We Turned a 10-Year Spreadsheet Nightmare into a Chat with Our Data

The Spreadsheet That Ate Our Week

Long-term planning is the kind of work that quietly devours a finance team's calendar. It's not glamorous, but it props up every strategic decision. At Snowflake, the FP&A team found themselves staring at a 10-year forecast that had morphed into something monstrous. Forty-plus entities, each with over 100 cost centers, hundreds of expense categories. The model had grown so complex that it wasn't just a finance artifact—it was a dataset that tax, treasury, HR, and the executive team all leaned on.

Tax needed breakdowns by goods versus services, legal entity, and jurisdiction. Treasury wanted the cash view. HR needed headcount assumptions. Execs wanted to see trade-offs between growth, margins, and free cash flow. One model had to serve all of them, and like any good finance team faced with a tool that no longer fits, they did what we all do: they built an enormous Excel workbook.

And it worked—for a while. But it became a Frankenstein. New tabs stacked on new tabs. Formulas patched onto formulas. Logic layered over logic. Every day, the model grew more valuable, but maintaining it turned into a full-time job in itself.

The Real Cost No One Talks About

Here's the thing about spreadsheet models: they demand constant feeding. Actuals had to be manually updated. Files saved, renamed, emailed. Versions multiplied like rabbits, and nobody could remember which one was the source of truth. The finance team wasn't analyzing the business—they were pushing numbers around a grid.

That's the real cost of a broken planning process. It's not the software licenses or the occasional formula error. It's the hours—days, honestly—that get sucked into maintenance instead of strategy. The team at Snowflake knew this all too well. So they decided to do something about it.

Rebuilding on Snowflake: A Platform, Not a Dashboard

About a year ago, they made a bold move. They scrapped the Excel monster and rebuilt the entire long-term planning model on Snowflake, with Streamlit as the UI layer. The result was Snowplan, an internal application that let analysts and executives interact with the forecast directly.

This wasn't just a dashboard. It was a real planning platform. The interface felt familiar to finance users, but underneath, it had Snowflake's scaling power, governance, and compute. Analysts could update assumptions in an editable Streamlit view, and those changes would write straight back to Snowflake, trigger the model, and refresh the outputs in real time. No more broken formulas. No more round-tripping files. No more guessing which version was current.

The architecture changed how planning worked. Instead of maintaining a giant offline workbook, they had an app connected to the actual source data, with governance baked in. Actuals flowed in automatically—no more hours spent updating files. Assumptions could be versioned. Scenarios could be compared. And different roles got different views: associates saw granular input pages, managers got visibility into logic changes, and executives saw the consolidated P&L and free cash flow.

Why the Platform Approach Saves Time

Building on Snowflake meant the model lived where the data already lived. That sounds obvious, but it's a big deal in practice. The team no longer had to manually pull actuals or reconcile offline data pulls. The model was embedded in the same environment as the finance data, permissions, and historical records. That made it easier to scale—a 10-year forecast across entities, cost centers, and expense categories is a heavy load, but Snowflake handles that kind of thing in its sleep. It also made governance simpler. Role-based access and row-level security meant everyone saw only what they should see. No more exporting different versions for different stakeholders.

But the biggest win was time. A planning process that used to take weeks of spreadsheet gymnastics could now be done in hours. The team could spend their energy on the actual work of planning—challenging assumptions, stress-testing scenarios, and aligning with the business—instead of babysitting a workbook.

CoCo Turns Planning into a Conversation

Streamlit made Snowplan usable. Snowflake CoCo made it conversational. Before CoCo, users still had to know where to click, which assumption to tweak, and how to interpret the downstream effects. That's a lot of institutional knowledge. CoCo changed the interaction model entirely.

Now, instead of navigating page after page of assumption configs, you can just ask. "Compare these two forecast versions and summarize the main drivers." "What changed between the plan we showed the board last year and the latest one we're preparing?" "What's the net impact on operating margin?" The answers come back as a conversation, not a pivot table.

This is powerful in executive planning because the real question is rarely "give me the latest numbers." It's "what changed, why, and what does that mean for our narrative?" CoCo compresses what used to be a manual comparison exercise into a back-and-forth dialogue. And because it's built on the same Snowflake tables and logic as Snowplan, the answers are grounded in the same governed data—not some black-box AI hallucination.

A Real Example: Tax Scenario Planning

One of the best examples comes from tax scenario planning. In the old days, this started with a meeting. Tax defines the issue, finance pulls the data, builds assumptions, updates the model, reviews the output, and then does sensitivity analysis. It's a slow, iterative process that usually takes days.

With CoCo, the flow is different. You can ask CoCo to summarize the potential tax change. Then you ask it to create a new forecast version assuming the change goes through. Immediately, the conversation starts: Is the tax passed on to customers or absorbed as a margin hit? What percentage can realistically be passed through? Which sales are affected? What's the impact on revenue, gross margin, and free cash flow?

Because the analysis is built on Snowflake tables, CoCo can identify which sales would be impacted, calculate the financial effect, and show the key metrics behind the numbers. It can also create sensitivity tables showing how operating margin dilutes under different pass-through assumptions. And it flags risks: a first-order model might miss the indirect costs of compliance and reporting. That's the kind of proactive thinking a good finance partner brings—and now it's automated.

CoCo can even draft an email to tax colleagues summarizing the analysis, key assumptions, and open questions. It's not just giving a number; it's helping move the process forward.

Trust Is the Real Foundation

None of this works if the numbers can't be trusted. That's why the architecture matters. CoCo isn't generating predictions out of thin air. It's interacting with the same governed data, assumptions, and logic that power Snowplan. Every scenario is versioned. Every change can be reviewed. Access control follows the same role model as the app. Analysts can compare before and after, understand what changed, and roll back if needed.

This is a crucial distinction. Finance leaders aren't being asked to trust a black box. They're using AI to operate a governed planning platform where data, logic, permissions, and outputs are all visible, explainable, and auditable. That's what makes AI viable for enterprise finance.

The Bigger Picture: From Tool to Strategic Platform

Snowplan has already outgrown its original purpose. The same foundation now supports multiple planning workflows: headcount planning, equity modeling, treasury cash forecasting, hedging, legal entity forecasting, COGS planning, and M&A scenario analysis. That's the advantage of building a platform instead of a one-off app. Each new workflow reuses the same governance, connects to the relevant data, and exposes a user-friendly Streamlit interface. And with CoCo, each one becomes easier to query, adjust, and explain in plain language.

The real ROI isn't that finance teams become more technical. It's that they get more time back for judgment. Long-term planning shouldn't be about maintaining a massive workbook. It should be about helping the company understand where it's going, where to invest, how to balance growth and profitability, and which risks are emerging. Snowflake and Streamlit gave the team a platform that scales and governs. CoCo made it fast and interactive. The result is that finance can spend less time updating models and more time working with executives on the long-term strategy.

That's what AI in planning should be about—not replacing finance, but removing the drudgery that slows it down.