Priceline helps millions of travelers book hotels, flights, rental cars, and packages every day. Its FP&A team turns that nonstop activity into the forecasts behind every investment decision, supporting dozens of planners across marketing, product, call center operations, and corporate finance. After four years on a legacy planning platform, the team needed an environment that could hold the whole P&L in one model and let them test an idea without taking the rest of the day to do it.
When the Tool Became the Limit
Priceline forecasts weekly, distributes those values down to the day, and rolls them back up to the month. The legacy planning platform had no formulaic way to connect those three cadences in one place, so the team ran weekly, daily, and monthly as separate models and moved data between them with manual imports.
Capacity made it worse. The platform capped each model at 125 gigabytes, and the weekly top-line model alone was already at 100. Across an environment of roughly 400 gigabytes, there was no option but to split the work five ways.
Every transfer between those models took 10 to 15 minutes, and a single input took seven seconds to calculate inside one model. Real-time forecasting, the thing the platform had been bought for, was never actually available.
From a formulaic perspective and from a capacity standpoint, we had to separate our models into five separate disconnected models, which required button pushes to transfer between them. That was a very time consuming process. David Hutchison, Director of Data Analytics and FP&A, Priceline
Why Scenario Planning Was the Breaking Point
Versions counted against model memory. Adding one sandbox version grew every model by roughly 25 percent and degraded performance for everyone, so the team could only ever afford one at a time, and standing that one up across every module was its own project.
The cost landed every forecast cycle. When the CFO asked for a sensitivity, a two point lift in hotel ADR or a higher take rate on air, the team changed the driver, pushed it through five models, exported the result to the consolidation tool because the legacy platform could not hold a fully rolled up P&L, then manually reverted everything to restore the baseline. One scenario took a full day. Two took most of a week.
After four years on the platform, with the contract coming up for renewal, the brief from Priceline's VP of Finance was specific. One interconnected model where a change at the lowest level, marketing gross orders, flows all the way to the final P&L with no button pushes. Calculation in under 30 seconds. And unlimited sandboxing with no performance penalty.
Why the Version Architecture Decided It
Fintastic entered the evaluation late, behind two established platforms already in use at sister companies in the group. Knowing it was a startup, the team took the meeting without much expectation.
From that first demo, we were blown away with what they showed us. We gave them a test case, and within two days they were able to replicate something that worked exactly as designed. David Hutchison, Director of Data Analytics and FP&A, Priceline
The deciding difference was structural. In the platforms Priceline knew, versions lived inside the model and consumed memory, which is why every sandbox came with a performance tax. In Fintastic, each version is its own environment. The team can duplicate the baseline and keep adding sandboxes without watching size or speed degrade.
The second piece was variance analysis. Pick any two versions on a dashboard and the comparison populates across the entire view automatically, in both absolute and percentage terms. In the legacy environment, roughly half of every build went into hand-coding variance structures, one view at a time. For a team judging how fast it could build, not just how fast the tool could calculate, that settled it.
In my finance career, it's rare to see an interconnected platform of this scale support iterative scenario planning without sacrificing speed or reliability. Marc Culver, VP of Finance, Priceline
A Live Feed From BigQuery and OneStream
Priceline's environment runs on two connection points. Daily transactional data comes from BigQuery. Monthly actuals come from OneStream, the parent company's consolidation tool. Between them, the models carry 10 to 20 million rows with heavy dimensionality behind them, so the question during evaluation was never just speed. It was whether a platform could consume that volume efficiently.
Some feeds import once a day. Others refresh every 30 minutes, so a change to a BigQuery structure appears in the forecast within half an hour with no manual intervention and no one pressing a button.
Fintastic's Delivery Team built all of the retrievals, which kept the setup off the FP&A team's plate entirely.
Opening the Model to the Whole Business
Between 40 and 50 people at Priceline work in Fintastic, with around 20 concurrent users at the peak of a forecast cycle. FP&A is the core, but marketing runs its own modeling with five to seven people entering inputs directly. The travel insurance team works in the tool. Call center managers own their own models and dashboards. Another 10 to 15 OpEx owners manage their own line items.
None of it routes through FP&A first, which is the point. Each group owns its own inputs and its own dashboards, and the FP&A team spends the forecast cycle on analysis rather than on collecting numbers from everyone else.
From a user standpoint, Fintastic has been a home run across the board. They can go through any of the versions that have been created, instantly see the impact, and compare across versions. David Hutchison, Director of Data Analytics and FP&A, Priceline
An AI Agent That Documents the Model and Hits the Number
The first thing David did with the Fintastic AI Agent was map and document the entire model, at a high level and in detail, covering how every element of the business interconnects.
Then he asked it to go back through each area and recommend improvements. It came back with inefficient build patterns and key inputs the team had missed, recommendations they are still working through to optimize the model.
The team has also started running complex goal seeks. David asked the agent to add 10 million dollars of incremental revenue to the 2026 bottom line EBITDA and gave it a weekly driver to work with. The agent distributed the change across every weekly input and landed the number exactly, replacing a task that would otherwise mean typing hundreds of weekly values by hand.
