How Loop Turned Back-Office Friction Into a $44 Million AI Business

When people discuss artificial intelligence in business, they often focus on visible applications such as chatbots, predictive analytics, autonomous systems, or AI-powered customer service.

However, some of the most valuable opportunities exist in less visible areas: accounting, procurement, payments, logistics, compliance, and data reconciliation.

Loop emerged from one such opportunity.

In 2021, Matt McKinney left his role as a data scientist at Uber to build Loop, an AI-powered finance and logistics platform. The company was created around a problem McKinney and co-founder Shaosu Liu had observed inside Uber’s freight operations: large volumes of invoices, contracts, spreadsheets, and supporting documents were being processed manually, often with significant errors.

The challenge was not a lack of data. It was the inability to structure and connect that data efficiently.

According to Inc., Loop generated $15.7 million in revenue in 2025 and reached a $44 million annualized run rate during the first quarter of 2026. The company subsequently raised a $95 million Series C in the second quarter.

The company’s growth offers a useful case study in how professionals can identify AI opportunities within traditional business processes.

From Uber’s back office to a startup opportunity

McKinney joined Uber in early 2019 as a data scientist. Liu, who later became his co-founder, was involved in his interview process and had worked as the head of engineering at Uber Freight.

Their experience at Uber exposed them to a contradiction that exists in many technology-driven companies: the customer-facing systems may be highly advanced, while internal operations remain dependent on fragmented and manual workflows.

Uber had developed sophisticated technology for transportation and logistics, but parts of its back-office operations still involved:

• Manual invoice processing.

• Ledger coding.

• Reconciliation between invoices and contracts.

• Comparing documents from different sources.

• Managing information stored in PDFs and spreadsheets.

• Correcting errors after they had already entered the workflow.

The scale of the problem made it particularly costly. McKinney told Inc. that approximately 30 percent of invoices contained errors, creating serious consequences for supply-chain operations.

For a smaller organization, an invoice error may lead to a delayed payment or a dispute. For a large enterprise, errors can affect cash flow, vendor relationships, financial reporting, procurement decisions, and operational planning.

This led McKinney and Liu to a broader hypothesis: if a company as large as Uber faced this issue, many other enterprises probably faced it as well.

Why invoice reconciliation matters

Invoice reconciliation is the process of comparing an invoice against relevant business records to determine whether the amount, goods, services, and terms are accurate.

A typical reconciliation process may require an organization to compare:

• A supplier invoice.

• A purchase order.

• A contract.

• A delivery confirmation.

• A shipment record.

• A payment history.

• Tax and compliance information.

In many companies, this information is stored across separate systems and formats. One document may be available as a PDF, another as an Excel spreadsheet, and another inside an enterprise resource planning system.

Employees may then need to manually compare the records, identify discrepancies, contact suppliers, and update accounting systems.

This process is repetitive, but it is not insignificant. It influences the accuracy of financial records and the efficiency of supply-chain relationships.

The underlying business issue is therefore not simply that employees spend too much time reading invoices. The larger problem is that critical operational information remains disconnected and difficult to use.

Loop’s central insight: structure the data first

Many companies initially approach automation by trying to eliminate a specific manual task.

For example, they may attempt to automate invoice matching or reduce the number of employees involved in document review.

Loop’s approach appears to have been broader. Rather than treating invoice reconciliation as an isolated task, the founders focused on organizing the underlying data.

This distinction matters.

An AI system that only flags invoice discrepancies may improve one workflow. An AI system that extracts, structures, and connects information from contracts, invoices, and logistics records can potentially support a much wider range of business decisions.

Large language models are particularly useful in this context because they can interpret unstructured information and convert it into structured data. This may include extracting:

• Vendor names.

• Payment terms.

• Product descriptions.

• Contract obligations.

• Quantities.

• Pricing.

• Delivery dates.

• Shipping details.

• Exceptions and discrepancies.

Once the information is structured, it can be used beyond the original accounting workflow. It may support procurement analysis, cash-flow planning, payment automation, supplier management, and operational forecasting.

This is one of the most important lessons from Loop’s model: the first automation use case may be only the entry point into a much larger data platform.

Moving beyond a single use case

Loop initially focused on accounting and reconciliation. Over time, it expanded into adjacent functions, including payments, procurement, and planning.

This expansion reflects a common pattern in enterprise software.

A company may begin by solving a narrow, urgent problem for one department. Once it has access to relevant data and earns the customer’s trust, it can introduce additional capabilities.

The strategic advantage is that each new function can build on the same data foundation.

This creates a stronger product than a collection of disconnected automation tools. The platform becomes more valuable as it understands the relationships between a company’s suppliers, contracts, invoices, shipments, payments, and planning decisions.

The legacy industry Loop challenged

Before AI-based systems became widely available, many companies outsourced document comparison and supply-chain reconciliation to specialized service providers.

These organizations often relied on large teams of employees to compare contracts, invoices, spreadsheets, and supporting evidence. Inc. describes this as a legacy “stare-and-compare” sector involving workers in countries including India, Colombia, and the Philippines.

Outsourcing helped companies manage the workload, but it did not necessarily solve the underlying data problem.

The documents still had to be reviewed manually. The information often remained trapped in unstructured formats. Businesses could process the records, but they did not always gain a reusable, searchable, and connected data asset.

Loop’s proposition was different: use AI not only to automate the existing review process, but also to organize the information generated by that process.

For enterprise customers, this can create two benefits:

1. Lower operational effort and faster processing.

2. Better access to information for future decisions.

The second benefit may ultimately be more strategic than the first.

Why vertical integration can create a moat

Loop’s founders emphasized the importance of vertically integrating parts of the technology stack. Their belief was that companies create defensibility by solving difficult technical and operational problems themselves.

Vertical integration can involve controlling several parts of the product experience, such as:

• Data ingestion.

• Document processing.

• AI extraction.

• Workflow automation.

• Exception management.

• Customer-specific integrations.

• Reporting and analytics.

This approach may require more time and capital than assembling a product from third-party tools. However, it can also give a company more control over quality, security, performance, and customer experience.

For enterprise AI products, this is especially important because customers often care about more than the model itself. They want:

• Reliable outputs.

• Auditability.

• Integration with existing systems.

• Data protection.

• Clear accountability.

• Consistent performance across large document volumes.

A generic AI model may provide the intelligence, but the surrounding infrastructure determines whether the product can function in a real business environment.

The founder’s decision to leave Uber

Leaving a large technology company to start a business is difficult under any circumstances. McKinney’s decision was made more complicated because he and Liu had recently become fathers.

A startup involves uncertainty around income, product-market fit, hiring, customer acquisition, and funding. In contrast, an established technology company offers stability, resources, and a predictable career path.

The founders nevertheless secured seed-round term sheets within eight weeks, according to the Inc. interview. Loop later raised its Series A from Founders Fund.

Their experience illustrates an important aspect of founder-market fit. Investors may be more willing to support entrepreneurs who have:

• Worked on complex systems at scale.

• Experienced operational problems firsthand.

• Developed relationships with potential customers or employees.

• Built technical credibility in demanding environments.

• Identified a problem from direct observation rather than market speculation.

The Uber connection also became part of Loop’s broader talent and investor network. Several Loop employees were Uber alumni, while investors included Garrett Camp, an Uber co-founder, and Ryan Graves, Uber’s first employee.

However, a strong professional network is not a substitute for a real business problem. It can accelerate fundraising and hiring, but customer value remains the foundation of sustainable growth.

What professionals can learn from Loop

1. Look for expensive inefficiencies

The best AI opportunities are not always the most glamorous ones.

Professionals should examine processes that are:

• Repetitive.

• Error-prone.

• Dependent on multiple documents.

• Spread across disconnected systems.

• Difficult to audit.

• Expensive to outsource.

• Important to cash flow or compliance.

These characteristics often indicate that a process may be suitable for AI-assisted automation.

2. Focus on the data behind the task

Automating a workflow is useful, but structuring the information behind it can create more long-term value.

For example, a company may initially want to automate invoice approval. Once invoice data is organized, it may also be able to understand supplier pricing, detect recurring discrepancies, forecast payment obligations, and improve procurement decisions.

The practical question is not only, “What task can AI perform?” It is also, “What valuable business data will become usable once this task is automated?”

3. Start narrow, then expand logically

Enterprise customers are often more willing to adopt a product that solves one clear problem than a platform that promises to transform every department.

Loop’s path—from accounting to payments, procurement, and planning—shows how a focused entry point can lead to broader adoption.

A sensible expansion strategy is to move into areas that share:

• The same customer.

• The same data.

• Similar workflows.

• Related business outcomes.

• Existing integration points.

4. Build for operational trust

AI systems used in finance and logistics must be accurate, explainable, and dependable.

An incorrect recommendation in a creative workflow may be inconvenient. An incorrect invoice decision can affect a supplier relationship or financial statement.

Enterprise AI companies therefore need to build more than a useful interface. They need systems that provide:

• Human review where appropriate.

• Clear exception handling.

• Traceable data sources.

• Confidence indicators.

• Access controls.

• Audit trails.

• Strong integration with existing enterprise software.

5. Treat traditional industries as technology opportunities

Industries with older software and manual processes may appear less innovative, but they can contain significant opportunities.

A company does not need to invent an entirely new category to build a valuable technology business. It may create substantial value by improving a process that thousands of organizations already perform every day.

The key is to find a problem where the cost of inaction is high and the benefits of improvement are easy to measure.

The broader significance for enterprise AI

Loop’s growth demonstrates how enterprise AI is moving beyond general-purpose productivity tools.

The next phase of AI adoption will likely involve systems embedded in specific business processes. These systems may not always be visible to the public, but they can influence how companies manage money, suppliers, inventory, contracts, logistics, and planning.

The most durable products may combine three capabilities:

1. Understanding unstructured information.

2. Executing or supporting business workflows.

3. Creating a reusable data foundation.

This combination is more powerful than simple document summarization or chatbot functionality. It allows AI to become part of the operating infrastructure of a business.

For professionals, the implication is clear: understanding business processes may be just as important as understanding AI models.

The strongest opportunities often arise where technical capability meets operational knowledge.

Final takeaway

Loop did not begin by chasing the most fashionable AI application. It began with a difficult and largely overlooked operational problem: the manual handling and reconciliation of fragmented finance and supply-chain information.

By turning unstructured documents into usable data, the company created a foundation for automation across accounting, payments, procurement, and planning. Its reported $44 million annualized run rate and $95 million Series C illustrate the commercial potential of solving high-value enterprise problems with AI.

For business and technology professionals, the central lesson is simple:

The next major AI opportunity may be hidden inside the processes companies have accepted as inefficient for years.

The companies that find those problems—and solve them deeply—may build more lasting businesses than those merely adding AI features to existing products.

Recently funded and expanding your team?

WorkNation helps startups and growth-stage companies hire top talent faster.

📩 hiring@worknation.buzz | 🌐 worknation.buzz/hiring-solutions