Spade knows where your money went. Does your bank?

Walmart ought to be easy to recognize. The name is on the building, the bag and, usually, the receipt. Inside a payment system, however, familiarity dissolves. Spade says it observed more than 165,000 merchant IDs associated with Walmart over two years. In three months, it saw more than 65,000 transaction descriptors. A household name had acquired the administrative habits of an international spy.

A very expensive spelling problem

Merchant IDs are assigned at the processor level. They can change when businesses switch systems or acquirers, and one brand can have many. Descriptors are equally unruly: abbreviations, branch numbers and processor prefixes crowd into a small space. The money arrives. The meaning takes a less direct route.

That distinction matters. A rewards program needs to know whether dinner happened at a participating restaurant. A fleet-card issuer needs to know whether fuel was purchased near the expected vehicle. Accounting software needs a consistent business identity. Each is asking a different question of the same scruffy record.

The map was too small

Spade’s early proposition was modest: put recognizable logos beside transactions. In a 2026 interview, co-founder Oban MacTavish described hand-labeling merchant data, scraping store locators and working in Photopea, a free image editor. Before the polished platform came a decidedly unpolished collection of human effort.

The first weakness was coverage. SafeGraph’s 2023 case study records Spade scraping roughly 500,000 to one million locations itself. That was substantial work and still an inadequate map. Buying access to SafeGraph Places brought millions of US merchant locations within reach. Breadth and data quality helped persuade the team to use a partner.

“We had our own data but it was just not as broad as we needed.”

Oban MacTavish / SafeGraph case study, 2023

This is the practical detail behind the company’s claim to proprietary intelligence. A useful database can incorporate outside records. The work lies in connecting, verifying and maintaining them for a particular decision. Scraping everything yourself is a business choice, and sometimes an expensive one.

Forty milliseconds to be useful

Spade now supplies APIs for card transactions, bank transfers and third-party transaction data. Customers send payment details; Spade searches its merchant records and returns structured context, including identities, categories and, where available, locations. Its tooling adds merchant search, category personalization and action triggers.

The company emphasizes verified business records as its starting point. Machine learning still has a role: its documentation describes a model that assesses confidence in matches. The distinction is between having a record of a real business and merely making a payment string look tidy.

Timing determines the product’s usefulness. Information that arrives after authorization can improve a statement, but cannot help decide that purchase. Spade reported processing latency below 40 milliseconds at the 99th percentile in March 2026. Its documentation also warns that integration patterns affect latency. A fast lookup still needs a sensible route home.

A reward is a promise with an address

Bilt makes the stakes unusually clear. Its local rewards depend on identifying participating businesses and specific locations. Spade’s case study describes inconsistent descriptors, manual mapping and investigations that could take hours. Rewards were sometimes issued despite uncertain attribution to preserve the member experience. Bad data had become a recurring operating expense.

Enrichment improved the foundation, but matching Bilt’s partner lists still required manual reconciliation. Bilt then adopted Spade’s Attribution Agent to automate validation against those lists and the merchant graph. The second product addressed the work left over after the first product had done its job.

Coast uses the same underlying context for a different purpose: fleet-card authorization. Merchant identity and coordinates let it apply proximity rules and merchant or category restrictions. Mercury’s case study focuses on consistent inputs for accounting and financial automation. The common purchase is trustworthy context; the resulting products are quite different.

The bill, and the bigger bet

Spade generally bills by enriched transaction volume, with a platform fee for automated workflows, according to MacTavish’s April 2026 interview. That makes the buying calculation concrete: compare the service and integration cost with manual investigations, incorrect rewards and unreliable decisions.

The company announced a $5 million seed in February 2023, a $10 million Series A that December and a $40 million Series B in March 2026, led by Oak HC/FT. The latest round supports expansion from enrichment into workflows. Spade reported 470% revenue growth for 2025 and peak daily volume of 1.9 billion transactions; peak volume should not be mistaken for an ordinary day.

Copy the question, then test the answer

Plaid Enrich and Ntropy also sell enrichment. An internal matching system is another alternative. Spade’s case rests on merchant records and speed, but buyers should test their actual transaction mix. A neat demonstration is no substitute for the awkward records your own customers generate.

There are useful limits in Spade’s documentation. Confidence scores are not probabilities. Location scores are unavailable for digital card transactions and transfer enrichments. Higher thresholds suit authorization and fraud decisions; lower thresholds may suffice for budgeting or display. Merchant context contributes evidence, while the customer remains responsible for the decision.

The lesson worth borrowing is the sequence: identify the missing fact, establish a reliable record, then automate the work around it. For Spade, a logo opened the conversation. Knowing which business stood behind it gave the conversation somewhere profitable to go.