A buyer who purchases steel, copper, or raw materials for plastics has seen prices move 20% to 30% within a single quarter, more than once in recent years. In such a reality, the question of whether we bought at the right time is worth not just a few percent of profitability, but sometimes all profitability. For years the answer rested on the buyer’s gut feeling and the last quoted price. Artificial intelligence is beginning to replace the gut feeling with probability calculations at a high level of reliability. <\/p>\n
The consulting firm Roland Berger frames the challenge in four practical questions. When to buy; how much to purchase in advance and store; what is the right mix between spot market purchasing and forward and futures contracts; and what indexation formula to set in the price adjustment clause with suppliers or customers. For each of these questions, machine learning can contribute a data-based decision. As they formulate it there, in an era of rapid volatility, Big Data analysis is the compass of commodity procurement. <\/p>\n
At its core, the model does one thing: it tries to predict where the price of a commodity will go. To predict well, it doesn’t only look at past prices \u2014 it also draws on many external data points: how much of the commodity is available in the market, what is the state of demand, geopolitical events, weather conditions that may affect crops, exchange rates, general economic indicators, and current inventory levels. All of these enter the model simultaneously. <\/p>\n
There is a variety of algorithms for performing this task. On the simpler side is XGBoost \u2014 a method that combines many simple decision trees into one powerful forecast. On the more sophisticated side are LSTM and GRU models, which simulate the way a human brain ‘remembers’ sequences of events over time \u2014 a critical capability when dealing with data where the historical order matters. <\/p>\n
At the technological frontier today, what are called ‘foundation models for time series’ are developing \u2014 tools like Chronos-2 published in 2025, trained on enormous amounts of data before being presented with specific commodity data. This approach is analogous to how a language model like ChatGPT knows the entire world before being taught a specific task \u2014 the advantage is greater flexibility and faster integration into new scenarios. <\/p>\n
similar to how a language model like ChatGPT knows the entire world before being taught a specific task \u2014 the advantage is greater flexibility and faster integration into new scenarios. <\/p>\n
AI is not only concerned with market prediction. The Should-Cost approach, or cost-based target price building, constructs a data-driven target price from the cost components themselves \u2014 raw material indices, energy, labor, and conversion cost \u2014 and then flags when a supplier’s price has deviated upward and triggers renegotiation. Tools in this category, such as Roland Berger’s CostIQ and market intelligence platforms like ChAI and PriceVision, illustrate the direction: turning spending data and market data into procurement signals and target prices. At the foundation of all this sits organized spend analytics. <\/p>\n
Here it is important to be alert and have a finger on the pulse. Commodity markets are inherently difficult to predict, because they are driven by shocks, geopolitics, and speculation that no model identifies completely. AI gives direction and probability, but you must know it does not give certainty. An extreme event, war, sanctions, or a blocked maritime strait, breaks any model trained on the past. The real value of these tools is not in perfect price prediction but in smarter timing, discipline in price protections (hedging), and the ability to challenge a supplier’s price offer with data instead of a feeling. And all this depends of course on spending data that is as accurate and structured as possible, which most organizations simply don’t have. Again, garbage in garbage out. <\/p>\n
In most organizations in Israel, procurement is still done based on gut feeling and the last price, and exposure to raw materials is almost not modeled. Value is created in the connection between market intelligence and disciplined procurement and risk management, for example through commodity risk management methodologies that Mashik works by in collaboration with the international consulting firm Kearney. The forecast is the input; the decision is the process. In Mashik’s procurement division we translate the graph into a procurement decision: when, how much, with what instrument, and with what indexation formula, so that market volatility transforms from a risk into an opportunity. <\/p>\n","protected":false},"excerpt":{"rendered":"
and improve procurement and hedging decisions. <\/p>\n","protected":false},"author":9,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[],"tags":[],"class_list":["post-43854","post","type-post","status-publish","format-standard","hentry"],"acf":[],"yoast_head":"\n