greater than 25 years of expertise in IT and product administration, at present serving to DNB/ISV clients obtain their enterprise targets.
Reagan Rosario brings over a decade of technical experience to his function as a Sr. Specialist Options Architect in Generative AI at AWS. Reagan transforms enterprise methods by strategic implementation of AI-powered cloud options, automated workflows, and modern structure design. His specialty lies in guiding organizations by digital evolution, preserving core enterprise worth whereas implementing cutting-edge generative AI capabilities that dramatically improve operations and create new potentialities.
Natalia Mayorga Prieto is an Affiliate Scale CSM at AWS based mostly in Bogotá, Colombia, serving to clients as a Migration Success Lead for MAP 2.0 migrations throughout LATAM. With a ardour for expertise and dance, she is devoted to creating her mark as a younger lady within the tech world, inspiring others women.
