How to Use AI to Perform SKU Rationalization

AIInventory
Written By Tracey Smith

*** I’ve recorded this process using AI and will be uploading it to YouTube. Subscribe to my YouTube channel so see the video when it launches. ***

SKU rationalization is one of the fastest ways to improve profitability, reduce operational complexity, and free up working capital. The challenge is not understanding why it matters… it’s knowing how to execute it efficiently across large datasets.

Artificial intelligence can accelerate this process, but only if you provide it with the right inputs and structure the analysis correctly.

This article walks through how to use AI to perform SKU rationalization and what information you need to give it to get meaningful results.

*** I’ll pretend that I know nothing about the data required for a SKU Rationalization as I enter the AI tool… ***

What Is SKU Rationalization?

If you’re not familiar with this activity, SKU rationalization is the process of evaluating your product assortment to determine which items should be:

  • Retained (kept active and continue to sell),

  • Discontinued (sell through existing inventory but make or buy any more),

  • Obsolete (stop selling immediately).

The goal is to eliminate underperforming SKUs while protecting (and ideally expanding) profit and customer value.

Why Use AI for SKU Rationalization?

Traditionally, SKU rationalization involves:

  • Manual Excel analysis

  • Multiple pivot tables

  • Time-consuming segmentation

AI changes this by allowing you to:

  • Analyze datasets quickly

  • Identify patterns across multiple variables

  • Simulate decision scenarios

  • Generate recommendations based on your criteria

However, AI is only as effective as the data and context you provide.

So, pretending I know nothing about the data needed, I headed into AI and asked it.

Step 1: Ask AI What It Needs

Question from Tracey: “What information do you need from me to perform SKU rationalization?”

Answer from AI: To perform a thorough SKU rationalization for you, I need a solid set of data. The more complete and accurate the information you can provide, the better and more actionable the analysis will be. Here’s exactly what I need (prioritized from most important to nice-to-have):

1. Core Data (Essential)

  • Full list of SKUs with: SKU code/ID Product description/name Category / subcategory (e.g., “Electronics > Chargers”)

  • Sales history for each SKU (ideally last 12–24 months): Units sold per period (monthly or quarterly is best) Revenue per SKU per period Gross margin % or profit per unit (if available)

2. Cost & Inventory Data (Very Important)

  • Cost of Goods Sold (COGS) or landed cost per unit

  • Current inventory levels (on-hand quantity)

  • Inventory holding cost % (or average storage/warehousing cost per unit)

  • Stockout history or lost sales data (if you track it)

  • Obsolescence / expiry dates (especially for perishables or tech)

3. Performance & Demand Data (Important)

  • Any demand forecast for the next 6–12 months

  • Seasonality patterns

  • ABC classification (if you already have one) or Pareto data

  • Return rates or defect rates per SKU

4. Business Rules & Constraints (Critical for recommendations)

  • Minimum acceptable annual revenue or margin per SKU

  • Must-keep SKUs (strategic, new launches, customer-specific, etc.)

  • Cannibalization relationships between SKUs

  • Supplier minimum order quantities or lead times

  • Storage or shelf-space limitations

  • Your overall goals (e.g., reduce SKU count by 30%, improve inventory turns, free up cash, simplify operations, etc.)

5. Nice-to-Have Data (Improves Quality)

  • Selling price per SKU

  • Channel breakdown (Amazon, website, retail, wholesale, etc.)

  • Customer concentration (does 20% of customers buy 80% of a particular SKU?)

  • Substitution possibilities (which SKUs can replace others?)

One thing I noticed is that AI did not recommend providing an indicator of whether a SKU was new or not. New SKUs often get flagged for elimination in automatic analyses since the demand has a short history, but let’s proceed and see if AI can recognize a short sales history and if it concludes that the SKU is a new one and should be kept.

So off I went and gathered a data set with some of the recommended information. Since this data set is for demonstration purposes only, I didn’t gather the fields which would take a lot of my time to obtain.

AI even recommended the file formats to use (Excel / CSV file (best), Google Sheet link, Paste a sample table directly here) and offered to provide me with a template.

I politely declined the template. We should be polite to AI, shouldn't we?

Step 2: Structure Your Data for AI

AI performs best when your data is clean and structured. At minimum, organize your dataset into a table of rows and columns with one SKU on each row. Provide clear header names so AI can more easily interpret what each column is.

Step 3: Define Your Rationalization Criteria

Before asking AI for recommendations, define what “good” and “bad” look like.

Examples of criteria:

  • Low revenue + low margin → candidate for removal

  • High revenue + low margin → pricing review

  • Low sales frequency → potential obsolescence

  • High inventory + low sales → excess stock risk

You can explicitly tell AI something like:

“Classify SKUs into Keep, Discontinue or make Obsolete based on revenue, margin, and sales frequency.”

Step 4: Let AI Perform the Analysis

Once your data and criteria are defined, AI can:

1. Segment Your SKUs

AI can group SKUs into categories such as:

  • High value (top contributors)

  • Margin drivers

  • Low performers

  • Dead stock

2. Identify “Profit Leaks”

AI can highlight SKUs that:

  • Generate revenue but little profit

  • Consume working capital without sufficient return

3. Detect Hidden Patterns

For example:

  • SKUs with high sales but long lead times (risk exposure)

  • Products with sporadic demand but high inventory levels

Step 5: Generate Actionable Recommendations

You can even ask AI to provide you with actionable recommendations for your SKUs.

Discontinue Candidates

  • Low revenue, low margin, low frequency

  • No strategic importance

Pricing Adjustments

  • High demand but weak margins

Inventory Reduction

  • Excess stock relative to sales velocity

Portfolio Simplification

  • Multiple SKUs serving the same purpose

Step 6: Validate with Business Context

*** AI does not inherently understand your business operations outside of the data you have provided, so all recommendations should be reviewed before implementation. ***

AI doesn’t understand things like:

  • Customer relationships

  • Strategic positioning

  • Future product plans

Before making decisions, consider the following:

  • Are low-performing SKUs tied to key customers?

  • Do certain SKUs act as “loss leaders?”

  • Are there contractual or branding considerations?

  • Are there other exceptions and obligations to consider?

Step 7: Iterate and Refine

SKU rationalization is not a one-time exercise.

With AI, you can:

  • Re-run the analysis monthly or quarterly, saving the exact prompts and data file format for speedy usage.

  • Adjust thresholds (e.g., margin targets).

  • Incorporate new data columns (seasonality, promotions).

Over time, this creates a continuous improvement loop as you work toward establishing a standard process for this analysis.

Final Thought & Takeaways

SKU rationalization is ultimately about focusing your business on the products that generate the most value.

AI doesn’t replace judgment, but it gives you the clarity and speed needed to make better decisions with your data.

Takeaways

  • AI can dramatically speed up SKU rationalization.

  • The quality of output depends on the data you provide.

  • Clear criteria are essential for meaningful recommendations.

  • Human validation is still required for final decisions.

Tracey Smith http://www.numericalinsights.com
Next

How AI Transforms Procurement: Practical Ways to Save Time and Money