How to build a demand plan from scratch

Most guides to demand planning are written for enterprise teams with dedicated software, data analysts, and a full S&OP process already in place.

This one isn't.

This is for brands that are starting from zero — maybe a spreadsheet, maybe some messy sales data, maybe a co-manufacturer who gives you 30 days notice and a minimum order quantity. The goal by the end of this guide is a working demand plan: a forward-looking view of what you'll sell, what you need to order, and when.

It won't be perfect. A demand plan is never perfect. But an imperfect plan built on real data is worth more than the most optimistic spreadsheet your marketing team has ever produced.

Let's build it.

Before you start: what you need to gather

Don't skip this step. The quality of your demand plan is a direct function of the quality of your inputs. Before you build anything, collect the following for every SKU you're planning:

Sales history

Two years is ideal. One year is workable. Less than six months makes statistical forecasting unreliable — you'll need to lean on judgment and industry benchmarks instead.

Get this at the SKU level, by month. Not blended revenue — actual units sold, by product, by channel if possible.

Current on-hand inventory

How many units of each SKU are physically in your warehouse right now.

Inbound purchase orders

Any orders you've already placed with your supplier that haven't arrived yet. These units are coming — treat them as inventory you'll have soon.

Lead time

The full end-to-end lead time for each SKU. Not just production time. Production, plus transit to your warehouse, plus any additional time before it's available to ship. If you're supplying retail, add transit to the retailer's distribution center and time from the DC to the shelf.

Most founders think their lead time is three or four weeks. End to end, it's usually seven or eight. Use the real number.

MOQ (minimum order quantity)

The smallest order your supplier will accept. This affects how you calculate order recommendations.

Safety stock

The minimum inventory level you want to maintain as a buffer against demand spikes or supply delays. If you haven't calculated this before, a reasonable starting point is two to four weeks of average demand depending on how variable your sales are and how long your lead time is.

Reorder point

The inventory level that should trigger a new order. Calculated as: average daily demand × lead time in days, plus your safety stock. When your inventory drops to this level, it's time to place an order.

Once you have all of this, you're ready to build.

Step 1: Clean and organize your sales data

Pull your historical sales data into a spreadsheet. You want one row per SKU, with monthly sales in columns going left to right across the page.

It will almost certainly come out of your system in the wrong format. Clean it:

• Remove cancelled orders, samples, and internal transfers
• Separate channels if you can (DTC, Amazon, wholesale) — demand patterns differ by channel and blending them makes your forecast less accurate
• Flag any months with unusual spikes or zeros that were caused by a one-off event (a stockout, a viral moment, a bulk wholesale order that won't repeat)

The cleaner this data is, the more accurate your forecast will be.

Step 2: Choose a forecasting technique for each SKU

This is where most guides lose people. The reality is simpler than it sounds.

There are a handful of statistical techniques commonly used for demand forecasting. Apply several of them to your historical data, then measure which one most accurately predicted what actually happened. Use that technique going forward for that SKU.

The techniques worth knowing:

Simple moving average

Takes the average of the last several months and projects it forward. Works well for stable, predictable SKUs with no strong trend or seasonality.

Weighted moving average

Same as above but gives more weight to recent months. Better when demand has been shifting — up or down — and you want the forecast to reflect recent momentum more than older history.

Weighted moving average with trend

Explicitly models whether demand is growing or declining and projects that trajectory forward. Useful for SKUs that have been consistently gaining or losing velocity.

Seasonality index

Models the seasonal pattern in your data — the months that are reliably higher or lower than average. Requires at least 12 months of data to calculate, and ideally two or more years. Most useful for products with clear seasonal demand (sunscreen, cold and flu remedies, holiday gifting).

Linear regression

Fits a straight line to your historical data and extrapolates it forward. Works well when demand has a consistent linear trend.

How to pick the right one:

For each SKU, run two or three techniques that seem like plausible fits. Then measure the MAPE — Mean Absolute Percentage Error — for each one. MAPE tells you, on average, what percentage your forecast was off from actual sales. Lower is better. The technique with the lowest MAPE for that SKU is the one to use.

In practice, a weighted moving average works well for the majority of SKUs. Seasonality index adds meaningfully when you have strong seasonal patterns and enough data to model them. Start simple and add complexity only where the data supports it.

For new products with no history:

You can't run statistical models on data that doesn't exist. Instead, use analogous products — how did similar SKUs in your portfolio or category perform in their first year? Build a range: a conservative case, a base case, and an optimistic case. Plan your inventory to the base case and hold a buffer for upside.

Step 3: Generate your 12-month forward forecast

Apply your chosen technique to each SKU and produce a monthly forecast for the next 12 months.

A few things to layer in on top of the statistical baseline:

Promotional lifts: If you have a confirmed promotional event — a retailer feature, a DTC sale, a Prime Day deal — add an incremental lift to the relevant months. Be conservative. Promotional lifts are almost always overestimated.

New distribution: If you're entering a new retailer or adding doors, add the expected volume for those new points of distribution.

Launches and discontinuations: Add new SKUs coming into the range. Remove or ramp down SKUs being phased out.

Do not layer assumption on top of assumption. Add one thing at a time, with a reason for each. The baseline statistical forecast is your anchor — every adjustment should require justification.

Step 4: Build the monthly inventory plan

Now connect your forecast to your inventory position.

For each SKU, for each month, calculate:

Projected available inventory:

On-hand inventory + inbound POs − forecasted demand

Add a 20 percent buffer to your forecast to account for normal demand variability. This isn't padding — it's a standard deviation adjustment that prevents you from running to zero every time demand comes in slightly above the model.

Units needed:

If your projected available inventory falls below your safety stock level, you need to order. The order quantity should be enough to bring you back above your safety stock, at a minimum your MOQ.

Order date:

Work backwards from when you need the inventory. If you need units available in March and your lead time is 45 days, your order needs to be placed by mid-January.

This is the piece most brands miss. The demand plan isn't just a forecast — it's a forward view of when to place every order for the next 12 months. That view is what turns a reactive ordering process into a planned one.

Step 5: Summarize into an order schedule

Pull your output into a clean summary: which SKUs need to be ordered, how many units, and by what date.

This is what you hand to your operations team, your co-manufacturer, or your buying contact. It gives them enough lead time to actually act on it. And it gives you the ability to see cash requirements coming — which orders are due in the next 30, 60, and 90 days, and what they'll cost.

Step 6: Maintain it every month

A demand plan built once and never updated isn't a demand plan. It's a historical document.

Every month, feed in your actual sales data and compare it to what you forecast. Calculate your MAPE for each SKU. If you were significantly off, understand why — and update either the model or the assumptions before you roll forward.

This monthly update is also when you incorporate new information: a changed promotional calendar, a delay from a supplier, a retailer that's ordering more or less than expected.

The plan should take a few hours to update each month once it's built. If it's taking longer than that, the model is too complicated or the data process isn't automated enough.

What good looks like

For a brand doing $5M to $15M in revenue, a working demand plan produces three things:

1. A 12-month forecast by SKU that reflects real demand patterns, not revenue targets.
2. A clear view of inventory against that forecast, including when you'll hit reorder points and by how much.
3. An order schedule that tells you exactly when to place each PO, so you're never ordering in a panic.

You don't need dedicated software to get there. A well-built spreadsheet handles this for most brands until they're managing 50-plus SKUs across multiple channels with complex promotional activity.

What you do need is clean data, honest assumptions, and the discipline to update it every month.

If you'd like to see what this looks like built for your specific products and channels, we're happy to walk through it.