From reactive to predictive: How AI is redefining supply chain planning

Think about the last time something in a supply chain went wrong. A shipment got stuck. A warehouse ran out of stock just when demand picked up. A truck reached the wrong place at the wrong time. For most of the history of logistics, this is how the industry has worked. Something goes wrong, and then people scramble to fix it. That is what we mean when we say logistics has been reactive. Decisions get made after a problem shows up, not before.

That is now changing, and it is changing quickly. Across procurement, warehousing, fleet management and last mile delivery, companies are using artificial intelligence to see problems coming before they happen. This is what people in the industry call predictive logistics. Instead of asking what went wrong, teams are now asking what is likely to go wrong next week, and what they can do about it today.

Procurement: The earliest point where things can go wrong
Before a product even reaches a warehouse or a delivery van, someone has to buy the raw material or the finished goods. This is procurement. It sounds like a simple business function, but it is actually one of the most important points in the whole supply chain. If a company buys too little, it runs short later. If it buys too much, it ties up money in stock nobody wants yet. Get procurement timing wrong, and everything downstream, including warehousing and delivery, ends up either short of supply or drowning in it.

For a long time, procurement teams have made these calls based on how much they bought last year, and a bit of gut feeling. Gaurav Baheti, Founder and Chief Executive Officer of Procol, an AI-driven procurement platform, believes the next step forward is what he calls Multi Agent Systems. This means using several specialised pieces of AI software, each watching a different signal such as supplier performance, market prices or inventory levels, and having them work together rather than in isolation. Instead of just showing a report to a human who then decides what to do, these systems can start a sourcing process on their own, suggest which supplier to use, and adjust the plan as conditions change, all within limits set by the business.

Baheti explained this shift in his own words. "Procurement shifts from reacting to disruptions to continuously anticipating and preventing them, creating a more resilient and agile supply chain."

Procurement shifts from reacting to disruptions to continuously anticipating and preventing them, creating a more resilient and agile supply chain
Gaurav Baheti, Procol

He also pointed to real results. ElasticRun, a logistics and fulfilment network in India, used Procol's tools to cut its sourcing cycle time by 40%. Another company, Quippo Oil and Gas Infrastructure, brought down the time it took to complete a transportation auction from several days to under four hours, while also cutting logistics costs by around 15%. These are not small numbers. They show that faster procurement decisions can genuinely change how a business runs.

Fleet planning: Knowing where your trucks need to be before you need them
Once goods are bought, they need to move. This usually means trucks, and lots of them. Fleet planning is about knowing how many vehicles you need, where they should be, and when. Get this wrong and you either have trucks sitting idle costing money, or not enough trucks when demand suddenly rises.

Nikhil Agarwal, President of CJ Darcl Logistics, made an important point that is easy to miss in conversations about AI. He said that before any company can use AI properly, it first needs its basic systems talking to each other. CJ Darcl built a system called MoveX, a cloud based platform that connects booking, planning, paperwork, tracking and billing into one place. Only once that foundation was solid did the company start looking seriously at AI for things like pricing and route planning.

Agarwal explained why this matters so much at the scale his company operates. "At our scale, with access to more than 955,000 partnered vehicles through a network of over 200 branches, forecasting extends far beyond estimating shipment volumes. It is about positioning the right capacity on the right lane at the right time."

He also raised a challenge that is often overlooked. As companies begin adding electric and alternative fuel trucks to their fleets, forecasting gets harder, not easier. These vehicles carry less weight because of heavy batteries, they need charging infrastructure that is still being built, and they behave differently on long routes compared with diesel trucks. Any forecasting model now has to account for all of this, not just how much freight needs to move.

Mapping and delivery: Learning from what actually happens, not what should happen
Most of us think of a map as something fixed, a picture of roads and distances. In logistics, this old idea of a map is no longer good enough. Rohan Anand, Vice President and Head of Data Science at Delhivery, explained why. His team built Delhivery Maps, a system that learns from real delivery data rather than guesswork.

Every single delivery a rider makes generates information. Did they reach the right address? How long did the last stretch of the journey actually take, compared with what a normal map would have predicted? Anand pointed out that most map companies simply do not have this kind of information, because they are not the ones making the deliveries.

He described what this means in practice. "The map stops being a static reference layer and becomes a living model, constantly retrained on operational outcomes."

The map stops being a static reference layer and becomes a living model, constantly retrained on operational outcomes
Rohan Anand, Delhivery

This has a real effect on how Delhivery plans its network. Instead of grouping demand by postal code, which in India can cover very different types of neighbourhoods, the company can now understand demand at the level of a single locality or even a rooftop. Anand also spoke about the scale involved. Delhivery handled over one billion parcels in the last financial year, and forecasting at that size has to work at three levels at the same time, the whole network, the city or region, and the individual delivery rider. He said the aim now is to go further than just predicting delivery times. The next step is predicting which deliveries are at risk of failing, and which routes are likely to get congested, before any of it actually happens.

Warehousing: Putting stock where it is needed before anyone asks for it
A warehouse is only useful if it has the right stock in the right place. This sounds obvious, but it is one of the hardest problems in logistics. Demand does not spread evenly across a country. Festivals, sales events and even the weather can shift where people are buying from, sometimes within days.

Kamal Kishore Kumawat, Cofounder and Chief Technology Officer at Edgistify, explained how his company deals with this. Edgistify runs a system called EdgeOS, which studies order patterns and seasonal changes across more than 75 warehouses. It can predict demand down to a very local level, even a specific postal area, and it takes into account things like festivals, online sale events and growth in smaller towns and cities.

Kumawat described the benefit of this in plain terms. "That lets us pre-position inventory closer to demand hotspots before a surge hits, instead of scrambling once orders spike."

That lets us pre-position inventory closer to demand hotspots before a surge hits, instead of scrambling once orders spike
Kamal Kishore Kumawat, Edgistify

He also spoke about a bigger change this is enabling. As customers expect same day and next day delivery more often, companies are moving away from a few large central warehouses towards many smaller ones spread closer to where people actually live. Kumawat said this kind of distributed setup would be too expensive and too inefficient without good forecasting. With it, stock can sit exactly where it will be needed next, rather than sitting somewhere convenient and hoping for the best.

Why data matters more than the algorithm itself
Here is something worth understanding clearly before we go further. All of this AI forecasting depends entirely on the quality of the information feeding it. If the data going in is messy, outdated or wrong, no amount of clever software can fix that. This is one of the most important ideas behind the whole shift, and it is worth taking seriously.

Optiflux made this point directly, arguing that the real barrier holding companies back is not a shortage of data, but a lack of trust in it. Shrinath Dakare, Co-founder and Chief Executive Officer of Optiflux, said many companies have data spread across different systems that do not talk to each other properly, with no single team owning it, and no shared way of defining what the numbers actually mean.

He put it this way. "Without clean, governed inputs and skilled interpreters, even sophisticated AI simply automates existing chaos faster."

Without clean, governed inputs and skilled interpreters, even sophisticated AI simply automates existing chaos faster
Shrinath Dakare, Optiflux

Dakare also pointed to where this is heading next. Rather than just predicting what might happen, the newer systems are starting to recommend what to actually do about it, such as suggesting which warehouse should send stock to another, or which delivery route to change ahead of expected traffic. He believes the next big step will be systems that can reason through a problem the way an experienced planner would, weighing up different pieces of information together rather than looking at them one at a time.

Seeing every single item, not just the overall stock count
RFID is a technology that lets a business track individual items using small tags, rather than relying on someone counting stock by hand. S R Srinivasan, CEO of QodeNext explained why this matters so much for forecasting.

Historically, one of the biggest problems in warehousing has been a gap between what a company believes is happening with its stock, and what is actually happening. A spreadsheet might say something is in stock, when in reality it has been misplaced, delayed or already sold. Srinivasan said this gap is exactly what item level tracking closes.

He explained the value simply. "It is essential to combine the two tools mentioned above, as AI can offer the right action, but with inaccurate data, it cannot be performed correctly."

Srinivasan also made a broader point about where forecasting is heading. He does not believe the future will be decided purely by better algorithms, nor purely by better data. He believes it will come from the two working together, with each one making the other more useful.

Learning from every truck on the road, in real time
Roadcast works with large fleets, collecting data continuously from vehicles on the road, including GPS location, driver behaviour, engine health and delivery progress. Rahul Mehra, Co-Founder of Roadcast, explained that the real value of this data is not simply watching where a truck is right now, but using patterns in that information to know what is about to go wrong.

Mehra referred to research from McKinsey and Company, which found that AI based forecasting can reduce supply chain forecasting errors by between 20 and 50%, and can cut losses from stockouts by as much as 65%. These are large numbers, and they show why so many companies are investing in this area.

Mehra summed up the shift happening across the industry. "Instead of reacting to disruptions after they occur, organisations can proactively adjust schedules, reallocate resources, and maintain delivery commitments with greater consistency."

Instead of reacting to disruptions after they occur, organisations can proactively adjust schedules, reallocate resources, and maintain delivery commitments with greater consistency
Rahul Mehra, Roadcast

On which single use case delivers the best returns, whether that is predicting delivery times, planning routes, predicting maintenance needs or planning capacity, Mehra made an important point. These should not be treated as separate tools competing with each other. The real value comes from using all of them together as one connected system, rather than picking just one and hoping it solves everything.

The view from a company that buys logistics, rather than sells it
So far, we have mostly heard from companies that provide logistics services. It is worth also hearing from a company on the other side, one that relies on logistics partners to get its products to customers.

N Vijay Kumar, Senior Director for Supply Chain, Logistics and Operations at Acer India, explained how quickly demand for technology products can shift. A new laptop launch, a big online sale, or even the start of a school term can change what customers want almost overnight. Acer now looks at far more than just past sales figures. The company studies how products are sold through different channels, how partners are behaving, and even pricing sensitivity, to work out where demand is about to move.

Kumar explained the benefit clearly. "A more predictive approach allows us to plan distribution more efficiently across online marketplaces, retail stores, enterprise channels and regional partners, ensuring that inventory movement is aligned more closely with real market demand."

He also had a clear message for logistics partners working with manufacturers like Acer. He said partners need to share far more than just where a shipment currently is. They need to flag early warning signs, such as a warehouse running low on space, a route becoming congested, or a delay caused by weather, well before it actually affects a delivery. Kumar believes companies that can offer this kind of early, honest information will become far more valuable to manufacturers than those who simply move boxes from one place to another.

What this all adds up to
AI has not solved forecasting completely, and it probably never will completely. Unusual events, sudden shocks and simple human unpredictability will always exist. What has changed is the starting point for making decisions.

Procurement teams are beginning to prevent shortages rather than just respond to them. Delivery networks are learning from what actually happens on the road, not just estimating it. Warehouses are moving stock ahead of demand instead of after it. Fleet operators are trying to catch problems with vehicles and routes before they cause delays.

If there is one thread running through this whole shift, it is this. The algorithm is only ever as good as the information behind it. As more companies, from manufacturers to warehouses to delivery fleets, start sharing better and more timely data with each other, the shift from reactive to predictive logistics will not be led by any single business working alone. It will be built together, one honest data point at a time.