Most finance and procurement teams can tell you what they spend on their SaaS vendors. Ask the same question about AI, and the answer isn’t as clear.
That gap is where our latest webinar started. Sam Voshall, Sr. Product Marketing Manager, and Sreesh Reddy, Director of Product, introduced AI Spend Visibility: SpendHound’s newest module built to consolidate AI usage and cost data across providers in one place.
The session covered the research behind the AI spend problem, why provider dashboards don’t solve it, a live look at the module, and what’s shipping next, from cost attribution tags to Claude Code analytics. It closed with how visibility connects to In-Platform Benchmarking, Deal Grader, and negotiation support.
Watch the full replay here.
Key takeaways:
Sam opened with the one thing that’s consistent across SpendHound’s customer base, from token-maxing power users to teams still validating their first pilot: nobody has really figured out AI spend visibility.
To put numbers behind it, SpendHound surveyed 172 CFOs, finance leaders, and procurement leaders, then paired those responses with actual spend data from the platform. Three findings stood out:
That last one takes a few forms: questioning a standalone AI vendor, re-evaluating an AI add-on from an existing SaaS provider, or asking whether the company is now paying multiple vendors for overlapping capabilities.
“AI has become a meaningful spend category before most organizations have really built the processes needed to manage it.” — Sam Voshall
The full research is available in SpendHound’s AI Spend Report.
Part of what makes the category so hard is that the old budgeting model doesn’t transfer. A typical SaaS contract has a set number of seats and a relatively predictable annual cost. AI spend scales with usage: tokens, models, API calls, projects, teams, and individual users.
Which means last year’s budget, or even last quarter’s, doesn’t give finance a reliable baseline for what’s coming. And the category is still accelerating. 81% of the leaders SpendHound surveyed expect their AI budgets to increase again in 2026, more than any other software category.
Ownership hasn’t caught up either. 22% of teams said there is no single owner for AI spend, and around half of organizations still can’t point to measurable ROI. For most, it’s simply too soon to tell, which makes visibility into usage and cost more important, not less.
The quotes Sam shared from recent customer calls made that point loud and clear:
The natural follow-up question, and one that came up in the chat: don’t the OpenAI and Anthropic dashboards already show this? Sreesh’s answer was that they’re great for high-level visibility into a single platform, but they run into four limits.
Before the demo, Sreesh laid out where AI Spend Visibility is headed: one place to track every dollar of AI spend. That means foundational model spend pulled directly from OpenAI and Anthropic, the next layer of coding tools like Cursor, Replit, and Lovable, AI spend routed through cloud providers (Bedrock, Vertex, Foundry), and the AI SKUs that traditional SaaS vendors are now adding to their contracts on usage- or token-based pricing.
What’s live today is direct integrations with OpenAI and Anthropic. Once connected, the AI Spend page populates for any time period, defaulting to month to date, quarter to date, and year to date, with custom ranges available. From there you get total spend and provider-level breakdowns, projected monthly spend (projected annual is landing in the next week or two), month-over-month growth, and a consolidated spend chart you can filter by provider.
It’s the same principle behind SpendHound’s broader SaaS and AI spend management approach: one system of record instead of a tab for every vendor.
The more useful half of the demo was everything below the summary cards, which is the detail the native dashboards make hard to reach.
For both OpenAI and Anthropic, you can see your top five workspaces and projects by cost, then open any one of them for token volume, spend, and active users, all filterable by time frame. There’s a parallel view for API keys and their token volume. If your team uses workspaces, projects, or API keys to separate internal from external work, or one team from another, that structure becomes your cost attribution.
Spend by model breaks your costs down model by model over time. Sreesh pointed out how visible model launches are in the data: when a major new model shipped, SpendHound saw daily token consumption spike 20 to 50% across customers, drop off when access was restricted, then rise more gradually when it came back.
The most forward-looking piece of the demo was the first drill-down page, in beta for customers with an Anthropic API plan or Claude Enterprise.
Because Claude Code APIs expose more than cost and usage, the tab reports on productivity too: total lines of code generated by AI versus what’s accepted or rejected, and the commits and PRs that spend produced.
Sreesh’s read on the acceptance rate was useful. If it’s flat or climbing, engineering is leaning on the tool and trusting its output. If it’s falling, the generated code may not be pulling its weight, and how the tool is used deserves a second look.
The same logic applies to cost efficiency metrics like cost per commit. Trending down is a good sign. Trending up may mean the team is reaching for the most performant model when a cheaper one would do the same work.
“When we think about usage at the individual level, we want to make sure users are actually using models effectively, that deliver the right business outcome at the right price point.” — Sreesh Reddy
The clearest theme from the Q&A was coverage. SpendHound expects coverage to expand to Gemini, GitHub Copilot, and Cursor within the coming months, with early access opening soon for customers using Cursor.
Also in flight: cost attribution tags so you can categorize spend as R&D versus general internal usage, anomaly detection that alerts you to unexpected spikes by provider or model, and a dedicated AI Spend user role, so you can give engineers or executives the AI module without exposing the rest of your software spend.
On the token limit question a customer raised, the goal is to forecast how soon you’ll run out of contracted tokens the way SpendHound already tracks available seats. Provider APIs don’t reliably expose contracted allotments yet, so the likely path is parsing an uploaded contract or letting you set the allotment manually.
Seeing your AI spend by provider, model, project, and user solves an important part of the problem. It still doesn’t tell you whether you’re getting a good deal, which is where the rest of the platform comes in. In-Platform Benchmarking shows how your SaaS and AI pricing compares against real contracts from more than 1,300 companies, including SKU-level pricing where available.
Deal Grader brings that benchmark signal into the application catalog and admin dashboard as a grade, so you can spot a poorly priced major vendor or confirm an upcoming renewal is already competitive without opening every contract one at a time. It applies to AI tools too.
From there it’s about action. Every SpendHound customer has unlimited access to the Procurement Expert team to interpret a benchmark, identify leverage, review contracts, and build a negotiation plan. For teams that would rather hand it off, Managed Negotiation puts SpendHound in the vendor-facing seat on a success-fee basis.
AI spend is going to keep growing and keep getting harder to see. If you want the full picture of yours, book a demo with the SpendHound team, or if you’re already a customer with company admin access, log in to access the module.
Book a demo below and we'll get you set up with our team.
