Production planning with AI: a practical guide for industrial SMEs

The owner of a small metalworking company described his Sunday night to me: sitting in the kitchen at eleven, building the week's production plan in a spreadsheet. He was cross-referencing by hand the confirmed orders, what he sensed would come in, the raw material stock and the machines' capacity. A lot of it was pure guesswork. If he produced too much, money froze in finished goods; if he produced too little, he could not deliver and let a customer down. And the following week, the spreadsheet and the guesswork all over again.

That scene is daily life for a good share of SMEs that manufacture. Production planning is a fiendishly difficult problem (demand that varies, limited capacity, raw material with lead times, orders that arrive at the last minute) and the serious tools for solving it — the big MRP systems — were designed for enormous industries, with prices and implementations that do not add up for an SME. So the SME ends up planning in Excel and on sheer effort.

The result is an expensive swing: weeks of excess finished goods followed by weeks of missed deliveries. Erratic demand is met with intuition, and intuition, however good, neither scales nor learns from past mistakes. Every plan starts almost from scratch.

In this article you will see why planning production by eye is so expensive, how a layer of AI on top of your ERP organizes the plan without a half-million-dollar MRP, a case of an industrial SME that stopped guessing, and the steps to implement it.

---

1. The problem: planning production by eye

1.1 Demand that will not sit still

The heart of the difficulty is demand. In an industrial SME it varies with seasonality, with large orders that land all at once, with customers who change their plans. Planning production demands anticipating that demand, and anticipating it by eye, week after week, is exhausting and imprecise. The owner's intuition is valuable, but it is not enough for a problem with this many variables.

1.2 The hidden cost of producing too much or too little

Every planning error costs, in one of two directions:

• Producing too much. Finished goods piling up: frozen capital, occupied space, obsolescence risk.
• Producing too little. Orders not delivered on time: annoyed customers, lost sales, urgent runs that blow up costs.

The hard part is that you are almost never wrong evenly: in an SME, surplus products and missing products coexist at the same moment. It is the same imbalance that shows up when reducing excess inventory, but on the production side.

1.3 Why the traditional MRP is not the answer

The textbook solution would be an MRP. But the big production planning systems were designed for industry at scale, with licence and implementation costs counted in the hundreds of thousands of dollars and projects measured in years. For an SME that is disproportionate: the cure costs more than the disease. That is why the SME stays in Excel, which is free but blind.

2. What AI does with production planning

2.1 Demand forecasting within an SME's reach

The AI layer starts with the hardest part: estimating demand. By crossing sales history, seasonality, orders in progress and commercial signals, it projects how much of each product will be needed. It is not a crystal ball, but it is vastly better than Sunday-night intuition, and it learns from each cycle's errors.

2.2 From forecast to plan

With demand estimated, AI helps translate it into a plan: what to produce, how much and when, taking into account the machines' real capacity, the availability of raw material and the lead times of the inputs. The plan stops being a static spreadsheet and becomes a recommendation that adjusts when something changes.

2.3 On top of your ERP, without a multi-year project

The key for an SME is that this does not require a giant MRP or a migration. The AI layer connects to the data you already have in your ERP (sales, stock, orders) and starts working within weeks. It is the same logic as purchase planning: adding intelligence on top of what already exists rather than replacing the system.

3. A real case: an industrial SME that stopped guessing

3.1 Before

An SME manufacturing injection-moulded plastic products, with several product lines and seasonal demand, planned production in a spreadsheet the owner and the plant manager built by hand each week. Excess of some finished goods (immobilized capital) coexisted with missed orders on others. They had looked at MRP systems, but the prices and timelines had scared them off.

3.2 A phased implementation

1. Forecast (month 1). The AI layer was connected to the ERP and a demand forecast per product was built, using history and seasonality.
2. Suggested plan (month 2). With projected demand, AI began suggesting the weekly plan taking capacity and raw material into account, replacing the Sunday spreadsheet.
3. Continuous adjustment (month 3). The plan started recalculating when large orders came in or input stock changed, so it stopped being a weekly snapshot.

3.3 After

The most immediate change was for the owner: he got his Sunday nights back. But what was measurable was the imbalance: excess finished goods came down (less frozen capital) at the same time as missed orders (fewer annoyed customers). The plan, resting on a forecast that learned, kept improving cycle after cycle. The SME achieved a level of planning it believed was reserved for much larger companies, without the cost of a traditional MRP, and connecting production to the rest of its operation, as in any product lifecycle view.

4. Step-by-step implementation

4.1 Get the demand history in order

The forecast feeds on the past. The first step is having a clean sales history per product in your ERP. It does not need to be perfect; it is enough that it be reliable and sufficient to see patterns.

4.2 Start with the lines that hurt most

Do not try to plan the whole catalogue from the outset. Pick the product lines where the imbalance costs you most (the ones with the greatest excess or the most missed orders) and start there. What you learn on those lines transfers to the rest afterwards.

4.3 Integrate capacity and raw material

A good production plan is not only demand: it is demand against real capacity and available raw material. Adding those constraints is what turns the forecast into an executable plan rather than a wish.

4.4 Iterate every cycle

Production planning is a living cycle, not a document. Each week the plan is executed, compared with what actually happened, and the forecast is refined. With each turn, AI learns and the plan becomes more reliable.

5. ROI and measurable benefits

5.1 What to measure

The key indicators:

• On-time order fulfilment (fill rate).
• Finished goods inventory and its turnover.
• Production stoppages from missing raw material.
• Time spent planning (from hours to minutes).

5.2 The typical return

The return comes from both sides of the imbalance: less capital frozen in surplus product and fewer sales lost to missing product. For an SME, freeing up that capital and stopping the missed deliveries usually has an impact on cash and on customer relationships that shows up fast, without the investment of a traditional MRP.

5.3 The benefit for whoever decides

There is a human return: the owner or the plant manager stops carrying an impossible problem alone, in their head and in a spreadsheet. Planning stops depending on one person's intuition on a Sunday night and becomes a process the company can sustain and improve.

6. Common mistakes in planning production at an SME

6.1 Planning only with the owner's intuition

The intuition of somebody who knows the business is valuable, but it neither scales nor learns from past mistakes. When the plan depends on one person's head on a Sunday night, every week starts almost from scratch and the good calls do not accumulate. Adding a forecast that learns from each cycle does not replace that knowledge: it frees it from the impossible part (anticipating demand for dozens of products at once).

6.2 Believing the only alternative is a giant MRP

Many SMEs assume planning well requires a half-million-dollar system and a two-year implementation, and since that does not add up, they stay in Excel. It is a false dilemma. A layer of AI on the ERP you already have puts the forecast and the plan within an SME's reach, in weeks and with no migration. You do not have to choose between blind Excel and the impossible MRP.

6.3 Forecasting demand and forgetting the constraints

A good demand forecast is not a plan. If you project how much you will sell but ignore the machines' real capacity, the available raw material and the inputs' lead times, the plan looks fine on paper and breaks on the shop floor. Estimated demand has to be crossed with production constraints so the plan is executable rather than a wish.

6.4 Making the plan once and never reviewing it

Production planning is a living cycle, not a document to be filed. If the plan is never compared with what actually happened and the forecast is never adjusted, it never improves. The value accumulates in the iteration: every week you execute, measure and correct, AI learns and the plan becomes more reliable. Treating it as a fixed snapshot wastes exactly what makes it useful.

Ready to plan your production without guessing?

Planning production by eye, in a spreadsheet, is expensive: excess coexists with missed deliveries, and everything depends on one person's intuition. You do not need a half-million-dollar MRP to fix it. A layer of AI on top of your ERP puts demand forecasting and the plan within an SME's reach, in weeks and on the data you already have.

Want to see how it works in practice? Book a demo and we will show you how to plan your production from your own history, without changing systems.