My jaw nearly hit the floor when a leader at a major oil and gas company described one of its first ideas for enterprise AI: managing its Power BI dashboards.
Over the years, the company had accumulated a dashboard for almost everything—production, inventory, service, demand, costs, reliability, sales, and on and on. Each had been created to answer a legitimate question. Collectively, however, they had created a new problem: thousands of dashboards that employees could neither navigate nor consistently interpret.
The proposed solution was to use the newest generation of technology to manage the complexity created by the previous one (certainly someone would eventually build another dashboard to track the performance of the AI managing all the other dashboards).
It is an amusing example, but not an unusual one. Industrial companies have spent decades collecting, connecting, modeling, and visualizing their data. They can see more of their operations than ever before. What many still cannot do is translate that visibility into a clear operating decision.
A dashboard can show a cracker's production rate, a reactor's reliability, current inventory, expected customer demand, logistics costs, and projected margin. All of that information matters.
But suppose the cracker goes down next month and ethylene becomes scarce. Should the company buy ethylene on the market to keep downstream assets running? Should it bypass the shortage by sourcing downstream intermediates instead? How much should it buy, from where, and at what premium? Which assets should continue operating at reduced rates? How should other precursor products be managed when an unaffected asset is feeding into the same downstream bottleneck? Which final products and customers should receive the constrained supply? And how should all of those answers change if the outage lasts twice as long as expected?
The company may possess every data point needed to analyze the situation and still lack an answer.
That is the next frontier for industrial companies: moving from systems that describe and predict the business to systems that help decide how to run it.
This gap is not the result of a lack of investment. It is the product of how industrial information technology evolved.
Enterprise resource planning systems gave companies a common system of record for transactions, orders, inventory, production, materials, and financial activity. Data warehouses and business intelligence tools made that information easier to aggregate and report. Leaders could more quickly understand what happened last quarter, last month, or during yesterday's shift.
Data science and machine learning added another layer. Instead of only describing the past, companies could begin predicting the future: expected demand, equipment failures, customer behavior, quality problems, and delivery performance. Data lakes and lakehouses made it possible to bring more information together, while modern visualization tools put the resulting metrics and forecasts in front of thousands of employees.
Each layer expanded what the enterprise could know:
Most technology stacks still stop before answering the question that matters most:
Given our objectives, economics, constraints, and uncertainty, what should we do?
That final question is fundamentally different from the first three. It cannot be answered by placing another chart beside the demand forecast. It requires the company to connect its information to a model of the decisions it can make and the economic consequences of those choices.
Consider a typical S&OP or IBP process. The team may have an excellent demand forecast for next quarter, but the forecast does not determine how much product to make, where to make it, what inventory to position, or how to allocate scarce material.
Even a perfect forecast would not answer those questions.
The plan still depends on production capacities, yields, changeovers, raw-material availability, intermediate balances, logistics constraints, contractual commitments, customer economics, inventory policies, and the value of preserving flexibility. It also depends on what the company is trying to accomplish. A plan designed to maximize volume may look very different from one designed to maximize contribution margin, protect cash, maintain strategic accounts, or reduce exposure to an uncertain outage.
Traditional KPIs can obscure these distinctions. A site may achieve its utilization target by purchasing an expensive intermediate to keep a unit running, even though the incremental production destroys margin. A commercial team may protect its service metric by fulfilling low-value orders while a more profitable customer goes short. A business may minimize one logistics cost while shifting a much larger cost somewhere else in the network.
Every function can hit its number while the enterprise makes the wrong decision.
Dashboards make these KPIs visible. They do not reconcile them.
Generative AI does not eliminate this problem. Large language models can make information easier to find, summarize, and interrogate. They can help employees navigate reports and ask better questions. But fluency is not the same as decision quality. Unless a system represents the company's economics, physical constraints, alternatives, and uncertainty, it cannot reliably determine which operating choice creates the best outcome.
Prediction is an essential input. It is not a prescription.
One of our clients is confronting this distinction now.
An ethylene shortage is constraining production across an interconnected manufacturing system. The obvious response would be to determine how much ethylene can be sourced externally and at what price. But that is only the first branch of a much larger decision.
The company may also be able to buy downstream products or import them from other assets, bypassing part of the ethylene shortage and allowing final production to continue at reduced rates. Whether that is economical depends on far more than the purchase price. The analysis must account for incremental freight, conversion yields, operating costs, available capacity, inventory already in the system, contractual obligations, product and customer margins, and the opportunity cost of using constrained equipment one way instead of another.
Then the material balances begin to interact.
If a downstream intermediate is purchased, which units should continue operating? At what rates? Does the purchased material displace precursor production from an unaffected asset? If that asset continues running, will it create inventory that cannot be consumed because the downstream unit is now the bottleneck? Should the company reduce rates, change the product slate, sell an intermediate, or preserve the precursor in anticipation of the outage ending?
Next comes customer allocation. The company does not yet know precisely which customers will be affected because the answer depends on the sourcing and production plan. It must determine which products will be short, which orders can be fulfilled from inventory, which commitments can be rescheduled, and where scarce material creates the most economic and strategic value. A high-volume customer is not necessarily the highest-value customer. Neither is the customer with the highest price if serving that order consumes scarce intermediates, adds expensive logistics, or prevents the company from producing a more valuable downstream product.
Finally, the company must make these choices without knowing how long the shortage will last, what spot material will remain available, how prices will move, or whether customers will order what they currently forecast.
A single base-case plan is not enough. The right decision for a short disruption may be the wrong one if the outage persists. A plan that produces the highest expected margin may also expose the company to an unacceptable downside if one key assumption fails.
The company therefore needs to evaluate a sequence of connected decisions:
This is not a dashboard problem. It is a decision-system problem.
Industrial Decision Science connects data and predictions to the economic and operational choices required to run the business. It combines models of the physical system, business economics, uncertainty, and prescriptive analytics to identify feasible actions and quantify their consequences.
At RedSynth, we organize this work around three pillars.
Industrial organizations often manage through KPIs because KPIs make a complicated business legible. But a proxy for value is not the same as value.
Economic realism means representing the full consequences of a decision: variable production costs, purchasing premiums, freight, yields, changeovers, inventory effects, contractual penalties, opportunity costs, and the economics of the customers and products being served. It also means recognizing that some considerations—such as protecting a strategic relationship or preserving future operating flexibility—may need to be made explicit rather than buried inside an arbitrary target.
The objective is not to maximize utilization, volume, service, or any other metric in isolation. It is to understand the economic tradeoffs among them and make a deliberate choice.
Most planning processes are built around averages: an average forecast, an expected outage duration, a standard yield, a typical lead time, or a base-case market price. Those averages make planning manageable, but they can also create a dangerously precise view of the future.
The business does not experience the average. It experiences one of many possible outcomes including the ones in the tails.
We describe the alternative as de-averaging the decision. Instead of asking which plan performs best under one expected set of assumptions, ask how the choices perform across a range of plausible futures. Where does the recommended action change? Which plan is slightly better in the base case but disastrous in the downside case? What option appears expensive on average but protects the company from a much larger loss? What commitment creates value now, and what commitment prematurely eliminates flexibility?
De-averaging does not mean predicting every disruption. It means designing decisions to be robust while acknowledging that forecasts will be wrong.
Descriptive analytics tells us what happened. Predictive analytics tells us what may happen. Prescriptive analytics evaluates the situation and tells us what we should do about it.
That means generating feasible sourcing, production, inventory, logistics, and allocation choices; calculating their economic consequences; exposing the constraints that matter; and recommending a path. The objective is not to hand an executive a mysterious black-box answer. A good decision system makes the tradeoffs explicit: what we gain, what we give up, which assumptions drive the result, and when a different action becomes preferable.
In the ethylene example, the output is not another visualization of the shortage. It is a plan: what to source, which assets to run, what to make, whom to serve, what it is worth, and how the plan should change as conditions evolve.
A disruption often triggers a familiar corporate ritual. Data is pulled from several systems. A new spreadsheet is assembled. Teams debate assumptions. Analysts work late to reconcile material balances and economics. Leaders review a small number of scenarios in a series of meetings. By the time a decision is reached, some of the inputs have already changed.
The spreadsheet may resolve the immediate crisis, but the capability disappears when the crisis does. The next outage, storm, waterway closure, war, supplier shutdown, or demand shock starts the process again.
That is not how we approach these problems.
RedSynth does not come in simply to calculate the answer to one decision. We build decision systems that allow companies to repeat and operationalize the process. As supply, demand, costs, operating conditions, and market opportunities change, the system can update the choices and their economics. The organization develops a durable way to make the decision month after month instead of another heroic analysis that lives on one employee's laptop.
This distinction changes how a company responds to volatility. While competitors are still collecting data, rebuilding spreadsheets, and determining how the latest disruption affects them, a company with an operational decision system can evaluate the possibilities, understand its exposure, and act. Its commercial team can make deals with confidence because operations and economics have been considered together. Its leaders can move quickly without pretending the future is certain.
The same capability that protects the business during a disruption also helps it recognize opportunity. A sudden change in feedstock economics, customer demand, asset availability, or competitor behavior can create a short window to act. A company that already understands its decision space can see the opportunity and chart a path while others are still debating which numbers are correct.
Dashboards are not the enemy. Companies still need reliable data, reporting, forecasts, and visibility. The mistake is treating visibility as the end of the journey.
For each critical recurring decision in your business, ask four questions:
If the answer is no, your company may not be short on data, analytics, or dashboards. It is short on decision systems.
The next era of industrial performance will not be defined by which company collects the most data or deploys the most AI. It will be defined by which companies can consistently turn information into economically sound action under uncertainty.
That is the promise of Industrial Decision Science.
RedSynth helps industrial companies build the decision layer between their data and their operations. If you have a recurring decision that still depends on averages, disconnected KPIs, and heroic spreadsheet work, that is where the conversation should begin.