A UAE-based trading firm spent about $12 million a year on freight across sea, air, and land shipments, but the vast majority of booking decisions were still made via email chains, spreadsheets, and manual quote comparisons. Each shipment sparked a well-understood choreography: send out requests for quotes to a small group of carriers and forwarders; wait 24-48 hours for responses; open PDFs and email attachments; normalize the charges manually; and then guess which was really cheapest based on freight, THC, documentation fees, surcharges, duration, and local charges, considering all.
It was not only the speed that was the problem. It was the decision quality. Various carriers had different ways of structuring the charges, rates fluctuated rapidly, and the team was unable to identify consolidation options at the overall level of shipments. A quote that looks cheapest on paper is frequently not the lowest landed cost. A pricier service could prove cheaper once you account for the risk of delay, the cost of working capital, or local delivery time.
At aTeam Soft Solutions, we developed an agent that gathered, extracted, normalized, compared, and ranked freight quotes in real-time, then extended into booking, consolidation, timing recommendations, and shipment analytics. The result: a 15% decrease in average freight costs per shipment, $1.8 million in annual savings, booking decisions that are made ten times faster, and a logistics staff that finally has time to think about strategy rather than e-mail administration.
Before building out the platform, we spent time observing how the company was actually buying freight. What we discovered was a process that looked orderly from afar, but was intensely manual, disjointed, and frustrating to scale under pressure.
The company processed over 3,000 shipments annually. Roughly 70% moved by sea freight, 20% by air, and 10% by land within GCC routes. They were importing from China, India, Turkey, Germany, and Southeast Asia and then distributing into the GCC. So that’s the logistics team, and they’re not dealing with one uniform shipping pattern. They were managing container bookings, LCL consolidations, air cargo scheduling, inland delivery coordination, route-specific carrier preferences, customs implications, and warehousing limitations within several trade routes.
The organization’s team had four logistics coordinators. That seems like enough for 3,000 shipments on paper. In reality, it was insufficient, as so much of their time went into repetitive freight comparison work.
A new shipment had to be booked, and so the coordinator began by collecting the essential information: place of origin, place of destination, weight, CBM, type of commodity, level of service, urgency of delivery, and any special handling requirements such as oversized cargo, dangerous goods, or temperature sensitivity. Then they sent out e-mails to freight forwarders and shipping lines to get the best rates. They typically reached out to five to 10 suppliers, according to route and urgency.
This section only added to the delay because the company was constrained by how many quote requests a user could realistically send and track. If a single coordinator was busy or a shipment was late in the day, the request cycle slipped. Then, all that waiting period. Quotes would arrive in the next 24 to 48 hours in a mixed array of formats. Certain carriers responded directly within the body of an email. Others sent out PDFs. Some of them used Excel sheets. Some provided all-in rates. Others broke down, separating every component.
That’s when the real difficulty in comparing started.
One quote might show only ocean freight plus a handful of surcharges. Another may have THC included in the line haul. Another might break out bunker adjustment, currency adjustment, documentation, local delivery, and customs fees in a different format. One forwarder might look cheaper until you realize a crucial local charge had been left out or bundled differently. And another might be more expensive in base freight, but cheaper once local handling is factored in.
So the coordinator had to manually normalize the quotations in Excel.
That seemed manageable for one shipment, but the company was repeating this over and over again across hundreds of moves. And because freight is volatile, the comparison had to be done quickly. A rate that was quoted on Monday may not be valid on Wednesday. A low spot rate might evaporate once a vessel is full. There could be a sudden spike for the season. Before booking, the team hardly had time to revisit rates, meaning some bookings were effectively going out on stale commercial assumptions.
The other issue was that “cheapest” was rarely a good choice on its own.
The logistics team needed to consider the transit time, inventory carrying cost, warehouse receiving schedules, promised customer delivery windows, and internal stock planning. A 35-day ocean option might look great on pure freight cost, but if the goods were high-value or time-sensitive, that slower route variation could generate more hidden cost than a 21-day service. The coordinators understood this instinctively, but they had no systematic means of pricing those trade-offs into every decision.
The process lacked visibility between cross shipments as well.
Coordinator A could be booking an LCL from Shanghai on Tuesday. Coordinator B might place an additional LCL from the same area two days later. Each person was maximizing their own shipment independently, because that’s what the workflow led them to do. No one had a live system to view the two shipments side-by-side that might have prompted the obvious question: Should we consolidate these into an FCL? This meant the company was losing out not just on route-by-route savings but also on network-level savings.
The team’s spending of its own time was an indication of how acute the problem was. About 60% of their time was dedicated to rate shopping and booking administration. Strategic work such as reviewing carrier performance, optimizing routes, coordinating with warehouses, and negotiating contracts, could now only be undertaken for about 40% of the time. A company that spends $12 million a year on its freight, that imbalance made a difference.
Here’s what caught our eye at aTeam Soft Solutions. The company was not just after speedier quote gathering. What it wanted was a genuine AI freight cost optimization workflow that could standardize chaotic rate inputs, intelligently compare total landed cost, look across all future shipments, and get the logistics team out of perpetual spreadsheet reconciliation.
The company had been examining freight technology previously as well. It was not that there was an absence of software on the market. The problem was that the majority of tools addressed only a fraction of the decision.
Tactically, the freight rate platforms could help give useful market benchmarks and some carrier access, but they were not representative of the full array of providers that this company used. That gap was significant because numerous regional Middle East airlines, as well as smaller consolidators, could still be found offering competitive rates through email workflows rather than full API connectivity. A platform that only sees the global carriers is not seeing the entire buying landscape.
Even when platforms did quote rates, they frequently didn’t normalize costs enough. Freight comparison is not just a headline rate exercise. The actual choice is found in all the fees around the shipment: THC, paperwork fees, BAF, CAF, local delivery, customs brokerage, storage risk, and maybe duty implications if one routing strategy affects classification or clearance cost downstream. Conventional tools are frequently sufficient for the purpose of finding rates. They fall short of depicting a company’s true total landed cost in the context of that company’s own operations.
The internal considerations of the company were important too. They had preferred carriers and negotiated discounts, they had volume commitment opportunities along specific routes and route-specific reliability histories, and they had internal scheduling constraints related to scheduling work in their warehouses. A standard rate platform doesn’t know what would be best for an organization unless it knows those organization-specific conditions.
That’s why the answer had to be an AI agent that was embedded in the company’s own freight workflow, and not simply a market lookup tool. We had to develop a system that could take actual quotes from the mentions, normalize, compare, think in landed-cost terms, and then turn around and recommend action in terms of cost, speed, trustworthiness, and consolidation potential. That’s the reason this became a true shipping rate comparative AI agent use case.
We rolled out the rollout in phases, as the logistics team was first required to have trusted visibility to the freight cost and quote structure before they would trust booking recommendations. So we began with rate intelligence, then expanded into automated quote solicitation and comparison, followed by cost optimization and booking, and ultimately lifecycle tracking and analytics.
The initial stage was all about one hard-bitten realization: the company didn’t have a single source of truth for where freight rates really lived.
Some of the carriers had APIs. Others didn’t. Some quotes were sent by email from forwarders who each time formatted the information in a different way. For the AI agent to be able to compare quotes properly, it had to see the rates in a consistent manner first.
We connected directly with major carrier rate sources where available and developed an email ingestion layer for smaller forwarders and regional carriers. All quotes came through a normalization pipeline, whether from API, email body, Excel, or PDF. The agent charged by the AI extracted the cost elements and broke them down to a common structure: freight, THC, documentation, surcharges, local charges, and other route-specific items.
That immediately made a big difference, as the logistics staff no longer had to guess if two quotes were really apples-to-apples. They could see a single structured view.
We also kept every quote historically. That is, the company was starting to create its own memory of freight rates by route, carrier, date, service level, and cost structure. Rather than considering every shipment as a new comparison from scratch, the AI agent was able to view the current market through a historical lens. That was useful later for timing and bargaining decisions.
This stage’s solution dashboard provided a clean view of today’s rate choices as well as evolved trends, route habits, and carrier performance information. Even before the booking logic had been automated, the company could at last view the shape of its freight market as it really looked, all in one place.
To the company, this is one of the first indications that an agentic AI logistics UAE system is functioning: the industry shifts from fragmented quotes to organized freight intelligence.
After the rate layer was stabilized, we proceeded to the actual quote request process.
When a new shipment was created, the AI agent selected the relevant carrier set for that route according to historical usage, route fit, reliability, and service availability. Then it automatically submitted quote requests to a wider group of carriers than the coordinators could realistically handle manually. Rather than five to ten carriers, the company could now query fifteen to twenty where appropriate.
When the responses are in, the AI agent immediately extracts and normalizes the pricing. The team didn’t hold their breath until someone could build the spreadsheet. The comparison view built itself dynamically.
But the real value here was not just speed. It was about the decision quality.
The comparison did not order options solely by headline freight cost. It displayed total landed cost, transit time, reliability score, available sailing or booking dates, and any special terms. A value score that balanced cost, speed, and reliability differently by type of shipment was also introduced. For extremely time-sensitive moves, the model put greater weight on service speed. When the shipment is in bulk or not critical, it’s weighted for cost efficiency.
It was relevant that the logistics decisions have a tendency not to be one-dimensional. A less expensive route that throws off inventory planning or customer commitments isn’t really less expensive. By presenting this trade-off explicitly, the AI agent made the ranking significantly more useful from a commercial perspective than a lowest-price table would have been.
After the coordinator looked over the ranked comparison, the booking could take off much faster. What used to be a full day or two of back and forth could now be moved out for a controlled decision in a fraction of the time.
The third stage is where the bulk of the savings was realized.
We enhanced the AI agent from passive comparison to active opportunity detection. If it observed a meaningful rate decline on a route the company regularly used, it notified the logistics team with quantified savings potential. That gave the team the ability to book efficiently where timing allowed.
But the greatest value was not in identifying the single lowest price for one shipment. It came from looking at all shipments underway.
The AI agent scanned the whole upcoming shipment pipeline of the company and found consolidation opportunities. In the case of several LCL shipments from the same origin region appearing to move with a compatible schedule, the system would inquire if a combined FCL move is more cost-effective and operationally feasible. And it also verified that the destination warehouse could handle the combined flow on the same day.
This was among the largest commercial breakthroughs in the project. Earlier, coordinators used to optimize within their own shipment queue. The AI agent optimized for the whole company queue. Those generated savings that humans could not easily see, because work was divided operationally.
We also applied the same logic to carrier contract negotiation. By consolidating route-by-route volume, the AI agent could indicate where carrier consolidation would likely result in discounts. Rather than negotiating from a vague sense of yearly spend, the company could negotiate from route concentration, timing patterns, and realistic volume distribution scenarios.
Another useful aspect of this stage was the inclusion of booking timing intelligence. Based on the rate history on certain dates in the past, the system started to detect periods when fares for specific routes usually drop. Still, not every shipment that didn’t rush should wait. Yet for non-urgent freight, the AI agent could suggest more optimal booking timing based on real pattern data rather than subjective market feel.
This is where the project went beyond being just a freight management automation tool. It turned into a true freight-spend optimization engine.
After the quote and booking optimization were working, we extended the system to track shipment and cost analytics.
The AI agent tracked the vessel and shipment status, port congestion signals, and carrier updates so the logistics team had live ETA visibility instead of dispersed carrier communications. This has enhanced the ability to collaborate operationally with receiving and downstream planning of delivery.
We’ve also been able to automate much of the shipping document process. Bill of shipping instructions, packing list support, and related shipment file preparation can now be derived from the structured shipment and PO information. This cut documentation errors drastically since the AI agent was working with structured data instead of repeated manual entry.
Freight cost analysis was also greatly enhanced. The company could now track quoted cost against final cost, carrier surcharge patterns, route trends, and spending by business line or product category. We also embedded the landed-cost model with customs and duty logic, so that freight decisions were not being made in a vacuum relative to the cost implications of importing.
At this point, the system was not just a quote comparison tool anymore. It was operating as a real-world intelligence agent for buying, booking, executing, and controlling the cost of freight.
We developed the core platform in Python with FastAPI, leveraging Celery and Redis for asynchronous processing of quotes, booking workflows, rate polling, and notification logic. Structured commercial and shipment data was handled in PostgreSQL, and TimescaleDB captured route-rate history and time-series pricing behavior to enable meaningful trend analysis and booking-window recommendations.
Claude did the dirty work of the workflow: reading email-based freight quotes, understanding natural-language charge descriptions, and assisting in cost component classification when provider formatting differed. Since rate entries came in HTML, PDF, Excel, and plain text, we developed a multi-format extraction pipeline that converted all information into a single freight cost schema.
The React dashboard provided our logistics team with a single operational screen to compare quotes, make booking decisions, identify consolidation possibilities, rate alerts, and monitor shipments. We integrated directly with the APIs of the largest carriers where they exist, and applied strong email parsing and monitoring in the absence of APIs.
Among the more valuable parts on the backend was the landed-cost engine. We didn’t stop at shipping. We factored in local fees, customs brokerage, and duty-related elements when applicable, such as links to HS code and customs logic. That’s what enabled the AI agent to sort options by real business cost as opposed to just line-haul rate.
At aTeam Soft Solutions, we believe such integration is a must for any real AI freight cost optimization solution project. A logistics decision is only as good as the underlying cost model.
The first major hurdle was quote extraction. Freight forwarders and carriers don’t submit rates in a uniform way. Some feature plain tables. Some conceal essential fees beneath paragraphs. Some combine multiple legs on one page. Others are inconsistent in their use of the abbreviations. We designed the extraction layer to accommodate these differences, and also to stop the uncertainty. If the program was unable to determine if a line was THC or documentation, it would throw a warning about an ambiguity rather than fabricating a structure.
Reliability of the API presented a further challenge. Carrier APIs are great, but they aren’t perfect. Rate limits, timeouts, stale bookable inventory, and response inconsistencies all introduced risk into the system. So we built in some fallback logic so the system could still function if one carrier source went unreliable.
The landed-cost model was also a bit more involved than it seemed at first. Freight was just a piece of the economics. Duty rates, short-haul trucking, container handling, customs brokerage, and receiving restrictions had to be taken into account. Without that, the ranking logic would have been too simple anyway.
But the key challenge was organizational rather than technological. The largest savings were from consolidation, and consolidation requires earlier visibility across shipments. So that meant coordinators had to share data about shipment planning earlier than they were accustomed to. The AI agent revealed the opportunity, but the company needed to change its internal planning behavior to realize full value.
The average shipping charge per shipment declined by around 15 percent on all routes. This added up to an estimated $1.8 million in annual savings on a yearly freight expenditure of about $12 million.
Quote comparison time went from 24-48 hours to around two hours, since the system performs request sending, response collection, and normalization much faster than when users do these tasks manually. That speed was important not only for the labor savings but also because the firm could lock in competitive rates before the market moved.
Consolidation accounted for a large portion of the value. In the first year, the company saved approximately $280,000 from FCL optimization by intelligently consolidating separate LCL bookings. Results of contract negotiations were also better, with some $320,000 more in negotiated discounts due to improved route-volume concentration and data-led carrier discussions.
The logistics department team became more efficient. The same four coordinators could handle an additional 50% of shipments, and their time mix shifted dramatically. Instead of spending the bulk of the week collecting and comparing rates, they’re now devoting a lot of their energy to strategic carrier management, route planning, and exception handling.
Rate shopping coverage also improved materially. Rather than contacting five to ten carriers, the team was up against as many as fifteen to twenty providers in some cases, as the AI agent took on the burden of administration. That broader net led directly to better decisions.
The booking process went from three to four hours per shipment to around half an hour. Errors in paperwork were reduced by approximately 80% as the system produced a single set of standardized shipment documents using complete structured data, rather than manually entering the same data repeatedly.
For the company, one of the most interesting results was how rapidly the company’s career choices became more evidence-based. Timeliness, surcharge behavior, and real route economics were no longer based on rumor. They were quantifiable. That’s the difference between a transport buying function and a logistics intelligence function.
The platform we developed in Python with FastAPI for the backend orchestration, Celery and Redis for asynchronous task execution, PostgreSQL for operational shipment and carrier data, TimescaleDB for freight-rate time-series storage, Claude for email and document-based rate extraction, React.js for the logistics dashboard, direct integrations with major shipping-line APIs where available, a custom email parsing and normalization pipeline for forwarder quotes, and AWS EC2, RDS, S3, and Lambda for infrastructure, storage, and scheduled execution.
The biggest takeaway was that the rate comparison alone doesn’t create the biggest freight savings.
Early on, it was tempting to say the main value would be comparisons among more carriers more quickly. That did produce savings. But the greatest value was in shipment consolidation and timing intelligence. When the AI agent had visibility into every upcoming shipment across the entire company, it saw opportunities that no single coordinator could see from their own desk.
We also found that “cheapest” is not a logistical decision that makes sense on its own. Transit time, carrier reliability, duty impact, warehouse constraints, and customer commitments are all important. This is why we designed the system around value scoring and landed cost rather than just freight rate ranking.
And eventually, we concluded that changing processes was as important as AI capability. The tool was only able to optimize what the business revealed soon enough. After the team began sharing shipment planning information earlier, the recommendations from the system began to be more useful. This is where a standard principle for the company in any logistics AI development India engagement comes in: the AI creates visibility, but the operating process must be ready to leverage it.
Trading and distribution companies throughout the UAE and the wider Middle East are regularly aware that freight is a big-cost factor, but they misjudge just how much waste exists within manual quote management, inconsistent landed-cost analysis, and consolidation chances that are missed.
At aTeam Soft Solutions, we develop systems where an AI agent can gather rates, normalize them, compare on the true total cost, and recommend bookings, as well as identify route and consolidation strategies that save real money. That is how logistics teams transition from email-based booking labor to dynamic freight cost management.
