−38.2%total supply chain cost, in the redesigned network
A global life-science manufacturer · antibody supply network
A digital twin of how the company makes and ships antibodies worldwide compared today's network with a simpler design before anything changed. The simpler network cut total supply chain cost by 38.2% and end-to-end lead times by 9.4%.
Grow into the plants you have, or build the next one on evidence.
+49.19%monthly output from the same line in simulation, with the same number of operators
A fast-growing mattress maker · five years of planned demand
A model of the production line tested whether it could meet five years of planned growth. Fixing the padding section, which capped the line, and moving operators to it raised potential output from 15,653 to 23,353 packed mattresses a month, enough for the demand forecast through 2016.
Test a new or closed site before you sign the lease.
−12%total cost in simulation, with 9 warehouses instead of 11
A manufacturer's spare-parts network · 6,800 customers
Fixed and transport costs for its 11 warehouses came to about $13 million a year. Simulating deliveries on a digital road map for each candidate layout showed the best answer was 9 warehouses, in new locations, cutting total cost by about 12%.
−20%inventory at a 99% service level, in a pilot study
One of the three largest detergent makers in China · about 400 products
A simulation of its factories and regional warehouses set the safety stock for each product and where it is held. A pilot on one major product category showed inventory could fall by 20% while keeping a 99% service level. The system now supports weekly inventory decisions.
~12%less spare-parts inventory value, with fewer aircraft grounded
A major US airline · 700+ aircraft, more than 40 stations
A simulation set how many of each spare part to hold at each station, against a daily target for aircraft out of service. Total inventory value fell by approximately 12%, simulated out-of-service events by 35%, and simulated service levels rose by 20%.
−4.77%fill rate lost to disruptions, with four weeks of stock
A large consumer products company's North American supply chain
A model of the company's plants, packaging site, distribution centers and one offshore contract manufacturer added its real disruption history. With four weeks of inventory, the fill rate fell by 4.77% to about 95%, thousands of sales units. After a six-week outage at the primary plant, recovery typically took longer than the outage itself.
Network sizeOur network has plants, DCs and hundreds of lanes. Is that too big to simulate?
No. We model the plants, DCs, lanes and stock the decision touches, at the detail it needs, and extend it only when the next question needs more.
Sites, lanes, policiesHow do you model our sites, lanes, reorder rules and lead times?
From your site list, lanes and transit times, reorder rules, lead times and costs. You watch orders and stock move through the network day by day, and every assumption is listed for you to change.
Disruption over timeCan it play out a port closure, a supplier failure or a tariff change week by week, not just as one average plan?
Yes. Close a port, lose a supplier or change a tariff on a given day, and watch stock drain, which customers run short first, and how long recovery takes. Disruptions run many times, so you see a range, not one plan.
Tariffs and moving supplyTariffs keep changing. Can we see what a new tariff does to cost and service before we move suppliers?
Run the network with the new rate and with the moves you're considering, and compare landed cost, stock and fill rate side by side. When the tariff changes again, you change one input and rerun it.
AI agents making decisionsAI agents are supposed to start running supply chain decisions. Where does simulation fit?
Before an agent or a planner changes a rule, you can run that rule against your network and see what it does to stock and service, including in a disruption. Every run is kept, which gives you the traceable record that governing AI decisions calls for.
ERP demandWe have ERP demand history and a lane list. What else do you need?
Stock by site, lead times, reorder rules, capacities, and the costs of holding, shortage and transport. Rough is fine to start, and the model shows which inputs actually change the answer.
Service and stock matchHow do we know it reproduces last year's service levels and stock?
First we replay last year with your real orders, lead times and disruptions, and compare fill rate and stock by DC with what actually happened. Only when they match do we test changes. Every run is kept, so anyone can trace how an answer was reached.
New DC or safety stockWhat's a good first study: a new DC, safety stock, or a disruption plan?
One scoped decision: a new DC, safety stock levels, or a disruption plan. You get the options side by side on your own demand, with the cost and service of each, and a model your team keeps.
Sourcing, inventory, transport and capacity. See the consequence across the network, not just at one site.
Pedestrian
For restaurant, passenger, venue and visitor operations leaders
Fill the venue, move the crowd and staff the counters on evidence, before the doors open.
Airports carried a record 9.8 billion passengers in 2025, Britain’s railways a record 1.83 billion journeys in the year to March 2026, and airports now have to get “far more out of the infrastructure we already have”. ACI World (opens in a new tab) · ORR (opens in a new tab)
Recorded · crowd through the doorsStadiums and arenas
Fill and empty the venue on time.
1 h 47 minto seat 501 fans with a dispatcher, against 3 h 13 min for the slowest policy, in simulation
A 2014 World Cup stadium in Brazil · one stand
A simulation of one stand compared four ways to get fans from the gate to their seats. With a dispatcher sending people to their rows, all 501 fans were seated in about 1 hour and 47 minutes. With numbered seats and no dispatcher, it took 3 hours and 13 minutes.
Gate and screening-lane allocation, concourse layouts, staggered entry times.
What we model
Arrival surges, screening times, groups and people with reduced mobility.
What you get
Entry and clearance times by gate, density hot spots, and the place that limits the whole venue.
Recorded · the platform filling, then boardingTransport hubs
Move people to the train, not into a queue.
7.53 saverage check-in wait, down from 24.3 s
One airline's check-in lobby at a mid-sized US airport
A model of curbside, kiosk, counter and online check-in compared ways to route passengers. Taking check-in off the counter, which then only weighed and tagged bags and handled special needs, gave the shortest waits: the average wait fell from 24.3 to 7.53 seconds, and average time in the system from 158.85 to 129.96 seconds.
43.44%less time in the vote center at best, with no added staff
A Los Angeles County vote center during a presidential primary
A model of check-in, ballot marking and drop-off tested the layout and staffing of a busy vote center. Processing provisional voters separately at check-in cut the average time voters spent inside by between 28.59% and 43.44%. The authors found wait times may be reduced with no additional resources or financial expenses.
31.4%faster passenger time through screening, in simulation
A security checkpoint at a major US hub airport
A simulation of one checkpoint, paired with an optimizer, set how many screening lanes and ID-check stations to open through the day. Against the benchmark staffing policy, it improved passengers' time through the checkpoint by 31.4%.
Timetable-led release, routes to the platform, screening and ticketing arrangements.
What we model
Arrival waves, processing times, connections and the space people wait in.
What you get
Queues and clearance before each departure, utilization by hour, and where the next bottleneck appears.
Recorded · a wave of riders off a train crossing the square past the doorsMuseums, exhibitions and pop-ups
Keep every room comfortable, all day.
100visitors let in every 30 minutes, in simulation
A world-famous art gallery in Rome
Visitors carried Bluetooth beacons so researchers could learn how people really move from room to room. The simulation built from that found an entry rule, 100 people every 30 minutes with no two-hour limit, that removed the crowding peaks and kept every room steady through the day.
Ticketing and entry rules, room order, where the most popular pieces go.
What we model
Room-to-room movement, dwell time, attraction to exhibits, group sizes.
What you get
Visitors per day, crowding in each room over time, and the queue at the door.
Recorded · the café line and pickup, colour shows what each customer is doingRestaurants and retail
Find the change that shortens the lunch rush.
69 sin store, down from 201 s
A coffee-and-doughnut chain's campus store
A simulation of the lunch peak compared ten set-ups, today's included. A third register cut the time drink-and-doughnut customers spent in the store from 201 to 69 seconds. Moving all food prep to the back counter and adding one dual-purpose server cut it by about 55% for every customer. Moving a cashier to that counter instead pushed time in store to about 18 minutes.
Registers, prep stations, staff roles and priorities between channels.
What we model
Order mix by hour, task times, shared equipment, pickup and seating.
What you get
Time in store by customer type, utilization by role, and what each option asks of the team.
Recorded · the café terrace, colour shows what each guest is doingVisitor attractions and hospitality
Size the experience for the crowd that actually comes.
20shuttle buses instead of 24, in simulation
A port authority's cruise terminals · 15,000 to 28,000 passengers a day
A simulation of passengers, shuttle buses and taxis sized a new transport centre for the terminals. It needed 20 buses, not the 24 a static capacity study called for, with 94.5% of passengers still away within 30 minutes. Directing groups to pick-up points instead of assigning them at random cut the share waiting over 30 minutes from 21% to 12%.
Crowd build-upWill it show where crowds really build up in our station or venue, and when?
Yes. We model the space and the people moving through it, so you see where density builds and at what minute, not just a daily total. A study starts with the one decision in front of you.
Gates, stairs, countersHow do you model our gates, stairs, escalators and counters, and the way people arrive?
From your floor plans, gate and counter positions, opening hours, and your timetable or event schedule. You watch the model run, point at what's wrong, and every assumption is listed for you to change.
Real behaviorDo people in the model behave like real visitors, with slow walkers and people who stop on the way?
Yes. Each person has somewhere to go, walks at their own pace, and decides on the way whether to stop, for a coffee or a ticket, and which queue to join. That's why crowding at a stair foot or a ticket hall builds the way it does on the day. You can follow any one person and see where they waited and why.
After the World CupAfter the World Cup, would a second exit route or a staggered departure change our worst crowding?
That's exactly the kind of change we test. Add a route, stagger the exit, or close a gate, and watch where density builds minute by minute, with the same crowd each time. You see which change actually takes the peak off, before you build or staff for it.
Transit arrivalsMost of our crowd will arrive by train and shuttle. Can we plan around the timetable?
Yes. Arrivals come in waves from the timetable and shuttle loops, so the model shows what each train does to platforms, stairs and security lines, and how long the queue takes to clear before the next one arrives.
Record daysOur checkpoint lines spike on record days. Can we test lanes and staffing before the holidays?
Run your busiest forecast day with the lanes and officers you plan to open, and see the wait at each hour. Then move a lane or a shift and compare, on the same passengers, before the day arrives.
Plans and countsWe have a floor plan and gate counts. Is that enough to start?
Yes, that's a good start. Gate counts give the arrival pattern and the floor plan gives the space. Service and dwell times can start from observation, and the model shows which of them actually change the answer.
Safety · worst dayCan we test the worst day: a closed exit, a delayed train, an evacuation?
Yes. Close an exit, delay a train or start an evacuation in the model, and watch where density builds and how long the space takes to clear. It supports your safety planning, and your safety engineer still signs off.
Before opening dayWhat can we test before opening day or a refit, and what do we get?
One scoped question: a new layout, counter staffing, an event-day plan or a refit. You get the options side by side on the same crowd, with waits, density and dwell for each, and a model your team keeps.
Recorded · the arm sorts each box to the robot or the beltRobot cells and cobots
Know what the robot adds before you buy it.
+31.2%throughput in simulation, one worker instead of three
A ready-meal factory · small manufacturer
A simulation of a manual tray-loading line found its bottlenecks and tested an articulated robot doing the loading. Throughput rose 31.2% in large-scale production and 12.0% in order-based production, and one worker could run the cell. The installed robot's output closely matched the simulation.
Manual work against a robot, a robot or a cobot, where it stands, how parts reach it.
What we model
Each task time, the robot's reach and grip, the people who feed and clear the cell.
What you get
Output per hour, use of the robot and of each worker, and the conditions under which the robot pays for its place.
Recorded · the loaded robot drives its box to the dockMobile robots and AGVs
Buy the vehicles the line needs, and no more.
+51%hourly throughput at best, in simulation, from better coordination of the vehicles
A battery-pack assembly line · people, robot arms and vehicles
A simulation of a high-mix assembly line with manual, automated and semi-automated stations compared ways for AGVs and mobile robots to share transport and material supply. The best concept raised hourly throughput by up to 51% against the benchmark, and pooling the two fleets raised throughput while cutting investment.
Fleet size, which vehicle serves which station, charging, traffic rules.
What we model
Routes and crossings, stations and what each one consumes, the people and robots at each station.
What you get
Production delay, vehicle use, and where vehicles queue or block each other.
Recorded · cartons off the end of the line, stacked onto a pallet; the hall is a backdropConveyors and end of line
Find what really caps the line, from the packer to the truck.
+8%pallets a day in simulation, about 30 more on about 250
A can-packing plant · four lines, two dock doors
A simulation of the plant showed the two dock doors, not the packing lines, capped output at about 250 pallets a day. A connection to the warehouse next door added about 30 a day; without it, more than five forklifts added nothing.
A new palletizer, a bigger buffer, a conveyor change, more dock capacity.
What we model
Packing lines, conveyor merges and sorters, palletizers, forklifts or mobile robots, dock doors.
What you get
Pallets per shift, where cases wait, and which option moves the limit.
Illustration · people at rework and mixed orders beside the robotsPeople, skills and shifts
Put your people where the line needs them.
−40.4%total production time on average, in simulation, from smarter task allocation
A precision machining workshop
A simulation compared one operator per machine, everyone trained on everything, and sending each task to the nearest free qualified worker. The last cut total production time by 40.4% on average, with walking distance close to one operator per machine.
Who does which task, cross-training, shift patterns, people beside robots.
What we model
Task times and skills, walking distance, breaks and shifts.
What you get
Output, each person's workload, walking and waiting, and the staffing the plan needs.
Illustration · a sorter sending each product to its laneMix, sequence and changeovers
Run more products on the line you already have.
−67%lead time in simulation, with batches more than halved
A carmaker's factory · machining, sub-assembly and final assembly
One group of lines needed 40 minutes to change from one variant to the next, which forced large batches. A model of the whole factory flow set batch sizes and safety stock together. Batches came down by more than half, lead time by 67% and storage by 31%, with delivery precision kept.
Robot cellsWe're adding a robot cell to a line that's already running. Will it show what that does to the rest of the line?
Yes. We model the cell together with the stations and buffers around it, so you see what the robot does to blocking, WIP and output downstream, not just the cell's own cycle time.
Routings and buffersHow do you capture our routings, cycle times and buffers without stopping production?
From your routings, cycle times, buffer sizes and shift patterns, plus a walk of the line with your engineers. Nothing on the floor stops. You watch the model run, point at what's wrong, and every assumption is listed for you to change.
Blocking and starvingDoes it show stations blocked or starved between each other, or just average cycle times?
Every station, buffer, conveyor and robot move is modeled, so you see exactly when a station is starved for parts or blocked by a full buffer. Breakdowns and repairs come from your own records and run many times, so you see a range, not one lucky run.
Reshoring with too few peopleWe're bringing production back to the US, but we can't hire enough people. Can we plan the line around the crew we can actually get?
Yes. The crew is an input like cycle times: you set who works each shift and what they can do. The model shows the output that crew really delivers, and whether a robot, a cobot or a different layout closes the gap. You can try running a shift short before it happens.
Tariffs and trade uncertaintyTariffs keep changing our suppliers and lead times. Can we see how much buffer and flexibility the line really needs?
Run the same line with longer supplier lead times, a different product mix or less stock, and see which plan still holds output and which one starves the line. The costs come from the rates you give us, so you can rerun it when the tariffs change again.
AI and digital twinsEveryone is selling an AI digital twin. What's actually different here?
You can open the model and see every station, robot and rule, and every run is kept, so you can trace why a result came out the way it did. AI helps frame the question and build the model faster. The answer comes from simulating your line, not from a guess.
MES dataOur MES data is patchy and cycle times vary by shift. Can we still start?
Yes. Start with what your MES and engineers have, shift-to-shift variation included, because that variation belongs in the model. It then shows which cycle times actually move throughput, so you know which few are worth measuring properly.
Throughput matchHow do we know the model's output matches what we ship today?
We run today's line first and check its output, WIP and waiting against what you actually ship. Only when it matches what you see on the floor do we test changes. Every run is kept, so anyone can trace a result.
Automation capexWhat's a sensible first study before we sign the capex for automation?
One scoped decision: a robot cell, an extra AGV, or rebalancing a line. You get the options side by side on your own orders, the business case for each, and a model your team keeps using. Robot safety and programming stay outside the study.
Recorded · totes from two infeeds merging and passing the scan arch; the hall is a backdropConveyors and sortation
Find the merge, the sorter or the chute that caps the building.
5–6%better cart use at the chutes, confirmed at 12 pilot sites
A large online retailer's fulfillment centers
A physics model of the chutes after the cross-belt sorters tested chute designs, sensor placement and merge logic before launch. The model showed a 6% improvement in cart use at the chute exit, and pilots at 12 sites confirmed a consistent 5–6%. The result set network-wide guidelines for where the sensors go.
95%singulation accuracy at 7,000 packages an hour, in emulation, in 64% less floor space
A large online retailer's package handling line
Engineers built and tuned the controls of a new small-footprint package singulator against an emulation of the line, before any physical machine existed. At 7,000 packages an hour it reached 95% accuracy, a 45% improvement over conventional singulators, in 64% less space and with no gapper.
Merge rules, sorter lanes and recirculation, accumulation space, packing stations.
What we model
Conveyor merges and diverts, the sorter, packing, and the lanes to each door.
What you get
Cartons per hour, where cartons wait over the shift, and which change moves the limit.
Recorded · the crane stores one tote and fetches anotherRacks and automated storage
Get more out of the racks before you lease another building.
+67.89%picking throughput in simulation, from re-slotting the racks
An equipment manufacturer's rack warehouse
Management had approved an extra off-site warehouse, months from opening, to add picking capacity. A simulation tested re-slotting the existing racks into zones by how often each item is picked. Picking throughput rose 67.89%, from 853 to 1,433 picks per run, and the share of time lift trucks spent waiting fell from 23.6% to 19.5%.
−10%picking distance against the best slotting rule in use, in simulation
An electronics maker's component warehouse
A simulation run on 20 working days of real production orders tested a new way to assign thousands of component types to storage locations. It cut picking distance by 32%, 20% and 10% against three slotting rules in use, the last being the best of them.
Where each item sits, racking against automated storage, how many aisles an order touches.
What we model
Rack and tote locations, forklifts and shuttles, decanting, and the time to pick each line.
What you get
Picks per hour, travel and waiting, storage space used, and when the building runs out of room.
Recorded · robot carts bring totes to the pickers; the hall is a backdropMobile robots and goods-to-person
Buy the robots the floor needs, and no more.
2:1robots per picker, where efficiency peaked
A study loosely based on a food-service distribution center
Robots met pickers at each pick location, and the simulation ran 48 mixes of 3 to 5 pickers and 3 to 12 robots. Efficiency peaked at about two robots per picker. Beyond that, pickers were more than 95% busy while robots began to sit idle.
Fleet size, pick stations staffed, charging rules and traffic rules.
What we model
Robot travel and charging, queues at each station, and the pickers who serve them.
What you get
Orders picked per hour, how long robots wait at stations, picker workload, and whether robots, chargers or people run out first.
Recorded · a worker takes cases off the rack and builds a pallet; the hall is a backdropPeople, skills and shifts
Staff each shift for the work that actually arrives.
19–39%faster container processing in simulation, with the same crew
A carmaker's plant warehouse · container unloading and replenishment
A simulation built on the plant's own processing times re-formed the teams that unload sea containers and trailers and replenish stock, and how jobs are assigned to them. Across nine workload scenarios, processing all containers took 19% to 39% less time than with today's teams.
5.25fewer man-hours a day in simulation, with every store shipment still on time
A convenience-store chain's cross-docking center
A simulation set how many operators of each skill to have in every hour, so that each store's shipment leaves by its fixed time. The plan needed 5.25 fewer man-hours than today's, and daily personnel cost fell from 40,892 to 34,880 yen.
Shift patterns, people per area, cross-training, and people beside robots.
What we model
Arrivals by hour, task times and skills, walking, breaks and shifts.
What you get
Hours needed per area, backlog by hour, each person's workload, and the staffing the plan needs.
Recorded · cartons sorted to four dock lanes; the hall is a backdropDocks, waves and cutoff
Make every truck's cutoff without building more yard.
No-goon a yard expansion: the dock doors, not the parking, were the limit
A large online retailer's fulfillment center truck yard
A simulation of the truck yard tested adding parking spots. Even with more parking the yard still gridlocked, because the bottleneck was the dock doors. The analysis ended in a no-go on expanding the yard, avoiding the construction cost.
+20%average pick rate in simulation, from how orders are released
A retail warehouse · pickers working with mobile robots
A simulation on the warehouse's real order data tested how orders are released and batched into waves for the robots. Combining zone-based release with a saving rule raised the average pick rate by about 20% against current practice.
Manual plus automationWill it work for our mix of manual picking, conveyors and robots?
Yes. Pickers, conveyors, sorters and robots are built from the same pieces, so a mixed operation is modeled as it really runs, including the handoffs between people and machines.
Slotting, waves, docksHow do you model our slotting, pick paths, waves and dock doors?
From your layout, slotting, order history, wave plan, dock schedule and cutoff times. You watch a day run, point at what's wrong, and every assumption is listed for you to change.
Robot trafficDoes it model robot traffic, charging and pickers waiting at handoff, or just units per hour?
Every robot trip, charge and handoff is modeled, so you see where robots queue in aisles and where pickers stand waiting for a tote. That's usually where an extra robot stops adding orders shipped. You see it on your own layout and order profile, against your cutoff.
AMRs going mainstreamEveryone is piloting AMRs. How many would we actually need, and where would they stop helping?
We run your orders and layout with different fleet sizes and show orders shipped against fleet size. The point where another robot adds little is usually a picker or handoff queue, and the model shows you which one.
Humanoid robotsShould we be looking at humanoid robots?
We can't tell you how good a particular robot is. We can model a robot doing a task at the speed and reliability its vendor quotes, alongside your people, and show whether it moves orders shipped. It's a cheap way to test the pilot before you run one.
Peak laborWe can't count on hiring for peak. What happens to our cutoff if we're short-staffed?
Run your peak week with the crew you expect to have, and see which waves miss cutoff and where work piles up. Then compare the fixes: re-slotting, a second wave, overtime or automation.
WMS historyWe have WMS order history and a layout drawing. What else do you need?
Task times for picking, packing and replenishment, your shift patterns and cutoff times, and any robot specs. A few weeks of orders is enough to start, and the model shows which numbers move the answer.
Pick rates and cutoffWill the model hit our real pick rates and cutoff misses before we test changes?
Yes, that's the first step. We run a recent week as it happened and check pick rates, backlog and cutoff misses against your own records. Only when it matches your floor do we test changes. Every run is kept.
Before buying robotsWhat should we test before buying robots or adding a shift?
One scoped decision: a robot fleet size, a second shift, re-slotting or a new wave plan. You get the options side by side on your own orders, the business case for each, and a model your team keeps.
