The problem: duplicate formulas hiding under different names
Plants, teams and acquisitions each bring their own codes. Over time the range fills with formulas that do the same job, and finding them means comparing spreadsheets line by line.
Duplicates rarely look like duplicates. One plant calls it a tint base, another calls it a standard, and an acquired business has its own code for the same pigment loading. Each version carries its own qualification history, stock and documentation, so the overlap stays hidden until someone compares the formulas themselves.
What this demo shows
- Scope the analysis to one application, aqueous dispersions for coatings.
- Choose the columns to compare, and set annual volume as the target and cost per kilo as the penalty.
- Run K-means clustering to group the 70 formulas into seven families and map them in two dimensions.
- Sort the phthalo blue family by efficiency and find three codes from three teams with the same pigment loading, dispersant and color.
Why it matters
Every extra formula in a range costs something to keep: raw material stock, QC specs, regulatory checks and change control all multiply with the code count. When a supplier changes an ingredient, each near-duplicate has to be reviewed on its own.
Clustering on the formula, its calculated content and its measured performance gives the team a shortlist to review instead of a spreadsheet exercise. Codes that already match can be merged with a re-qualification, and the formulas that remain are the ones that earn their place.
See it with your own data
This video is part of Formulation. To see how it would work with your own formulations and records, request a demo. Our team replies within 30 minutes during business hours.
Read the transcript
Three plants and an acquisition left us with 70 pigment dispersions. Which of them are the same product under another name? Rationalization finds formulas that do the same job under different names, so you can merge them and simplify the range. We scope it to one application, aqueous dispersions for coatings, so we only compare like with like.
The columns decide what gets compared. Here that's the formula itself, pigment and dispersant content from input calculations, plus the measured color it delivers. Any column can also take a role. The target is what you want more of, like annual volume or margin. The penalty is what you want less of, like cost or ingredient count.
Target and penalty rank formulas inside a family. They only shape the families if you turn on this option. With the columns in role set, one run does the math. Every column goes onto the same scale, then K-means clustering groups the formulas that sit close together. The platform tries different group counts and keeps the cleanest split.
Here that's seven families. The map squeezes every column into two axes, so similar formulas land together and odd ones stand out at a glance. Efficiency is the target divided by the penalty. Here that's annual volume divided by cost per kilo, so a higher score earns its place. Take cluster six, the phthalo blues, 12 formulas for one product line.
Sorted by efficiency, two formulas lead by a wide margin. The ones below them are where to look for overlap. Three different codes from three different teams. The Gen standard, an acquired chrome scratch, and a key account tint base. Same pigment loading, same dispersant, same color. Found in minutes, not by comparing spreadsheets line by line.
Merging codes that already match means a quick re-qualification, not a reformulation. Plant versions of one code stay since they exist for local reasons. The value is in codes nobody knew overlapped. So, each family runs on fewer formulas with less to maintain.
