Every team that needs forecasting at scale ends up in the same conversation. Should we build our own forecasting engine? Should we buy a planning platform? Should we wire in a forecasting API and own the layer above it? The conversation usually runs for two months and lands wherever the loudest opinion sits, which is rarely the right place.
There is a clean decision framework hiding underneath. Three questions, and the answers determine which of the three paths fits. (If your current engine is a spreadsheet, start with Geneva Forecast API vs Excel — it covers exactly where spreadsheet forecasting breaks before you ever reach build vs buy.)
The three options, briefly
Build means your team writes the forecasting engine. Time series models, parameter tuning, seasonality detection, evaluation, monitoring — all of it owned in-house. The reference implementation is a small team running statsmodels, Prophet, or a custom ensemble inside an internal service.
Buy means licensing a planning platform — SAP IBP, o9, Anaplan, Kinaxis, Blue Yonder. The forecasting engine is one component of a larger workflow product that includes consensus planning, scenario management, governance, and audit. The deployment is months, the contract is large, and the engine is married to the platform.
Embed means using a forecasting API as the engine and owning the layer above it — the data pipeline, the UI, the workflow, the integration. The reference architecture is one HTTP call from your existing system to a forecasting service that returns the forecast and confidence interval.
Question 1: Is forecasting the differentiating capability?
If your product or business sells better forecasts to customers as the value prop, build it. A forecasting accuracy company, a quant trading firm, a niche industry forecaster — the engine is the product, and outsourcing the engine outsources the differentiation.
If forecasting is a supporting capability inside a larger value chain — demand planning at a CPG, capacity planning at a SaaS company, revenue forecasting at a B2B sales org — do not build. The engine is not what your customers pay for, and the cost of building and maintaining a competitive engine is large enough to crowd out work on the actual differentiator.
This filter eliminates ninety percent of build conversations in five minutes.
Question 2: Do you need the surrounding workflow?
If you need consensus forecasting across demand, supply, finance, and commercial, with audit trails, approval chains, scenario versioning, and a planner UI — buy. This is what enterprise planning platforms exist to do.
If you already have a workflow — a planning sheet that works, a BI tool, a custom internal app — and just need a better forecast feeding it, do not buy. The platform’s value is the workflow you already have, and you would pay for it twice.
Question 3: What is the engineering cost of integration?
Embed only makes sense if the integration cost is small. The right shape is one API call from your existing system on a schedule, with the response written back to wherever the planner reads it. If that integration costs more than a few engineering weeks, you are either embedding the wrong API or your data plumbing is the actual problem.
The framework in one sentence each
- Build when forecasting accuracy is the product you sell.
- Buy when you need the planning workflow more than the engine.
- Embed when you have the workflow and need a better engine inside it.
The vast majority of mid-market companies — and a meaningful share of enterprise companies that have a working planning process — are in the embed bucket and do not realize it because the build vs buy framing skips the third option.
What goes wrong when teams pick incorrectly
Build when you should embed. Six months in, the team realizes they have built a worse version of what an API would have given them in a week. Forecast accuracy in production is below the API benchmark, the team is maintaining infrastructure instead of solving business problems.
Buy when you should embed. Eighteen months in, the platform is live and the forecast accuracy is roughly the same as the legacy forecast it replaced, because the platform was sold as a forecasting accuracy improvement when it is actually a workflow improvement.
Embed when you should buy. The forecast engine works fine but the planning process around it is held together with spreadsheets and Slack. Forecast accuracy keeps improving and inventory keeps misbehaving because the consensus process is broken.
The hidden hybrid pattern
Real architectures often combine two paths. The most common pattern at maturing mid-market companies: embed an API for the forecasting engine, buy a planning platform for the workflow, and use the API as the engine inside the platform — either replacing the platform’s native forecasting or running alongside it on specific workflows where it outperforms.
When embed means Geneva
Geneva is designed for the embed pattern. Forty-five algorithms, automated selection, confidence intervals, sub-200ms latency, Python SDK, REST API, Claude MCP server. The free tier runs without a credit card. The integration cost is genuinely one API call.
For teams working through the build vs buy vs embed decision and wanting an outside read, Contact RoadMap.
Leave a Reply