Predictably Unpredictable: Top Vendors Driving Unpredictable Spend Variance (and How to Tame It)

Usage-based pricing creates spend variance where actual costs exceed contracted amounts, with AI-native tools like Cursor growing spend 4,000% year-over-year and OpenAI contributing dramatic budget swings. Analysis of $18 billion reveals vendors including Google, Twilio, AWS, Anthropic, and MongoDB drive highest variance through consumption-based models lacking governance controls. Six tactics tame variance: track spending live with real-time alerts, forecast using historical patterns with 10-30% buffers for volatile vendors, build contract flexibility with price caps and renegotiation clauses, automate spend tracking triggering notifications at thresholds, bring procurement into operational planning before usage spikes, and schedule regular contract reviews rather than waiting for renewals.

Remember pay-as-you-go cell phone plans? That was usage-based pricing in its simplest form. It was a smart way to stay flexible, until a heartfelt three-hour call turned into a $500 bill.

Today’s software contracts aren’t so different. Usage-based models offer real advantages: you scale with demand, avoid overcommitting, and pay in proportion to value, which is especially helpful as AI tooling matures and vendors refine how they price around consumption.

But if your usage outpaces tracking without clear guardrails - or contract terms aren’t built for scale - your budget will spiral. 

This is spend variance - the distance between contracted spend and actual spend - and based on our analysis of over $18B in spend, it's a major problem in procurement. More importantly, it's one that most finance and procurement teams are still fighting reactively, when the battle needs to be won months earlier.

Who Are The Top Vendors Contributing to Unpredictable Spend Variance?

While all vendors exhibit some spend variance, here's a full list of top highest-variance culprits we're seeing across the market from our data right now:

Notice how these vendors aren't obscure tools hiding in your long tail but sticky platforms your teams use every day.

• Google tops the list by both spend variance and shadow spend frequency. Cloud auto-scaling, Workspace seat drift, and Maps API consumption create compounding variance that's genuinely hard to model without real-time visibility.
• OpenAI and Cursor are the newest additions with the most dramatic trajectories. Cursor grew spend by over 4,000% year-over-year in our dataset. Anthropic grew by roughly 2,000%. These aren't rounding errors — they're entire budget categories that didn't exist two planning cycles ago, now showing up as material variance drivers.
• Twilio and AWS remain perennial high-variance vendors. Twilio's messaging costs scale with product usage in ways that aren't always captured in the original use-case model. AWS continues to generate the most acute version of this problem: a 3-day compute spike following a feature release can create 6 months of budget repair.
• LinkedIn, MongoDB, JetBrains, and Navan round out a list that spans marketing, data infrastructure, developer tooling, and travel. What they share: variable pricing structures, usage patterns tied to business activity rather than fixed license counts, and limited native controls for spend governance.

Here's what connects all of them: a tool that's easy to adopt without approval is equally easy to expand without approval. That's a shadow spend problem, a variance problem, and a procurement problem all at once.

AI Makes The Spend Variance Problem Worse

Spend variances have always existed - inevitable to say the least. But something has changed in the past 18 months that has made it materially harder to manage: AI.

The tools driving the highest spend variance in 2025 all share a common trait in using consumption-based pricing models. The flexibility is nice to have, but the invoice often isn’t.

What's new is the speed at which AI-native tools are scaling inside organizations. Across our customer base, AI-native spend grew 94% year-over-year for mid-market and enterprise companies. For SMB, it grew 24%. And when a tool with token-based or credit-based pricing suddenly becomes core infrastructure for a certain team, finance doesn't always get the memo until a quarter-end reconciliation turns really uncomfortable.

The problem compounds when you factor in how vendors are changing their pricing structures. Over the past 12-18 months, we've tracked what we’re calling the AI Tax: AI-driven price increases of 20-37% at renewal, far exceeding the typical 3-9% annual uplift. One of the four primary tactics we're seeing: Credit-Based Obfuscation, where vendors shift to consumption "credits" that make it genuinely hard to know what you're spending until after you've spent it.

The result? Businesses are now managing variance on both ends - unplanned overages from usage-based tools and aggressive repricing from legacy vendors leaning into AI as a justification for double-digit uplifts.

The True Cost of Variance Goes Beyond the Invoice

Think about it like this: a 20% overage on a $2M contract is $400K. That's three senior hires, a significant marketing campaign, or several months of runway depending on your stage and scale.

And when spend variance hits, it lands in the CFO's lap at the worst possible time: during a board review, a fundraise, or an annual planning cycle where your FP&A model is suddenly explaining itself.

Variance undermines trust. It creates the impression that finance doesn't have visibility into the business. It puts procurement in a defensive posture when they should be playing offense. And it puts pressure on the entire organization to over-correct (i.e. freezing spend, delaying approvals, or building in such aggressive buffers that you're actually over-budgeting for tools that don't need it).

The forecasting challenge is real and asymmetric. You can negotiate a $100K contract and end up with a $140K bill. Or you can over-commit and watch utilization stay at 60%. Neither outcome is acceptable. Both are common.

According to our analysis, fewer than one-third of companies can currently tie AI investments to measurable P&L impact. That’s partly because efficiency gains are captured by employees rather than converted into redeployable business capacity, and partly because the spend models haven't kept up with the pricing models.

How to Play Offense: Six Tactics to Tame Spend Variance

Finance and procurement teams need to build systems to navigate variance and volatility.

1. Track Variance Live: Use a tool that automatically spots overspending, underspending, and shadow spend in real-time. Waiting for the month-end close is too late.
   
   
2. Forecast Based on Real Patterns: Use historical data to identify typical usage arcs. Pair that with predictive analytics to generate a more realistic baseline for future consumption. (Pro Tip: Create upper and lower bound models for high-variance vendors. Forecast a 10–30% buffer based on past volatility.)
   
   
3. Build Flexibility into Contracts: Add price caps, volume tiers, and renegotiation clauses. (Example: moving Twilio off pure usage to a tiered model with thresholds.)
   
   
4. Automate Spend Tracking: When usage creeps beyond the baseline, someone should know immediately. Set up notifications when spend nears limits or diverges from normal patterns. Don't wait for the invoice to see the problem. (Especially helpful for AWS, where 3-day spend spikes often result in 6-month budget damage.)
   
   
5. Bring Procurement and Ops In Early: Some contracts require cross-functional stewardship. If GTM changes are about to trigger usage increases, bring procurement to the table before the spike. The earlier procurement is looped in, the better the contract and operational alignment.
   
   
6. Schedule Contract Reviews: Don't wait for renewals to examine performance. Usage-based contracts should be reviewed over regular checkpoints. Build this into your cadence.

Guess What? Negotiation Can Work Too

We’ve found that more strategic negotiation works, even on AI-native tools. It just requires a different strategy.

For fast-growing vendors like Cursor, Anthropic, and Clay, the standard discount conversation is a dead end. These vendors have pricing power right now and they know it. Their contract counts grew 400-650% year-over-year.

Discounts aren't the lever here - price caps, uplift protection, and usage guardrails are.

For legacy SaaS vendors applying the AI Tax, our data shows that negotiation reduces initial uplift asks by roughly 55% on average. Vendors were asking for 20-37% increases; companies that engaged early and pushed back landed at approximately 12%. The delta between those outcomes is material at any spend level.

One counterintuitive finding worth flagging: in 2025, the deepest discounts came from short-term contracts (31.9%), while 12-24 month deals averaged just 26.3%. This is the inverse of the conventional wisdom that longer terms earn better pricing. 

The hypothesis: vendors know that AI is rapidly changing their product roadmap, so they're willing to discount short-term deals, betting they can recapture on uplift at renewal. If that's true, the implication for buyers is significant - you may have more leverage in a shorter commitment than you think, but only if you're prepared to revisit and renegotiate before the vendor resets the baseline.

Finance's Job: Build the Muscle to Handle Variance

Think about your biggest cost centers: people and software. For most modern businesses, 70%+ of non-headcount spend touches some form of variable pricing.

This means your FP&A model is only as good as your ability to:

• Normalize volatility
• Plan for upper limits
• Course correct mid-quarter

Here's a better approach to treat budgeting:

• Scenario planning: Build plans that flex based on usage sensitivity.
• Input-output mapping: Understand what drives cost in tools like Datadog or OpenAI (and who owns those drivers).
• Capital allocation by confidence level: Allocate more precisely when you have better controls. Use buffers where you don't.

The best teams treat usage-based contracts like strategic bets, not reactive costs. They know which vendors are price elastic and where the spend cliffs are.

Procurement's Job: Move From Gatekeeper to Growth Partner

Usage-based pricing requires a higher level of procurement orchestration:

• You need smarter intake processes
• You need data-sharing across finance, IT, and engineering
• You need muscle memory for how to renegotiate early

Too many teams delay procurement involvement until the renewal. By then, your leverage is gone, your budget is off, and you're on your back foot.

Instead:

• Involve procurement months before during the sourcing stage
• Utilize price benchmarks to see where others are landing
• Understand vendor-specific negotiation levers to get better terms
• Schedule monthly spend reviews with internal stakeholders

Architect Around Usage-Based Vendors and Variance

Spend variance reflects how modern software is priced and consumed, and it's only growing as AI becomes core infrastructure rather than experimental spend.

Your top 10 vendors still command nearly three-quarters of your software budget. That concentration hasn't changed even as AI tools proliferate. What has changed is the pricing complexity layered on top of those relationships.

Build the muscle to anticipate it, contain it, and negotiate around it.

For more data like this, grab our full Spend Report here.