Back to the Prolego reference library

Selected delivery accounts and what they teach

Prolego has worked exclusively in AI since 2017. The accounts below represent only a small fraction of our work. They highlight major engagements in which Fortune 1000 companies and other large organizations hired our small expert teams for sustained, high-impact work, often over multiple months.

This is a selection of substantial client engagements, rather than a count of individual use cases. A single engagement can include several applications, strategic decisions and stages of delivery. Our broader experience includes many smaller projects, training exercises, executive workshops, preliminary prototypes, scoping assignments and confidential work that are not listed here.

These anonymous accounts show the depth and range of that experience: the business problem, our contribution and what happened through delivery. They span engineering, strategy and capability development across our history in AI.

The dollar and ranking labels identify historical profiles under which the work was shared. Publicly named clients appear separately in our company history.

Publicly traded personal loans company

Company scale

At year-end 2025, the company had approximately 9,300 employees and more than 1,300 branches, with branch and digital operations serving customers across 48 states. It generated about $6.2 billion in annual revenue and originated $14.4 billion in loans during 2025. Its managed receivables totaled $26.3 billion across approximately 3.8 million customer accounts.

Turn around a failing project by building reliability

The company had spent a year on a failing policy-and-procedure Q&A project. Its CTO asked Prolego to take over. The previous team had built the solution and then tried to address reliability.

The original design assumed employees would ask clear questions that made source retrieval straightforward. Actual questions included short queries such as "fraud?" The system first needed to understand what problem the employee was trying to solve.

Prolego built a clarification workflow that offered possible interpretations and asked follow-up questions before retrieving source material and answering. Practical examples and expert feedback guided that engineering. This is how Performance-Driven Development brings the biggest risk forward and solves for reliability while building the system.

Within the first 30 days, the client confirmed 90% pre-production accuracy in an evaluation framework built around practical examples. The previous team's less-than-5% accuracy was self-reported. These figures have different measurement bases; they are not a matched comparison or a production-accuracy result.

Build a foundation, then expand through reuse

Operations Intelligence is the collection of systems that enables AI to run reliably. The initial chat created capabilities that subsequent applications could reuse.

Prolego's team has been about two full-time equivalents, varying somewhat over the engagement. It supports chat while the work expands into Coach, sales preparation and Workbench development.

Usage was five times that of the incumbent solution immediately after launch. The policy-and-procedure chat is deployed across the national branch network. The client's training team handled enablement without difficulty, mainly policy guidance and encouragement. Employees could ask the AI itself how to use it; Prolego did not need to run a change-management program.

Engineering speed depends on operational agreement

Coach was intended to help employees advise personal-loan customers about consolidating outstanding debt, including credit card balances. The work exposed different calculation methods across groups and systems.

Roughly three or four key stakeholders needed to agree on the organization's method before Coach could provide consistent guidance. Additional engineers could not make that business decision. Prolego works with stakeholders daily to accelerate decisions in an organization accustomed to quarterly cycles.

Operational agreement has been the main rollout constraint. DevOps issues existed but did not drive the schedule.

The next applications

Automated sales preparation extends assistance from answering questions to preparing loan explanations, offers, sales collateral and objection-handling guidance.

An AI Workbench, such as Claude or ChatGPT connected to Operations Intelligence, gives key knowledge managers and decision makers a flexible way to analyze information and maintain guidance. Uses include consolidating policies for expert review, analyzing recurring employee questions and preparing decision briefs.

Sales preparation remains in development. Workbench rollout has begun and continues as new uses are added. Approved improvements to guidance could also support chat and Coach through the shared foundation.

Coach's estimated financial potential

The client supplied these annual benefit estimates for Coach. Rollout is still underway, so these are expected benefits rather than realized returns.

The client separately estimated $3 million annually for AI infrastructure. These categories and the infrastructure estimate are not a reconciled net return or investment multiple.

What the case illustrates

• Build reliability around real work. Clarifying unclear questions addressed employee behavior the original design had missed.
• Expand through reuse. Production chat established Operations Intelligence that supported another interface and subsequent development.
• Match engineering speed with business decisions. Consistent guidance required agreement on how the business should operate.

Global aerospace manufacturer and systems provider

Company scale

The business reported approximately $30.2 billion in 2025 sales. Its published company profile describes more than 80,000 employees, an engineering workforce above 20,000 and over 250 sites worldwide.

The manual process

Several hundred engineers perform a tedious, custom process to prepare material supporting Federal Aviation Administration (FAA) approvals. They manually copy and paste information to assemble testing material and trace FAA testing requirements down to software code.

Prolego's contribution

Prolego deployed two AI workflow solutions, including generation of FAA testing material and traceability linking requirements to code. Both are in production, automating preparation that engineers previously performed by hand. Workbench rollout has also begun as the work develops.

Prolego continues to work with the company as one contributor to a larger effort.

Status and expected value

The client estimates that the broader automation effort could save more than $50 million annually. Both Prolego solutions are in production. Prolego's contribution is one part of that broader initiative.

Fortune 100 Bank

Leverage AI to automatically detect compliance risks

The objective was to reduce more than $100 million in compliance fines by detecting misalignment between 500 regulatory rules, including FINRA, SEC and CFTC rules, and 20,000 internal policy sections. The historical account records an existing system at 16% accuracy. Prolego's phase-one embeddings and matching work achieved 50% accuracy. Phase two was deploying LLMs for retrieval and filtering toward an 80% accuracy target at the time of that account. The bank intended to use the solution as the foundation of its compliance program.

Technology: Embeddings, LLMs (Mistral-7B, Mixtral 8x7B), custom text parsing, ML search ranking

Delivery and planned handover

The current delivery account records generative AI components delivered since 2023 and handed to the client's internal team before production, as planned. The historical phase results and later handover describe different points in the engagement. This work is separate from the personal-trade compliance case below.

PE-backed regulatory services company

Automate application preparation

The company prepares applications for submission to the U.S. Food and Drug Administration (FDA).

Prolego built an automated workflow to ingest and process application material and produce draft applications. The company's team could then review and modify the drafts before submitting them to the FDA.

Deliver the foundation and hand it over

The workflow reached production after three months. Prolego provided the workflow foundation, then the client's internal team took over the work.

At handover, the client told Prolego that it believed the solution was on track toward its goal of reducing application preparation from six weeks to two weeks.

$1B Self-Regulatory Organization

Analytics strategy and an $80 million multi-year funding request

Prolego was hired to build the organization's advanced analytics strategy and $80 million multi-year funding request. We ran an internal process with business-line leaders to identify regulatory opportunities, and we organized them into foundational capabilities. Strategy included business cases such as cost savings from automating examinations, and increased regulatory coverage from graph technology and predictive analytics.

Build foundational AI systems for processing millions of documents at scale

Regulators process millions of pages of text from filings, policies, and correspondence. Prolego was hired to build the organization’s first text-processing solutions based on NLP transformer technology. The major scaling challenge is building the training data without human labeling. We developed custom techniques for scaling label data including label functions and active learning.

Technology: BERT transformers, embeddings, custom labeling functions, active learning

Scale investigatory graph to thousands of nodes with entity resolution

Prolego was hired to develop custom engineering solutions for scaling a graph. Regulators leverage graph technology for detecting bad actors and performing investigations. A major challenge for populating graphs is reconciling entity relationships within documents (i.e. John Smith works at X bank). Prolego developed the foundational libraries for resolving these entities across different regulatory documents.

Technology: BERT transformers, custom text parsing libraries, embeddings

Fortune 50 Bank

A business case targeting $2 million in annual cost reduction

Prolego was hired to give the bank’s compliance department its first “quick win” with AI. Prolego conducted a survey of business line opportunities and recommended starting with personal trade and compliance. At the time of the engagement, employees were required to submit brokerage statements as part of trading approval, and the bank processed 20% manually. Prolego developed the custom software to detect transaction tables within 20-page PDF statements and extract key table entities. Prolego transitioned the technology to the bank’s internal team for integrating it into their existing trade compliance workflow.

Technologies: Computer vision, OCR, NLP transformers, custom zoning & text processing

Top-20 Life Insurance Carrier

• Develop a custom solution for detecting contractual risk using NLP transformers

• Built custom models for extracting named entities from policies with the goal of accelerating acquisitions using pre-trained Q&A models.

Prolego fine-tuned BERT models to detect high-risk clauses in this life insurer's contracts and put an NLP transformer into production in 2019, using the same architecture behind today's LLMs. We are not aware of an earlier production deployment of this kind in life insurance.

Top-10 Life Insurance Carrier

• Automate underwriting through inverse reinforcement learning

• Detect unresolved customer chat conversations with NLP transformers and topological data analysis

$18B Automotive Remarketer and Retailer

• Developed company’s first vehicle product recommendation engine using collaborative filters

• Presented strategy for new revenue opportunities from selling data services

$100B Pharma Corporation

• Developed messaging strategy for AI and MLOps initiatives

• Created narrative-based innovation communication package for centralized AI services team.

Fortune 100 Defense Contractor

• Trained 500 managers and executives on AI opportunities across 5 business units

• Helped the Chief AI Officer develop the organization’s 5-year AI roadmap

What carries across the cases

The contribution varies with the need: a production workflow, specialist components for an internal team, a shared foundation, strategy or organizational capability. The consistent work is understanding the business, making expert judgment concrete and building the system around it.

Use the fit guide and scoping approach to compare these accounts with your task, sources, experts and operating responsibilities.