Introduction
Expense data is a goldmine for understanding which AI services businesses are actually paying for, rather than just talking about. A recent survey from fintech firm Ramp, based on its clients’ expense records, reveals that 34.4% of participating businesses are paying for Anthropic services, while 32.3% pay for OpenAI. This guide will show you how to conduct a similar analysis—using the Ramp example as a case study—to gauge real-world AI adoption trends among companies.
What You Need
- Access to aggregated expense data from a platform like Ramp (or a similar corporate card/expense management system).
- A defined set of AI service providers to track (e.g., Anthropic, OpenAI, Google, Microsoft).
- Data analysis tools (spreadsheet or analytics software) to categorize transactions and calculate percentages.
- Knowledge of vendor names—how each AI provider appears on merchant statements (e.g., “Anthropic PBC,” “OpenAI”).
Step-by-Step Guide
Step 1: Collect Comprehensive Expense Data
Start by gathering expense records from a representative sample of businesses. Ramp’s data comes from its own clients, which include thousands of companies across industries. If you’re not using Ramp, you can partner with a corporate card provider or use your own organization’s purchase orders. Ensure the dataset includes at least several hundred businesses to make the comparisons meaningful. Back to top
Step 2: Identify AI Service Transactions
Scan the expense records for payments to AI service vendors. For the Ramp study, they looked at line items labeled as subscriptions or usage fees for services like Anthropic’s Claude and OpenAI’s ChatGPT. Create a list of all distinct AI providers, noting their official merchant names. For example, Anthropic payments might appear as “Anthropic PBC” or “Anthropic AI.” OpenAI might show as “OpenAI, LLC” or “ChatGPT.” Back to top
Step 3: Categorize Businesses by Provider
For each business in your dataset, determine which AI services they have paid for during your analysis period (monthly, quarterly, or annually). A single business may subscribe to multiple AI services, so you’ll want to count each business independently per provider. In Ramp’s survey, 34.4% of participating businesses had at least one payment to Anthropic, and 32.3% had at least one to OpenAI. Use a pivot table or SQL query to tally unique companies per vendor. Back to top
Step 4: Calculate Adoption Percentages
Divide the number of businesses paying for a specific provider by the total number of businesses in your dataset. Multiply by 100 to get the percentage. For example: (businesses paying Anthropic / total businesses) × 100 = 34.4%. Do the same for OpenAI and any other competitors. Ensure the denominator is consistent—Ramp used only businesses that had any expense activity, not all possible businesses. This gives a clear market-share picture. Back to top
Step 5: Compare and Interpret the Results
Once you have percentages for each provider, rank them to see who leads in paid business adoption. Ramp’s data shows Anthropic edging out OpenAI by 2.1 percentage points. This difference, while small, suggests that Anthropic has successfully attracted a slightly larger share of paying business customers. However, consider factors like trial periods, free tiers, and bundled services that might skew raw numbers. Interpret the data in context—such as the specific time frame and the sample’s industry mix. Back to top
Step 6: Validate with Additional Metrics
Expense data alone doesn’t tell the full story. Cross-reference your spending percentages with other signals like client feedback, product announcements, and contract lengths. For instance, Ramp’s survey also looked at total spend amounts, not just number of businesses. You can repeat Steps 1–5 using total dollars spent instead of counts to see if the same pattern holds. This dual analysis can reveal whether Anthropic’s lead in customer count also translates to more revenue. Back to top
Tips for Accurate Analysis
- Watch for sample bias: Ramp’s clients are often startups and mid-market companies, so the data may not reflect Fortune 500 firms. Always note the demographic of your expense data source.
- Account for free tiers: A business might use ChatGPT extensively via a free account and never show up in expense records. Paid usage only captures the tip of the iceberg.
- Update regularly: AI adoption shifts quickly. The 34.4% vs 32.3% split may change within a quarter. Run your analysis periodically to stay current.
- Combine with surveys: Expense data is objective, but supplement it with direct business feedback to understand why companies choose one provider over another.
- Use consistent provider names: Variations like “OpenAI” and “ChatGPT” could create duplicate counts. Standardize vendor names before analysis.