How We Built a Conversational Long-Term Planning Platform for Active Living

Why Long-Term Planning Is So Hard

Long-term planning is one of the most important things a finance team does, but it's also one of the hardest to scale. At Snowflake, we had to build a 10-year forecast for a business that had grown incredibly complex: more than 40 legal entities, each with over 100 cost centers, and hundreds of spending categories. We needed that granularity because our long-term plan wasn't just a finance artifact anymore. It had become a dataset that other teams relied on.

The tax team didn't just want a high-level expense number. They needed to see the split between goods and services, the legal entities involved, the jurisdictions, and the operational details that mattered for their work. Treasury needed a cash view. People planning needed headcount assumptions. Executives wanted to understand the trade-offs between growth, margins, investment, and free cash flow. So the model had to serve all of them at once.

Like many companies, when our business outgrew our existing tools, we did what finance teams often do: we built a massive Excel model. It worked. But over time, it turned into the monster that most big financial models become. We kept adding tabs, patching in new formulas, and piling new logic on top of old logic to keep up with the changing business. The model became more valuable, but it also became harder to maintain, govern, and scale. That was the first thing we had to fix.

From Spreadsheet Monster to a Real Planning Platform

A little over a year ago, we rebuilt our long-term planning model on Snowflake, using Streamlit as the UI layer so analysts and executives could interact with the forecast directly. That became Snowplan, our internal long-term planning app. We didn't want to build a dashboard. We wanted a real planning platform, one that felt intuitive to finance users but had Snowflake's scalability, governance, and compute underneath.

In Snowplan, analysts update assumptions through an editable Streamlit interface. Those changes write straight back to Snowflake, the model runs, and the new outputs show up in the app immediately. No more broken formulas. No more saving files back and forth. No more wondering which version is the source of truth. This architecture changed the whole planning flow.

Instead of maintaining a giant offline workbook, we now have an app that connects to the actual source data of the business, with governance built in. Actuals flow into the model automatically, so nobody has to spend hours updating files. Assumptions can be versioned, scenarios can be compared, and different roles can work in the same platform at whatever level of detail they need. Individual contributors get granular input pages, assumption management, scenario creation, and version control. Directors and managers get visibility into logic and assumption changes for review and approval. Executives get a consolidated P&L, free cash flow, and key scenario views.

That matters because long-term planning is never just a modeling exercise. It's also a way to align the organization. The more time finance spends maintaining the model, the less time they have to actually sit with the business and think through strategy.

Building the Model Where the Data Lives

The most important decision we made was to build the model right where the data already lived. Because Snowplan runs on Snowflake, it's naturally connected to our raw data sources and our governed data models. We don't have to manually update actuals or reconcile offline pulls. The model sits in the same environment as our financial data, permissions, logic, and history.

That brings several advantages. First, the model can really scale. A forecast that spans 10 years, across entities, cost centers, spending categories, headcount, revenue, balance sheet, and free cash flow, generates a ton of data. That's exactly the kind of workload Snowflake is built for. Second, it's easier to govern. Access is managed through Snowflake's role-based permissions and row-level security, so different roles only see what they should. Executives don't need the same interface as analysts, and analysts don't have to export separate versions for every stakeholder. Third, the same platform can support people planning, equity modeling, treasury cash forecasting, hedging, legal entity forecasting, COGS planning, and M&A scenario analysis. That's the bigger story. Snowplan isn't a one-off planning app. It's becoming a financial planning platform.

CoCo Makes Scenario Planning a Conversation

Streamlit made Snowplan scalable and usable. Snowflake CoCo made it conversational. Before CoCo, Snowplan already gave us a better way to manage long-term planning. Analysts could update assumptions, run different scenarios, and compare outputs. But you still had to know where to go, which assumption to tweak, and how to understand the downstream impact of those changes.

CoCo changed that interaction model. Now, instead of clicking through pages of assumption configs to find an answer, I can just ask a question in plain English. I can ask CoCo to compare two versions of the forecast and summarize the key drivers. I can ask what changed between the version we showed the board last year and the latest one we're preparing now. I can ask about the net impact of those changes, the key drivers of margin expansion or dilution, and which assumptions deserve the most attention.

That's incredibly powerful in executive planning. Because when you're preparing for a board discussion, the real question is rarely 'Can you give me the latest forecast numbers?' It's 'What changed, why did it change, and how does that shift our narrative?' CoCo compresses what used to be a manual comparison into a conversation. The value isn't just speed. It means finance can keep iterating while the strategic discussion is still happening.

A Real Example: Scenario Planning Around a Potential Tax Change

One of the best examples is scenario planning around a potential tax change. In the old world, you'd start with a meeting. You'd talk to the tax team, define the affected sales, pull the data, build assumptions, update the model, review the output, create sensitivity tables, and then decide who else needed to be involved. With CoCo in Snowplan, that process becomes much smoother.

I can start by asking CoCo to summarize the potential tax change. Then I can ask it to create a new forecast version that assumes the tax change goes through. That immediately turns into the kind of back-and-forth finance teams have in meetings: Is the tax passed on to customers or absorbed as a margin hit? What percentage can realistically be passed through? Which sales will be affected? What's the impact on revenue, gross margin, operating margin, and free cash flow?

Because the analysis sits on Snowflake tables, CoCo can identify which sales would be affected, output the financial impact, and show the key metrics behind the numbers. It can also build sensitivity tables that show how operating margin gets diluted under different pass-through percentages. Just as important, it can flag risks and caveats. For example, the first-order model might not include the extra indirect costs of supporting compliance or new reporting obligations. That kind of reminder is exactly what a good finance partner raises before anyone treats a scenario as a conclusion.

CoCo can even help draft next steps, like an email to the tax team summarizing the analysis, key assumptions, open questions, and decision points. So the system isn't just giving you a number. It's helping you frame the problem, identify subject matter experts, and keep the process moving. At that point, it's not just AI-assisted modeling. It's AI-assisted planning.

Why This Matters for Finance Teams

Finance teams are constantly asked to answer strategic questions faster than traditional planning processes allow. What if we accelerate growth? What if we open a new office? What if cloud costs improve by 25 basis points? What if compensation inflation runs hotter than expected? What if a tax or regulatory change hits certain sales? What if we reallocate investment across functions? These aren't theoretical questions. Executives throw them out in real time.

The problem is that traditional planning tools and big spreadsheet models aren't designed for that level of iteration. They're built to produce a plan, not to support an ongoing strategic conversation. By building Snowplan on Snowflake with Streamlit, we made a planning platform that scales with business complexity. Adding CoCo turned it into a conversational system. That combination changes what finance can actually do. Instead of spending time updating actuals, maintaining formulas, reconciling scenarios, or manually comparing versions, finance can spend more time on the work that really matters: challenging assumptions, aligning executives, weighing trade-offs, and shaping long-term strategy.

Trust Is Everything in AI-Driven Planning

For finance teams, conversational planning only works if the numbers are trustworthy. That's why architecture matters. CoCo isn't generating a forecast out of thin air. It's interacting with the same governed data, assumptions, and logic that power Snowplan. When it compares versions, explains drivers, or creates scenarios, it's using the Snowflake data models and planning logic we already rely on. Every scenario can be versioned, every change can be reviewed, and access controls follow the app's role model. Analysts and executives can see what changed, understand it, and roll forward or back as needed.

That's a key distinction. We're not asking leaders to trust a black box. We're using AI to operate a well-governed planning platform where data, business logic, permissions, and outputs are visible, explainable, and auditable. That's what makes AI truly useful in corporate finance.

From Planning Tool to Strategic Platform

The most exciting part of Snowplan is that it's already outgrowing its original use case. Once we moved the model to Snowflake, the architecture became reusable. The same foundation now supports, or can support, multiple financial planning workflows: headcount planning, equity modeling, treasury cash forecasting, hedging, legal entity forecasting, COGS planning, and M&A scenario modeling. That's the advantage of building it as a platform rather than a one-off app.

Each new planning workflow can reuse the same governance foundation, connect to the right data sources, and expose a finance-friendly Streamlit interface. And with CoCo, each one becomes easier to query, adjust, and explain through natural language. I think more finance teams will follow a similar path: first, move the model to where the data lives. Second, build an intuitive app layer for users. Third, use AI to make the planning process conversational.

The Real ROI

The real ROI of Snowplan isn't making finance teams 'more technical.' It's giving them more time for judgment. Long-term planning shouldn't be about maintaining a giant workbook. It should help teams understand where the business is going, help leaders decide where to invest, how to balance growth and profitability, which risks are emerging, and which trade-offs matter most. Snowflake and Streamlit gave us a platform that makes long-term planning scalable, governed, and connected to real-time data. CoCo is making it faster, more interactive, and more strategic.

That's the change. Finance teams can spend less time updating models and manually tweaking assumptions, and more time iterating with the executive team on the company's long-term strategy. For FP&A teams, that's the real value of AI in planning. It's not replacing the finance function. It's taking away the manual labor that slows finance down, so they can focus on what they should actually be doing.