09/15/2026

How Predictive Analytics Cycle Counting Works

How GEODIS built its predictive analytics cycle counting model, and the root cause analysis and workforce benefits it delivers for clients.

Predictive analytics cycle counting provides an alternative option for GEODIS clients who want to maintain and improve inventory accuracy. PACC uses a mathematical model to predict cycle count locations that have a high probability of mismatch in a warehouse, based on historical cycle counting data and other contributing factors. The benefits of PACC include enhanced root cause analysis and remediation to prevent inventory errors, redeployment of labor to revenue-generating activities, and a more engaged workforce that benefits the client.

Key takeaways

PACC uses a mathematical model to predict cycle count locations that have a high probability of mismatch in a warehouse, based on historical cycle counting data and other contributing factors.
The benefits of PACC include enhanced root cause analysis and remediation to prevent inventory errors, redeployment of labor to revenue-generating activities, and a more engaged workforce that benefits the client.
The model is continually self-correcting: it takes in the results of each cycle count and matches this against its predictions, adjusting its analysis and algorithms to increase predictive accuracy over time.

Part 2 of 3 in a series on GEODIS predictive analytics cycle counting (PACC) approach. Part 1 covers what predictive analytics cycle counting is and why GEODIS built it. Part 3 covers the pilot and live-client results, and how you can use PACC.

Predictive analytics cycle counting replaces standard cycle count approaches

Predictive analytics cycle counting provides an alternative option for GEODIS clients who want to maintain and improve inventory accuracy. PACC replaces our standard inventory approach by counting a representative sample of inventory items that are more likely to be inaccurate. It allows for more rapid counting and reconciliation of stock levels and warehouse locations, particularly across high-volatility products that have had discrepancies in the past.

The PACC approach gathers historic data for each SKU and location type, including previous inventory counts, variances, and product groups. The model then uses several factors to predict the areas where an inventory cycle count would be most valuable, mainly with a focus on areas that have a high probability of cycle count errors, so that warehouse staff can identify, investigate, and resolve discrepancies.

The specific criteria for how the model decides on what gets counted can be partly defined and configured by the client. For example, one PACC approach may look at high-velocity items, while another may focus on high-value products.

The advantages of a PACC approach include:

More focused counts based on likely discrepancy rates.
Earlier identification of errors, allowing for better RCA.
Less workforce labor required for cycle counting, allowing redeployment to higher-value activities.
More frequent inventory reconciliation.
Reduced frustration from employees upon encountering an inventory variance during picking.
Reduced delays for employees waiting for inventory issues to be resolved.

Discover how GEODIS can help you manage your inventory more precisely. Get in touch with GEODIS.

Button Contact us

woman_clipboard_checking_inventory_warehouse_2

Removing standard cycle counting drawbacks through a predictive analytics approach

PACC resolves standard cycle counting issues by avoiding stagnant locations and identifying the inventory items and warehouse areas that are most likely to have inaccuracies or discrepancies. These areas are then counted more frequently, allowing for earlier identification of inventory issues, faster and deeper RCA, and more effective remediation.

The PACC model uses limited resources more effectively, resulting in more accurate inventories over time, less workforce effort, and better client results. This supports longer-term, continual improvement, driving benefits for both the client and GEODIS.

One of the main features and benefits of PACC is how it enhances RCA of inventory inaccuracies.

Developing the predictive analytics mathematical model

The mathematical model that GEODIS has created is designed around several key principles:

Uses multiple information inputs including product, location, and inventory data.
Trains and refines itself based on historical, real-world data.
Captures and identifies important correlations that guide accurate analysis and algorithms.
Predicts areas that should be counted based on the likelihood of errors.
Outputs prioritized cycle count areas for counting and reconciliation.
Self-corrects, based on the latest data.

We carried out the following steps to build the model for our predictive analytics cycle count approach:

Establish a representative inventory dataset for a pilot client.
Gather different types of inventory data from the dataset.
Carry out exploratory data analysis.
Find correlations and matches between data types and inaccurate counts.

Establish a dataset

We used the dataset from a popular direct-to-consumer ecommerce retail brand.

Data types used in data exploration

The GEODIS project captured the following data points from the inventory dataset. The model includes these attributes, categories, and calculations for training and cycle count prioritization.

Item attributes used to train the PACC model

Product name and description.
Product SKU and unique identifiers.
Product group and status.
Product color, size, and category.
Product value.
Product velocity.

These attributes are modified further in particularly high-risk product groups.

Location attributes used to train the PACC model

Location and storage types.
Aisle, picking, and putaway zones.
Items in surrounding locations.
Item distances, dimensions, and positions.

GEODIS primarily uses two storage locations, floor and rack, with rack being the more popular. Most cycle counts occur in rack locations. Storage locations, rather than pick faces, are where most replenishment activity occurs. This means storage locations are cycle counted more frequently than pick faces.

Some locations are excluded from the data based on feedback from inventory control teams. This is mainly due to buffer zones and other locations that cause "noise" in the data.

Inventory attributes used to train the PACC model

Days since the last cycle count.
The number of previous cycle counts.
The number of pick and replenishment events.
The quantity of product shipped.
The total inventory for the product.
Shipping quantities and rates.
Total and available quantities.

Findings from exploratory data analysis

Our analysis identified several interesting insights:

Most cycle counts passed and there was no inventory mismatch.
Approximately 1% of all cycle counts failed and caused a mismatch.
Item and location attributes showed promising correlations with cycle count failures.
Certain location features, such as location type and aisle, indicated that the largest number of cycle count failures occurred in specific areas of the warehouse.
Traditionally, picking and replenishment activity at a location is thought to contribute to a greater number of failed cycle counts. However, the data does not show a strong correlation between picking and failed cycle counts.
This could be because locations with high activity are heavily cycle counted and inventory levels are closely monitored.
Personal intuition about what types of inventory may experience cycle count failure is not always supported by data.

Contributing factors to failed cycle counts

The data uncovered some fascinating insights about cycle count failures:

Storage locations have a higher probability of cycle count mismatches, compared to pick faces.
There is a correlation between specific product groups and higher levels of cycle count mismatches.
Specific aisles in the warehouse correlated with higher numbers of failed cycle counts.
Product groups with the largest ship quantities over the previous ten days had a slight increase in failed cycle counts.

Scan d'un vêtement pour la préparation de commande

Training the predictive analytics model

GEODIS data scientists analyzed the data that influences cycle count accuracy and built a neural network and algorithms that use the information to test and predict likely cycle count variances. These outputs were then tested against actual inaccuracies, and the results were fed back into the mathematical model. This process was repeated several times, helping to further refine and correct cycle count predictions.

This resulted in a mathematical model that can predict problematic cycle count locations with a high degree of accuracy. The model produces a weekly snapshot of inventory data. The weekly snapshot reviews all of the inventory for a particular client at a specific moment in time.

The model then scores every part of the inventory captured in the snapshot with the likelihood of that location and product being inaccurate. The model uses this scoring to predict, prioritize, and rank locations and products for cycle counting in the following week. These recommendations are passed to the IC team and form the basis of their cycle count activities.

The model is continually self-correcting. It takes in the results of each cycle count and matches this against its predictions, adjusting its analysis and algorithms to increase predictive accuracy over time.

Future PACC development

One PACC area that we are continuing to develop is how we specifically define the items that we want to count. We can configure PACC item and location count recommendations using many different factors, and understanding the right combination is central to using our inventory count resources in the most effective way.

For example, a strong starting point for scoring the weekly snapshot might be combining:

Item velocity: how frequently that item is ordered, picked, and packed.
Item value: the retail value of that particular item.
Item location: how frequently we discover error discrepancies in a location type.
Historic discrepancies: how often the inventory count has been incorrect for that item in the past.

This might then give us a list of 100 items that we count daily and a further 600 that we count weekly. We are continually testing the PACC model to balance SKUs, locations, and count frequencies so we can optimize our workforce efforts.

We're developing the model to move beyond the most important areas of the weekly snapshot so that we can get the best return for time invested on other parts of a client's inventory. This is happening as we mature the PACC model further, bring in more inputs, train the algorithm, and go through a continual feedback and review process.The benefits of adopting a predictive analytics approach

Resolving inventory inaccuracies through root cause analysis

The variety and volume of order fulfillment and inventory picking creates an environment where errors can naturally occur. Examples include:

Products stored in incorrect locations.
Incorrect items picked and packed.
Too many or too few items picked.
Mistaking an outer box (that may contain several items) for an inner box (that only contains one).

GEODIS maintains very high standards for inventory picking and has extremely low shrinkage rates, but these rates can be reduced even further. The main method we use to achieve this is root cause analysis.

Find out how GEODIS can optimize your inventory management. Get in touch with GEODIS.

Button Contact us

The importance of root cause analysis

RCA is a process that identifies why picking discrepancies happen. When we identify an issue with inventory discrepancies, we investigate the factors that caused that error: this tells us the root cause.

When we understand the root cause, we can take steps to remediate the problem so it does not happen again. These remediating actions can take many forms: changes to standard operating procedures, updating signage, staff training, and other measures. We capture these root causes using "reason codes" that allow for further analysis and improvement.

For example, RCA may tell us that an employee has accidentally mistaken an outer box that contains multiple inner boxes, for an inner box. They have then picked and packed that outer box thinking it is only one item. Once RCA has identified this mistake, we would update our procedures for that item to ensure that all outer boxes are opened and disposed of when storing the product so that only inner boxes can be picked. This would resolve the underlying problem and ensure it did not recur.

Effective RCA continually improves inventory management, resulting in lower shrinkage rates and more accurate picking.

How PACC improves root cause analysis

PACC makes RCA faster and more effective. Because PACC focuses on products and locations that have had issues in the past, without improvement, it is more likely those areas will have issues in future. This affects RCA in two main ways:

More frequent counting of "repeat offenders" results in earlier identification of problems, making RCA easier.
Limited investigative resources can be prioritized to the areas with the most likely discrepancies.

We can then identify and address negative trends to do with training, process improvements, the customer relationship, and other areas. This results in faster and wider-ranging problem remediation, leading to lower shrinkage rates. We can reassign cycle-count labor to RCA, moving from simply checking locations to preventing those issues from happening in the first place.

Correcting inaccurate inventory levels and locations sooner improves the chances that an inventory variance won't impact a teammate during order processing and picking, ensuring that SLAs are met.

Redirecting inventory control labor to value-added client initiatives

Predictive analytics cycle counting allows GEODIS to redirect our inventory control labor to check locations that are more likely to have discrepancies. This means we spend less effort on areas that are already accurate, and can refocus that attention on problematic locations.

Even with this redirection of IC labor, PACC also drives an overall reduction in IC workforce resources. This allows us to redeploy IC labor to drive improvements in other aspects of the client's business.

warehouse_people_logistics_checking_inventory

Greater workforce engagement and incentives as a result of PACC

PACC has direct benefits for our warehouse and logistics workforce, and, by extension, our clients.

It's helpful to understand how GEODIS assigns our employees between "indirect" and "direct" functions. IC employees are part of our "indirect" workforce: these are the functions that, although important, do not directly contribute to client revenue. Examples of indirect functions also include customer service and facilities management.

The other part of the GEODIS workforce is employees in a "direct" function. These are activities that directly generate revenue for the client and include receiving, picking, packing, and shipping products. Direct functions are financially incentivized by GEODIS to perform well: the faster and more efficiently that staff can fulfill orders, the greater their wage bonuses.

Prior to PACC, our indirect IC staff would spend the vast majority of their time simply counting items, so they could not benefit from direct activities and higher bonuses.

PACC changes this balance. More efficient counting results in these staff spending less time on indirect inventory counting, and more time on direct, revenue-generating functions. This means they have similar opportunities to increase their paychecks, resulting in greater staff engagement. We have seen this demonstrated in our pilot project metrics: the ratio of indirect to direct functions decreases significantly after introducing PACC.

PACC also enhances employee engagement in several other ways:

RCA activities often involve staff training, and GEODIS provides a bonus for training other staff. More RCA-related training activities results in greater bonuses for those employees.
GEODIS encourages promoting from within. PACC allows employees gradually to transition out of traditional cycle counting, which can be a repetitive, low-interest task, and into other roles that add more value for the employee, the client, and GEODIS.

Taken together, PACC-related workforce initiatives allow for stronger career development, supported by financial incentives. This gives our employees wider exposure, such as for internal promotions and more challenging and fulfilling roles.

Part 3 covers what this looked like in practice, with results from both the PACC pilot and a live client in production.

Paul Maplesden

Former Lead Content Strategist

This is Part 2 of a three-part series. Part 1 covers what predictive analytics cycle counting is and why GEODIS built it. Part 3 covers the pilot and live-client results, and how you can use PACC.