Mahesh Vemula

Identifying the real constraint in any operational process is one of the hardest parts of improving a process. People often focus on symptoms like delays, spikes in workload, or complaints, but in reality, the problem might be deeper somewhere in data patterns, policy rules, decision sequencing, or how the system works. AI has a big advantage here because it looks at processes without being influenced by internal assumptions, past biases or any emotional reactions. AI can often find limits that people miss by looking for patterns in millions of records, simulations, and contextual signals. In the air cargo domain constraints are rarely static when it comes to pricing, capacity planning, booking, etc. AI can find these hidden blocks such as rule interactions, operational variability, decision delays, or capacity leakages, areas where human judgment alone is often not enough. Here are few detailed examples where AI was able to find the real constraint in processes better than us: 1) Performance of Booking request We thought that the high API response time was because there were a lot of requests. AI's study of booking requests found that 41% of problems happened when there weren't many requests. The request traffic wasn't the real problem; the business rules were too complex. Some combinations, such as pharma with weekend restrictions and embargoes, made the validation logic longer to process. The real constraint was the business rule settings complexity. 2) Capacity Planning We assumed that flights with low load factors are not constraints, but lack of demand is the issue. AI simulations found that inconsistent early-week bookings created artificial constraints. A lane with an average load factor of 78% showed Monday booking spikes 3.5 times higher than other days causing last-minute rejections despite overall capacity availability. The real constraint was booking variability and not the load factor. 3) Pricing Decision Sales team believed that price is the biggest constraint and customers often reject due to high rates. Based on the AI Observations, booking logs over 100K transactions showed that urgent shipments accepted higher rates as long as confirmation was faster. The hidden constraint wasn’t price, but it was slow confirmation workflows and approval dependencies. Again here the real constraint was overall latency in the confirmation process and not the pricing strategy. Here is a quick comparison of human judgement over AI based on various aspects: Aspect Human Judgement AI supported approach Finding Hidden Patterns It depends on experience and might miss constraints that aren't obvious. Finds patterns in millions of data points and spots small problems. Assumptions and Bias Habit, gut feeling, and departmental views all played a role. Not affected by cognitive bias; looks at pure data. Speed Slow, and often needs workshops or manual review. Fast analysis and simulation in real time. Inter-dependencies It's hard to see multi-factor constraints. Easily deals with interactions between more than one variable, like rules, capacity, and behavior. Adaptability Constraints are often updated late and in a reactive way. Always learns and see how rules are changing. Making decisions and interpretation Has a good understanding of the context and thinks about edge cases. May get the wrong idea about cases that don't have enough context or policy detail. Scalability Limited by how much people can do. Easily scales the data of the whole organization. So, what are the pros and cons of identifying the real constraints: Pros Cons Finds hidden limits that people can't see. Without context from the domain, AI might not understand patterns correctly. Gets rid of bias, feelings, and tribal assumptions. Needs data that is clean, consistent, and of high quality. Can handle complicated interactions with large datasets. AI can't fully grasp the subtleties of contracts or regulations. Gives you measurable information to help you make decisions. Teams may not want to hear insights that go against what they have always believed. Allows for ongoing monitoring instead of just one-time analysis. Relying too much on AI could make people less able to make important decisions. Here are some suggestions: Use AI for early detection and humans for understanding the true constraints. Allow AI to analyze the usual process flows and test alternative scenarios. Make a "Challenge Framework for Assumptions" like every time AI finds a constraint, it should make you think about the beliefs or operational policies that go along with it. Use data from multiple dimensions, such as operational, behavioral, and system. Air cargo restrictions can include things like how prices work, how people book, how well ratings work, and how to plan for capacity. Invest in Explainable AI and it helps teams figure out why AI flagged a certain constraint, which makes it more likely that people will use it. Continuously train the AI models so that it continues to learn the market changes. AI can make it much easier for us to find the real constraint in any process. People tend to focus on symptoms and obvious problems, whereas AI looks at the whole data ecosystem without any biases. In air cargo, where pricing, capacity, booking patterns, or performance can all change, AI helps teams find bottlenecks they didn't know were there. AI and people working together can get the best results. This combination helps us to understand how things work better and make decisions that are more resilient and based on data. This is exactly what modern cargo operations need.

Based on my experience, AI is great at peeling back the layers of operational truths we take for granted in air cargo, where decisions about pricing, routing, and loading depend on things like "demand always spikes predictably" or "partners share data reliably." These blind spots can cost millions in lost profits or delays. Unspoken beliefs that teams accept as fact just because "that's how it has always been" affect every part of an air cargo operation, from pricing to forecasting to booking to capacity planning. These assumptions are often hidden until the system acts in an unexpected way or there are problems with performance. AI, especially conversational agents and pattern-based models, can bring these hidden beliefs to light by questioning long-held operational shortcuts, showing where perception and data don't match up, and checking to see if our internal logic still matches real market conditions. Studies show that over 60–75% of process deviations in airlines originate from outdated assumptions or undocumented knowledge. Here are three specific air cargo processes where unspoken assumptions are hiding, and AI brings them to light through things like anomaly alerts, simulations, or counter-proposals. 1) Cargo demand is stable on key routes Process Area: Forecasting & Capacity Planning Most cargo teams think that demand for general cargo will stay the same on long-established routes (like FRA–JFK and DXB–LHR). Planners often make predictions with a small range because they feel that demand fluctuations will be minimal on these routes. How AI Exposes the Assumption When an AI forecasting model runs multi-factor simulations (seasonality, competitor schedules, macro-economic indicators), it often identifies unexpected volatility caused by: sudden e-commerce surges, geopolitical restrictions shifting flows, competitor network changes, regional promotions from major forwarders. The AI’s prediction variance may be far wider than the historical forecast range used manually. Revealed Hidden Assumption “General cargo demand doesn’t change much” is outdated. Demand does swing, we just stopped noticing because our forecasting process assumed stability. 2) Customers would rather have lower rates than faster confirmation Process Area: Cargo Pricing and Booking Behavior Teams often think that forwarders care more about price than anything else. The idea that rates are the strongest lever guides pricing decisions. How AI Shows the Assumption An AI agent looking at thousands of booking patterns might show that: Forwarders with high shipment urgency will pay more if they get confirmation right away. Some small and medium-sized businesses would rather have consistent rates than the lowest rate. Some customers value capacity assurance more than discounts. An AI-powered chat agent might also notice that users keep asking for "quick confirmation" instead of "lower price." Exposed Concealed Presumption It is too simple to say that "price is the main factor in customer decisions." AI shows that in real life, speed, certainty, convenience, and contract consistency often matter more than price. 3) Cargo rating process takes longer during peak load Process Area: Revenue Management & System Performance Teams often think that performance problems only happen during certain times, like the end of the month or during sales. How AI Exposes the Assumption AI that monitors performance might find: slow responses during mid-week, non-peak times micro-delays caused by specific origin–destination pairs higher latency when certain product types (e.g., pharma, special handling) are included API sequences that are using unnecessary recalculations. Revealed Hidden Assumption We believed that peak load was the cause, however AI revealed the root cause is more complicated than just one thing, it was a combination of specific origin/destination, logic, data spikes, and business rule complexity. Below are Pros and Cons of using AI to surface hidden operations assumptions: Aspect Pros Cons Decision Accuracy The right choices AI checks data, which helps people avoid making mistakes based on their gut feelings or personal opinions. If the data they are based on is incomplete, noisy, or biased, AI insights can be wrong. Process Improvement Making the process better Improving processes makes it easier to redesign them by revealing hidden problems and inefficiencies. If teams depend too much on AI, they may miss important human or operational details that AI can't see. Root Cause Identification AI can find small, complicated patterns that people often miss, which speed up the search for the root cause. AI might find connections that seem important but don't really cause anything to happen, so people need to look over them very carefully. Customer & Market Understanding It shows how real customers behave and what they like, which goes against what people used to think. AI can make mistakes when it comes to behavior if it doesn't know the cultural, contractual, or relationship-based context. Organizational Learning Learning promotes a culture of always learning, asking questions, and using data to improve things. Teams may not want to use AI results if they go against established practices or strongly held beliefs. Operational Efficiency Automates the process of analyzing large datasets, which saves time and speeds up the decision-making process. If teams start to rely on AI instead of using it as a tool to help them, they may not be able to think critically as well. Here are few recommendations: Use AI in weekly reviews of prices, forecasts, and performance to keep finding gaps in your assumptions. Make a "Challenge the Assumption" dashboard that shows where AI and human predictions are different. Combine AI insights with domain knowledge to make sure that both points of view are taken into account when making the final decision. Use conversational AI to find undocumented workflows or "tribal rules" that analysts use but have never written down. Teach teams about cognitive bias so they know why people make assumptions and how AI can help break down old ones. Validate AI anomalies as soon as possible, because they often show the most useful information. AI is more than just a way to automate tasks. It is also a way for organizations to look inward. AI helps find the hidden assumptions that guide decisions in air cargo operations by bringing to light differences from long-held operational beliefs. AI makes teams face facts that data shows but people often miss. This is true whether they are predicting volatility, customer behavior, or performance changes. Organizations that take on this challenge will grow faster, create processes that are more flexible, and make decisions based on facts instead of tradition.

Can AI Truly Be Creative — or Does It Just Remix Human Ideas? In my view, AI cannot truly be creative but it is a masterful re-mixer, not an originator. Drawing from experience in semiconductor and telecommunication industry in the past and my current deep dives into air cargo logistics, I have seen AI shine in generating ideas feel innovative, but peel back the layers and it is always recombining human patterns looking at research papers, trial data, governance frameworks and stories from ops floors. Everything it produces, no matter how impressive or new it looks, is built by intelligently combining patterns it has seen in human-generated data. True creativity belongs to humans because only humans can step outside existing data, driven by emotion, curiosity, or pure stubbornness. A spot-on creative task in my domain is generating improvement ideas for real-time process problems, for example yield crashes in fabs or capacity crunches in cargo holds. This isn’t about simple optimization but it requires envisioning adaptive, multi-stakeholder fixes that balance trade-offs like cost, compliance, and resilience, often in high-stakes chaos where one bad call costs millions. In my experience, AI's creativity doesn't show up in abstract art tasks. Instead, it shows up in how it re-frames operational problems, suggests alternatives we might not think of right away, and creates scenarios that get teams to think of new ideas, again all this based on the past human-generated data. Here are three examples from the real world that show the difference between "looks creative" and "truly creative." 1) Air Cargo - Protecting Shipments That Go Bad in Severe Weather Problem: A storm in Europe suddenly makes 38% less room for cargo on flights to London. Medicines and flowers could go bad. AI-made creative solution: Send partners a short, safe message that says, "Available space decreased by 38% from 13:00 today for 18 hours. Let AI agents automatically re-book shipments to Amsterdam, trade slots with other airlines, and hold a bid for the space that is still available. 95% of temperature-sensitive cargo was delivered on time Each airline group saved between $10 million and $15 million in fines and lost goods. Though the above looks creative, but every part is a remix of human ideas: ONE IATA came up with record standards, the EU fixed noise injection, and yield pools were first used by passenger airlines in the 1980s and then copied to cargo airlines. 2) Semiconductor Fab: How to Fix a Sudden Drop in Yield Problem: A lithography machine drifts, and the defect rate goes up by 25%. This is very costly during a chip shortage. Creative solution made by AI: Don't give partner factories the secret recipe, just tell them the result ("defect signal +15 % on layer 7"). Automatically adjust machine settings in real time and move wafers to backup lines. Results from 2025 fab deployments that have been reported: Yield errors went down by 30–50% Time to fix defects cut by 20% Each mid-sized factory saves between $50 million and $100 million a year Once more, impressive but just a remix: federated learning (taken from banking fraud systems), real-time tuning (standard predictive maintenance), and capacity trading (taken from air cargo yield pools). 3) Recovering from a telecom network outage Problem: Over 100,000 customers are affected by a fiber cut. Message and action from AI: "We've already borrowed spare network capacity, full service will be back by 20:15." Get a free 10 GB gift. Result: Customer satisfaction stays high, and the risk of losing customers goes down by 12–18%. The tone, gift size, and timing are all taken from millions of other scripts and retention studies written by people. Some of the Pros and Cons of the AI creativity includes: Aspect Advantages of AI (Remix) Limitation of AI (No True Creativity) Speed Seconds instead of days Cannot imagine solutions never seen before Scale Handles thousands of variables at once Misses rare events outside training data Consistency Same high quality every time No emotional insight or ethical gut feel Cost Saves millions Risk of hidden bias or accidental collusion Novel combinations Often beats human teams by 10–30 % Still only combinations, never a genuine invention So, suggestion for best results is human and AI partnership: Let AI generate 50 ideas in seconds and then humans pick the 3 most promising. Always keep a human approval step for high value decisions. Regularly audit data and add deliberate “noise” so AI does not copy bad patterns. Use open standards (like ONE Record in cargo) so small companies can join. Run cross-industry experiments, many of the best “new” ideas are simply good ideas moved from one sector to another. AI is already making a difference, with yields going up by 20% to 50%, millions of dollars saved, and faster recovery from problems. These are real numbers from 2025 deployments in the semiconductor and air cargo industries. But every improvement is just a smart mix of things that people have made, standards, research papers, and past trials. AI doesn't make things the same way people do. It doesn't have feelings or intuition. But in real business situations, AI shows a kind of augmented creativity when it comes to coming up with training scenarios, new pricing ideas, or operational hypotheses. People are still the ones who come up with the real creative spark, the idea that has never been in any dataset. AI is not a replacement for us, it is our most powerful partner. When both work together, we can do more than either AI or people could do on their own.

How Will AI-to-AI Collaboration and Competition Reshape Markets? When AI agents from either the same company or different companies start talking to each other about things like setting prices in real time, moving inventory through supply chains, bidding in ad auctions, or working together to improve logistics, markets will stop moving at the speed of people and start moving at the speed of machines. This transition is happening faster in air cargo, where AI-powered systems are already handling dynamic pricing for perishable goods, rerouting shipments when there are problems, and making the most of belly-hold space on passenger flights. These interactions promise faster revenue capture and stronger networks, but they need new rules to keep the market from changing in ways that are hard to predict. Below are a few situations that are starting to come up in air cargo, mostly from the point of view of revenue management, pricing, and planning. Here are few aspects showing how competition, pricing and value creation will change with AI-to-AI collaboration: Aspect With Human Involved With AI-to-AI collaboration Speed Rate quotes and route approvals take from hours to days Rate and route offerings can be done in matter of seconds along with counteroffer and approvals Pricing Manual and time-consuming process with lot of data analysis AI can help negotiate prices in matter of seconds Competition Require contracts and sales relationships for forwarders to work with airlines Smaller forwarders can use federated AI to access the capacity without direct contracts Value creation Based on experience and often reactive planning Shared AI forecasts can improve overall network utilization by 10 to 15% Below are few examples that are already happening in air cargo world: Managing revenue in the flow of perishable goods AI agents from exporters and carriers now automatically bid on and accept space for time-sensitive shipments like flowers and medicines. In a recent rollout, systems from multiple airlines worked with forwarder platforms to predict demand spikes (like holiday produce rushes), dynamically allocating belly space and changing yields. This led to load factors that were 20–30% better and over $50 million in annual revenue across a mid-sized network. Negotiating dynamic pricing for spot rates Freight forwarder AI platforms now handle multi-round request for quotes (RFQ) where agents from different carriers can make counter offers in real time, taking into account lane-specific factors like fuel surcharges and cut-off times. This has cut the time it takes to negotiate from 24 hours to less than 5 minutes. It has also cut empty leg rates by 15% on transatlantic lanes and stopped people from bidding too much when demand is high. Planning a network for rerouting multiple carriers During the weather problems in Europe in 2025, AI systems from airlines and ground handlers worked together to reroute pharma cargo through different hubs (for example, from FRA to AMS), automatically negotiating interline hand-offs and slot swaps. This kept 95% of high-value loads on time, improved network flows, and saved each major operator $10–15 million in penalties. Making the most of belly space across alliances AI agents from alliance partners (like one-world carriers) now share anonymous data to improve mixed loads on passenger-cargo hybrids. They also bid on more space for e-commerce overflow. There has been a 12% rise in use since deployment, which has turned flights that were not being used enough to make more than $20 million. Predictive Yield Pools for Seasons with High Demand Forwarders and carriers set up AI-powered yield pools where agents guess and trade future capacity blocks for delivering gadgets. This has made peak-season volatility less of a problem in Asia-Pacific trials, and it has also increased total revenues by 8–10% through shared prediction models that take into account changes in trade tariffs and demand. Some of the emerging advantages with AI-to-AI collaboration include: Yield Precision and Resilience: AI-to-AI forecasting reduced the error by 30-50%, which lets you make proactive capacity swaps that can help keeping the revenue stable during disruptions like port strikes. SME Empowerment: Smaller businesses can connect to global networks through plug-and-play agents. They can compete on data insights instead of size, for example, dynamic pricing tools make spot market access available to everyone. Sustainability Gains: SAF rules for 2025 say that optimized routes and loads should cut fuel use by 10–15%. This also makes green premium price levels. New Revenue Streams: Shared AI platforms make money by offering additional services like offering a product like temperature control when shipping fish or flowers, that can add up to 5 to 7% to the value of each shipment. There are many emerging risks as well as ethical dilemmas, some of those are listed below: Using algorithms to send price signals: When there are shortages, agents who interact with each other could unintentionally converge on higher spot rates, which looks like collusion. For example, tests in 2025 showed 10–15% yield inflation on constrained lanes before randomization fixes were added. Network Weakness from Too Much Dependence: If one agent makes a series of mistakes in their forecast (for example, misreading how tariffs will affect things), it could cause a lot of reroutes, which would make delays worse, like the Asia-EU corridor glitch in Q1 2025. Bias in Yield Distribution: Training data that favors shippers with a lot of shipments may not be fair to perishable goods or small and medium-sized businesses, which raises questions of fairness. In opaque AI negotiations, who makes sure that everyone has equal access? Data monopoly risk: Big forwarders could use exclusive datasets to get ahead of their competitors that could keep their market shares giving them advantage. Regulatory Gaps for Exclusion of People: By the end of 2025, more than 70% of B2B bookings will be fully automated. This could leave behind people who don't use AI, like old-school small and medium-sized businesses, which could lead to discussions about making it mandatory for everyone to work together. However, many of the above can be mitigated using the following: Noise Injection in Bids: Randomness in pricing algorithms stops people from working together without telling each other, as shown in EU cargo exchanges. Federated Learning Networks: Agents work together on models without sharing raw data, which keeps privacy while improving group predictions. Oversight Agents: Neutral AI monitors (like IATA compliant ones) flag unusual behavior in real time, and people can stop high-stakes deals. Open Protocols for Equity: Standards like ONE Record make it easy for everyone to work together, so smaller companies aren't left out. Bias audits and simulations: Stress tests for a variety of situations that must be done before deployment, as required by the 2025 AI Act updates for logistics. So to conclude, AI-to-AI collaboration could have a huge impact on markets, making them faster, more dynamic, and more efficient. If done with careful planning, governance, and risk management, this could make logistics, cargo, and supply chain industries much more efficient and profitable. If we don’t, we might have algorithmic monopolies, supply cycles that aren't stable, or unfair competition.

The ability of AI systems developed by different organizations to collaborate has become a practical requirement rather than a desirable feature in a world where businesses rely more and more on AI. Setting up a system that allows partners or competitors to share just the information they require while safeguarding trade secrets, consumer privacy, and legal compliance is definitely not an easy task. Some of the important things to be considered are: Data format compatibility: Information is stored and organized differently in every business. If formats are not standardized, as an example, one system may display 95% probability, while another system might expect the value as a decimal e.g. 0.95. Even these kinds of minor discrepancies lead to major mistakes. Data Governance and Trust: Companies need to agree on how open they will be and how to check each other's work to keep the trust. It is not negotiable to have clear contracts, audit trails, and common ethical standards. Security and Privacy: Though data sharing is required, at the same time, it can be risky if any personal information is being shared. Strict access controls, data masking, and encryption are very important. To give few examples based on my experience working with various systems, Airlines and its partners (specifically for Cargo): Airlines might be using revenue management AI models to determine the cargo capacity and pricing, while logistics partner(s) can be using different AI models for freight routing and overall cargo belly space optimization. Both models must respond quickly when inclement weather unexpectedly reduces the amount of cargo space available. As an example, the airline sends a brief, structured message instead of giving over the entire algorithm e.g. “Effective 14:00 IST, cargo capacity on routes BLR-LHR reduced by 38% for next 18 hours.” The partner's system can reroute shipments and modify bids without ever knowing the airline's pricing logic. Just like weather models predict how severe and how often storms might occur without exposing all the underlying science, AI setups can surface risk-weighted signals that help teams make decisions together. Financial Services: Multiple banks can team up to spot fraudulent transactions using AI, where each bank can use different models and utilize their own bank’s private data on which models are trained. In this case federated learning techniques can be used where models gain knowledge together without exposing raw data. Just like a project charter establishes scope, constraints, and expectations at the start, federated AI teamwork sets the data boundaries, compliance rules, and shared goals needed for collaboration. Supply Chain Networks: For example, makers and sellers of automobiles use different AI models to optimize their inventory. Overstock and stockouts can cost millions of dollars for both makers and sellers, instead auto makers can publish few standard indicators such as lead time variability, forecast accuracy and stock levels using secure APIs, where the API hides maker’s precise cost structure or the manufacturer's upcoming model launch schedule. These standard KPIs through APIs meeting ISO standards ensure quality across suppliers. Also, along with use of APIs, it is also important to explain the reasons behind it, for example, port congestion in Vizag increase delay risk by 60%. To summarize the following best practices to be followed for AI systems from different companies to collaborate effectively: Follow standard protocols to stick to accepted methods to share the data. Offer reasoning behind decisions without giving away sensitive or proprietary details. Establish a clear governance to set clear roles and responsibilities, and outline who should take action if any issues come up. Last but not least, keep human involvement to ensure people verify and approve any major decisions. AI systems from different companies can definitely achieve collaboration by sharing useful information while keeping sensitive intellectual property as private. Consider it as a trust layer between AI setups, like how Lean Six Sigma quality frameworks ensure the same standards are met without exposing every detail.