Platform

Know what next year costs.

Where this month lands, what next quarter looks like, and what the bill becomes as usage grows. The month-end figure updates as soon as spend moves, so a month heading for five times normal says so in its first days.

How do you forecast AI spend?

Camaze forecasts from usage drivers rather than from a spend trend line. It separates volume from unit cost, models each workload on its own growth pattern, and applies the pricing actually in force including committed tiers and discounts. Output is a range rather than a single figure, and it can be run under scenarios: usage growth, a feature reaching general availability, a model change, or an optimization plan being executed.

The problem

Last month's number with a percentage on top is not a forecast.

AI spend does not grow smoothly, so extrapolating the total is close to useless. It moves in steps.

More detail

A feature reaches general availability and volume multiplies overnight. A workload switches model and unit cost halves while volume is unchanged. An agent ships and creates a category of spend that did not exist last quarter.

A forecast built on the aggregate line cannot see any of that coming, which is why AI is the number in the plan with the widest error and the least confidence behind it.

  • Step changes at launch, not gradual growth
  • Unit cost and volume moving in opposite directions in the same month
  • New workloads appearing between planning cycles
  • Prepaid credits that hide the real run rate until they are exhausted
Total AI spend All providers
This month to date$151,400+12.8% vs last month
Forecast, month end$168,200Range $161,000 to $176,000
Identified savings$61,90041% of run rate
Monthly AI spend over twelve months with a three month forecast Spend rises from about 41 thousand dollars to about 151 thousand dollars over twelve months. The forecast tail projects roughly 168, 187 and 208 thousand dollars over the next three months, inside a shaded confidence range. 0 $65k $130k $195k $260k FORECAST AugOctDecFebAprJunAugOct
Actual Forecast Confidence range

Illustrative product view. Figures are examples.

What we do about it

Forecast the drivers, not the total.

Each workload is modeled separately on the things that actually determine its cost, then the results are aggregated. That is what makes it possible to answer a question about one launch without rebuilding the whole model.

  • Volume and unit cost forecast separately, then combined
  • Per-workload growth patterns, including seasonality and weekly shape
  • Current contracted pricing applied, including committed tiers and discounts
  • Confidence ranges, with the drivers of uncertainty named
  • Known future events included: launches, migrations, contract renewals
  • Re-run on every change, so the month-end figure is never stale
Month end

Where this month lands, from the fifth onward.

An intra-month forecast is the most immediately useful output, because it is still actionable.

More detail

It updates as spend arrives, accounts for the weekly shape of your usage rather than assuming a flat run rate, and shows the range as well as the point estimate.

This is what makes day 5 worth anything. A workflow going to five times its normal rate on the third of the month moves the projected month-end total the same day, so the size of the problem is known while there are still 28 days to do something about it. Waiting for the invoice tells you the same thing three weeks after the decision point has passed.

When the projection crosses a budget, the alert fires against the forecast, not the actual, which is typically weeks earlier.

  • Projection updated as usage arrives, aware of weekday and weekend patterns
  • A 5x day visible in the month-end number immediately, not at close
  • Range shown alongside the point estimate
  • Per-team and per-project projections, not only the company total
  • Forecast breach alerts routed to the budget owner
Platform team budget Alert sent day 5
Budget burn-up for the month, crossing the limit before month end Actual spend runs above the even-pace line all month and the projection crosses the ninety thousand dollar budget on day 28, ending the month at about one hundred and four thousand dollars. 0 $30k $60k $90k Budget $90k Even pace Over on day 28 Day 1Day 8Day 15Day 22Day 31
Actual spend Projection Even pace

Illustrative product view. Figures are examples.

Scenarios

What happens if.

The questions that actually get asked in a planning meeting are conditional. What does this cost if the feature goes to every user instead of ten percent.

More detail

What happens if we move that workload to a cheaper model. What if usage doubles and we do nothing.

Scenarios can be built on any workload and compared side by side, so a decision to expand an AI feature arrives at the budget meeting with a cost attached rather than an assurance.

  • Rollout scenarios: from a pilot cohort to general availability
  • Model change scenarios, with the unit cost effect isolated from volume
  • Growth scenarios at several rates, with ranges on each
  • An optimization scenario showing the plan executed, for comparison
Scenario planning Monthly run rate
Five quarter AI spend forecast under three scenarios On the current trajectory monthly spend reaches about 447 thousand dollars. Rolling the feature out to everyone reaches about 662 thousand. Following the optimization plan holds it at about 238 thousand. 0 $175k $350k $525k $700k $447k $662k $238k NowQ1Q2Q3Q4Q5
Current trajectoryFeature rolled out to everyoneWith the optimization plan

Illustrative product view. Figures are examples.

Against plan

Variance you can explain in a sentence.

Every forecast is held against the plan you loaded, and variance is decomposed rather than reported as a single unexplained figure.

More detail

Volume against plan, unit cost against plan, and new workloads that were not in the plan at all are reported separately.

That is the difference between telling a board that AI came in over budget and telling them that volume was on plan, unit cost fell after a model change, and the overrun is one new workflow that shipped in month two.

  • Variance split into volume, unit cost and scope
  • Plan loaded from your existing budget, at your own level of detail
  • Rolling reforecast on your cycle, monthly or quarterly
  • Prior forecast accuracy tracked, so the range can be trusted
Scheduled reports 6 running
ReportRunsLands in
Monthly close pack1st of the monthExcel, emailed to finance
Chargeback journal1st of the monthERP export, journal-ready
Board AI summaryQuarterlyPDF and slides
Cost by teamEvery MondayGoogle Sheets, live
Raw cost and usageNightlySnowflake
Budget variance digestEvery FridaySlack, #finance

Illustrative product view. Figures are examples.

Put a real number in the plan.

Bring next year's AI assumption to a 30 minute session and we will show you what a driver-based version of it looks like.

In practice

Fitting into the budget process you already run

No new cycle, no new tooling for the FP&A team.

Step 1

Load the plan

Your existing AI budget goes in at whatever level you hold it: company, department or project. No restructuring required.

Step 2

Reforecast on your cycle

Camaze produces a rolling reforecast on your schedule, with variance decomposed and the largest movements listed first. It exports to Excel, Sheets or your warehouse.

Step 3

Plan the next year with scenarios

Annual planning runs on scenarios rather than a single line. The AI number arrives with its assumptions attached, which is what makes it defensible when it is challenged.

FAQ

Questions people ask

How accurate is the forecast?

Accuracy depends on how stable your workloads are, so Camaze reports a range rather than a single figure and tracks its own prior accuracy so you can see how much weight the range deserves. Intra-month projections are typically the tightest, because most of the month is already actual. Annual figures widen considerably, which is the honest answer for a market where model pricing changes several times a year.

Can it forecast a workload that does not exist yet?

Yes, as a scenario. You supply the expected volume and the workload shape, and Camaze prices it against current model pricing and comparable existing workloads. It is an estimate and it is labeled as one, but it is built on your own unit costs rather than a vendor's list price.

What happens to the forecast when we switch models?

Unit cost updates and volume carries across, because workload identity persists through a model change. You see the effect isolated from volume movement, and the forecast reprices from the switch date rather than restarting.

Does it account for our committed spend and discounts?

Yes. Contracted rates, committed-use tiers and volume discounts are applied to the forecast, including the point at which a commitment would be exhausted or a discount tier reached. Procurement uses the same output when sizing the next commitment. See procurement.

Can we export the forecast into our own model?

Yes. Forecasts export to Excel, Google Sheets, Snowflake and via API, at the level you hold your plan. Most FP&A teams keep their model and treat Camaze as the driver source.

Forecast the AI line properly.

Book 30 minutes and we will build a driver-based forecast on your own usage during the call.

5 minute setup. Read-only. No engineering time needed.