By Nikhil Gilani, Director – Goldratt Bharat
Organisations are deploying AI faster than ever. Predictive analytics, automated workflows, intelligent reporting — the investments are real, and the expectations are high. Yet a frustrating pattern keeps emerging: significant AI spend, limited business impact. The technology isn’t the problem. The application is.
Most organisations start by asking “Where can we use AI?” instead of “What actually needs to improve?” Without that clarity, AI gets deployed across the business in a spray-and-pray fashion — optimising things that don’t move the needle, while the real bottleneck sits untouched. There is one question every enterprise must answer before deploying AI: Where is the constraint?
The Constraint Controls Everything
Every organisation is a system. That system has a limiting factor – a constraint – that governs how much value it can generate. Improve anything outside that constraint and you get local efficiency gains. The system as a whole will not improve.
This is the core insight of the Theory of Constraints: performance is dictated by the weakest link. Apply AI without finding that link, and you end up with better dashboards, smoother workflows, and more automation, but no meaningful lift in revenue, profitability, or cash flow. AI shouldn’t try to improve everything. It should elevate the constraint.
Four Constraints. Four Distinct Problems.
Enterprise constraints typically fall into one of four categories. Each demands a different focus.
Cash. When liquidity is the bottleneck, no amount of customer engagement tooling or reporting automation matters. The organisation can’t procure, can’t produce, can’t fulfil. AI’s highest value here is in reducing cash to cash cycle time — predicting receivables, prioritising collections, optimising inventory, and improving visibility into working capital requirements l. Success looks like faster cash velocity.
Supply. When materials aren’t arriving reliably, production delays and missed commitments follow. AI can help through supplier performance analytics, demand-supply synchronisation, and early risk detection. The goal isn’t cost reduction – it’s continuity of flow. Success looks like fewer stockouts and higher material availability.
Orders. When there isn’t enough demand to fill existing capacity, the constraint is at the front of the pipeline. This is as much a delivery reliability problem as a sales one — customers don’t come back if you can’t commit and deliver. AI can sharpen segmentation, improve order pipeline visibility , and align delivery promises with actual capacity. Success looks like higher throughput from the same resources, improved delivery performance and customer delight..
Operations. When internal processes are the bottleneck, flow breaks down somewhere in production or delivery. This is also where AI gets deployed most reflexively — often before anyone has identified the constraint. The key insight here is that total time equals cycle time plus waiting time. In most systems, waiting time is the bigger culprit — caused by poor scheduling, batching, or coordination gaps. AI should target those delays at the bottleneck specifically, not operational efficiency in general. Success looks like increased throughput at the constraint, better flow and on time delivery
Stop Optimising for Cost. Start Optimising for Throughput.
The default framing for most AI initiatives is cost reduction. It’s the wrong frame. Cost reduction has a ceiling. Cut enough, and returns diminish. Throughput — the rate at which the organisation generates value — has no such ceiling. More revenue, faster delivery, healthier cash flow: these are the metrics that matter. AI investments should be evaluated by whether they move those numbers. If they don’t, the initiative isn’t delivering real impact — regardless of how technically sophisticated it is.
Measure What the Constraint Produces
Without clear metrics tied to system-level outcomes, AI projects drift. The discipline is simple: identify the constraint, define the metric that reflects its performance, then deploy AI to move that metric.
- Cash constraint → Receivable days, cash to cash cycle time
- Supply constraint → Material availability, stockout rate
- Order constraint → Order inflow, conversion rate
- Operations constraint → Throughput, On Time in Full (OTIF) performance
This linkage turns AI from an activity into an investment with a measurable return.
AI Is a Lever. Use It Where It Has Leverage.
To quote Archimedes
“Give me a lever long enough and a fulcrum on which to place it, and I shall move the world.”
AI is a powerful tool. But a lever only works if it’s placed at the right point. Applied to the constraint, AI can deliver outsized impact. Applied elsewhere, it risks making the wrong things run faster. The starting point for any serious AI initiative isn’t technology selection. It’s system understanding. Get the constraint right, and AI becomes a genuine driver of growth and competitive advantage. Get it wrong, and you’re just automating noise.
