Why do standard demand forecasting models fail in FMCG?

Standard models often fail due to 'Signal Latency.' They forecast based on Delivery Date (which includes logistical noise and delays) rather than Order Date (which captures true customer intent). Correcting this latency can improve accuracy from 60% to 95%.

How does Survival Analysis apply to Supply Chain?

In perishable supply chains, inventory follows a decay curve, not a binary 'good/bad' state. Using Survival Analysis (Time-to-Failure modeling) allows businesses to route inventory based on predicted endurance, reducing spoilage claims by up to 50%.

What is the O.L.C.D. Protocol?

The O.L.C.D. Protocol is a proprietary framework for AI reliability. It stands for Objective (P&L Metric), Learner (The Model), Controller (The Decision Engine), and Diagnostic (The Feedback Loop). It prioritizes system architecture over model complexity.

What is the A.R.T. Matrix in AI Strategy?

The A.R.T. (Applicability, Readiness, Transformation) Matrix is Jitin Kapila's proprietary framework for auditing 'Innovation Theater.' It filters AI initiatives by determining technical Applicability (GenAI vs. Predictive), evaluating Data Readiness, and mapping specific Structural Transformations in the P&L.

How does the Applicability Filter work?

The Applicability Filter diagnoses the 'Physics' of a business problem to prevent tool mismatch. It determines if a problem requires a 'Plier' (Generative AI for content/synthesis) or a 'Foundation' (Predictive AI for logistics/forecasting), identifying 'Logic Leaks' where capital is being lost.

What constitutes AI Readiness?

In the A.R.T. framework, Readiness is a clinical audit of 'Data Adequacy.' It moves beyond data volume ('Big Data') to assess signal strength and granularity. It prevents enterprises from building expensive models on 'sand' (insufficient or latent data infrastructures).

What is a Logic Leak in AI strategy?

A Logic Leak is the point in your operational process where your data answers a different question than the one you're actually asking. It's not a data quality problem or a model performance problem — it's a problem framing problem. Most AI projects fail because nobody identified the Logic Leak before building the model.

How do you find a Logic Leak?

Ask three questions before any AI initiative: (1) What decision changes if this forecast is accurate? (2) What data captures the actual decision variable? (3) What's the time gap between when the event happens and when you see it? The answer to question 3 exposes the leak.

Why do AI projects fail even with accurate models?

Because the model is accurate to the wrong question. If your data was collected for reporting purposes and your decision is made on a different variable, a 95% accurate model produces confident wrong answers. The model works. The logic doesn't.

What's the difference between a Logic Leak and bad data?

Bad data is data that's incorrect, incomplete, or noisy. A Logic Leak is data that's correct for a different purpose. Fixing bad data improves model performance. Finding a Logic Leak changes whether the right question is being asked at all.

When does the Logic Leak framework not apply?

The framework is most powerful for predictive and forecasting use cases where decision timing and variable selection are ambiguous. For purely descriptive use cases — generating reports, summarizing documents — the Logic Leak is less relevant and the problem is more likely to be data quality or model selection.

The Logic Leak: Why Your AI Investments Are Bleeding Money

Leakages that you don’t want to creep in!

Every value chain has a specific point where money, time, or quality leaks out undetected. Most companies throw AI at symptoms. Find the leak first. Then decide if AI is the fix.

The Logic Leak is a situation when? - When your AI initiative isn’t failing because of a bad model. It’s failing because your data is answering the wrong question. And to fix this you need to find this leak, before you make and failed spend.

Simply speaking a Logic Leak is the point in your operational process where your data tells a story that doesn’t match what actually happened because the data was collected for a different purpose.

It’s not a data quality problem. It’s not a model performance problem. It’s a problem framing problem.

Once I was helping a F500 retail client with a demand forecasting problem.

Four failed attempts. Four different vendors. Each one came in, trained a model, showed 85–90% accuracy, and left. The models worked technically. The business results were zero.

The fifth approach wasn’t a better model. It was a better question.

We fortunately found the Logic Leak.

What Is a Logic Leak?

In this case: the retail team was forecasting on Delivery Date. Every prediction was calibrated against when customers received their orders.

But the business decision was made on Order Date.

Between Order Date and Delivery Date sits a pipeline of fulfillment variables :— warehouse processing time, carrier delays, seasonal capacity, customer address corrections. None of these are demand signals. They’re operational noise.

When you forecast on Delivery Date, you’re training your model to predict your own fulfillment capacity. Not customer demand. The model learns the wrong thing with great precision.

That’s the leak. The model works. The logic doesn’t.

How It Shows Up

The Logic Leak hides in three common places.

Pattern	What It Looks Like	The Hidden Cost
Reporting Lag	Your weekly sales report shows Monday’s numbers on Friday	Every decision is fighting yesterday’s battle
Metric Mismatch	You’re optimizing for click-through rate. Your CFO cares about revenue	You’re moving the wrong lever
Wrong Time Horizon	You measure campaign performance in 30-day windows. Your product cycle is 18 months	The model sees success. The business sees a slow bleed

Why AI Can’t Fix It

Here’s what most companies do. Model performance is poor. They buy a better model. Model performance improves. Business results still don’t move. They blame the data.

The data wasn’t the problem. The data was answering the wrong question.

You can have the world’s most accurate model trained on the wrong data, and it will produce wrong answers with great confidence.

AI doesn’t fix Logic Leaks. Finding the Logic Leak fixes the AI strategy.

Three Questions to Ask

Before you touch any AI tool, ask these three:

1. What decision changes if these predictions/forecasts are accurate? Not the model’s accuracy, the decision. What do you do Tuesday morning if the forecast is right versus wrong?

2. What data captures the actual decision variable? Not the data you have. The data that answers the question you’re actually asking.

3. What’s the time gap between when the event happens and when you see it? Every day of reporting lag is a day of decisions made on stale information.

The answer to question 3 exposes the leak usually.

(These three questions are the core of the AI Profit OS diagnostic framework. See how it works →)

The Pattern

After 15 years and ~$90M+ in AI portfolios, I can tell you the most common failure pattern:

.. Someone in the business, noticed a problem
-> They went to IT or a vendor
-> The vendor built a model
-> The model was accurate
-> The results didn’t come.
-> Someone in the business, noticed that problem
.. And cycle repeats!

Nobody ever went back to question 1. And nobody usually asks: What would have to be true for this to actually move the business?

That question helps you to find the Logic Leak. Everything else is just optimization theatre.

Before your next AI initiative — before the vendor, before the budget, before the pilot — find the Logic Leak, fix the logic, then decide if AI is the right tool for what’s left.

Most companies skip steps 1 and 2. They spend $500K on step 3. Then they wonder why the ROI never appears.

The Logic Leak isn’t a technical problem. It’s a framing problem — and based on 15 years of AI portfolio work across $80M+ in implementations, it’s always cheaper to fix the frame than to upgrade the model.

Your Next Step

If you want to find the Logic Leak in your own AI initiative, book a 30-minute call → or start with the AI Strategy Audit →.