Why Optimization Is the Missing Link After Modeling
You built a strong predictive model, so now what? The leap from accurate predictions to optimal actions is where value happens. Minitab’s new Response Optimizer inside the Predictive Analytics Module closes that gap by recommending the best combination of inputs to hit your goal, whether that’s maximizing chemical yield, minimizing call escalations, or achieving a precise diameter target.
Minitab’s Response Optimizer has always been a key tool when analyzing a designed experiment. For example, the Body Interior Six Sigma team at Ford Motor Company used Design of Experiments (DOE) together with Minitab’s Response Optimizer to fine‑tune needler machine settings for vehicle carpet. Their challenge was to eliminate brush marks while preserving plushness, durability, and stain resistance. By applying the Response Optimizer, they achieved full elimination of defects, unexpected improvements in plushness, and brought the process under control in just 12 days from defining the problem to implementing the solution.
What’s New in Minitab’s Response Optimizer
And now after you create a model with TreeNet®, Random Forests®, or MARS® in the Predictive Analytics Module, you can click Response Optimizer right from the toolbar at the top of your model’s results and obtain the best settings for your predictors to achieve your desired outcome.
Tell Minitab what “success” looks like—Maximize, Minimize, or Target—and set any practical limits for your inputs (e.g., temperature must stay between 180–200°C). The Response Optimizer updates instantly, so you can drag sliders to run “what-if” scenarios and see trade-offs in real time.
Examples
1. Manufacturing On-Time Delivery
Let’s say you want to increase the probability that custom orders ship on time. You train a TreeNet Classification model using predictors like supplier lead time, order complexity, and expedite status. From the model results, now launch Response Optimizer and choose Maximize for the binary response “On-Time.” The optimizer recommends the combination that yields the highest probability of on-time delivery.
Here, the probability of an on-time delivery is maximized when the black lines on each graph are at their highest point, so tightening the supplier’s lead-time window and standardizing the sub-assembly process should do the trick. You may want to play around with these settings, though. For example, you can change the quantity ordered and see how that impacts the probability of an on-time delivery. You can change the values on the graph to obtain predictions for other settings as well. You might even find a better solution in the process!
Suppose expediting and standardizing the sub-assembly process reduces late deliveries by 20%, cutting penalty costs or lost customer revenue. If the company has $1M in costs tied to late deliveries annually, that’s a $200,000 annual benefit. Combined with small cost of changes, the ROI (benefit ÷ cost) could easily exceed 5×.
2. Call-Center Escalations
Now let’s say you train a Random Forests Classification model to predict whether a call to customer service will escalate. Predictors include agent tenure, training hours, staffing level, and call type. In Response Optimizer, set the goal to Minimize the escalation probability. Because you want to minimize the probability of call escalation, look for settings where the black lines are low. For example, here are some good settings for a “Billing” call type:
But what if instead, the type of call is “Technical”? With each call type, you will need to tweak the settings of the other factors to minimize the probability of the call being escalated. The interactive Response Optimizer let’s us do just that and obtain the corresponding predictions on the fly.
For this type of call, notice that a more experienced agent is needed to reduce the chance of call escalation.
In call centers, efficiency gains unlocked through analytics have showed ROI in the 20-30% range in cost savings by reducing escalations, wait-times, and agent overstaffed hours (CallCriteria). A ClearSource case study noted that reducing turnover and improving first call resolution using data-driven tools can save $5,000-$12,000 per agent in recruitment and training costs.
Suppose optimizing agent training hours and staffing levels reduce escalation rates by 30%. If each escalated call costs an extra $50 and there are 10,000 escalated calls annually (extra cost $500,000), you save $150,000. If implementing the changes costs $30,000, ROI is 5×.
3. Process Yield
Finally, you might need to model chemical yield with MARS® Regression to capture nonlinear interactions among temperature, residence time, and catalyst ratio. Once again, open Response Optimizer, choose Maximize for yield, and add Constraints to impose safety and quality bounds. Within minutes, they arrive at actionable set points, along with prediction intervals to understand variability risk around the recommended settings.
Instead of a graphical view of the relationship between each setting and the response, you can also use the table view shown below to see what settings lead to the best chemical yield of 95.8%
Suppose optimizing temperature, residence time, and catalyst ratio improves yield by 5% in a facility producing, say, 1,000,000 units/year with margin of $10/unit. That’s $500,000 more margin in yield. If the cost to tune the process (additional control, monitoring, minor adjustments) is $50,000, ROI is 10×.
Tips & Guardrails for Reliable Optimization
- Model first, optimize second. Always start with a well-validated model before optimizing. The optimizer can only be as good as the model it uses.
- Set realistic bounds. Use engineering knowledge and business rules to constrain the search. It keeps solutions practical.
- Pick the right goal. Choose Maximize, Minimize, or Target based on the KPI.
- Use the interactive plot. Drag sliders to stress-test the solution. Small tweaks can reveal robust settings with similar performance.
Get Started in Three Easy Steps
- Fit your model with TreeNet, Random Forests, or MARS in Minitab’s Predictive Analytics module. Or use Automated Machine Learning to have Minitab pick the best model for you.
- Click Response Optimizer on the model’s results toolbar.
- Set your goal, then click OK to use the Optimization Plot to evaluate and select the best settings.
It’s that easy!
Talk to Minitab
Ready to turn your predictions into optimized, real-world decisions that will save you money?
