Given the large number of 3PLs in the U.S., with variations in both their size and in the services they offer for transportation and logistics, it's probably not a surprise that there is also a big variation in the analytics capability between 3PLs.
A large analytics gap exists between the 3PL ‘haves’ and the ‘have-nots’. Some organizations have invested heavily in everything… including data science teams, large-scale technical infrastructure, and expensive software tools. They are "all-in" on building everything in-house for themselves. But looking more broadly, Penske's recently published 2024 3PL Study reported technology issues in this area ranging from a lack of clear business cases (56% of 3PLs), to a lack of capital (46%), to a lack of adequate talent (43%). At the "have-not" end of the spectrum we see small budgets and a need to rely on what's provided by Excel, or one or two vendor systems their use (like a TMS). Staff is often expected to get reports out the door by hand, on a scheduled basis, as a side-activity.
4 Types of Analytics
Descriptive Analytics
Most 3PLs work hard to deliver descriptive analytics that report “what happened” or “what is happening”. The most common delivery method there is Excel reports, and many departments still build those reports manually, but at the end of the day historical analysis gets into the hands of customers in a quarterly fashion (at least). Timely delivery of these reports to customers, at a daily or weekly frequency, may still be an issue. If a 3PL has invested in additional analytics technology, they might offer customized Power BI or similar tools that offer users additional pivot & summarization capability, often by lane or by carrier.
Predictive Analytics
Fewer groups deliver predictive analytics that look forward to forecast & model what will happen in the near future. Forecasts can range from tracking volume trends, or carrier spend, to even “predicting disruptions”, but the common goal here is to let the data tell us what will most likely happen in the near future. A common predictive analytic (for transportation-management-focused 3PLs) is the prediction of late shipments and predicting pickup & delivery times. When a system does provide a predictive forecast, it is worth noting how those forecasts are created. The simplest forecasts use averages, moving averages or linear regression algorithms. A more advanced forecast applies algorithms like ARIMA to potentially detect and account for seasonality. AI algorithms have also proven useful for forecasting and can train using a large number of features from historical data. Also, some forecasts are derived using only historical data and a single source (e.g., historical shipping data), while other forecasts use multiple data sources and may include external data to improve specific forecasts (e.g., the Federal Reserve Global Supply Chain Pressure Index, lane market analysis from FreightWaves, DAT, etc.).
Diagnostic Analytics
An even smaller number of groups branch off into diagnostic analytics, where the goal is to identify “why” something happened, or to detect anomalies. Note that when teams deploy AI-based solutions, many of the classic AI algorithms are notoriously bad at explaining why and how they come to their decisions and forecasts. Classic correlation and other statistics techniques can be used for this type of analytics, but it is hard to find resources to do the analysis without error and also report the results in a business-friendly manner that staff and customers understand (and find useful).
Prescriptive Analytics
And finally, the least-seen analytics type is prescriptive analytics that offer decision support or decision automation. The most common scenario where this kind of analysis is found is likely to be built into the software tools used by a 3PL (like a TMS, WMS, or inventory system).
Anomaly Detection
It is also worth looking at 3PL capabilities specifically regarding anomaly detection. 3PLs commonly have systems that enable them to get basic alerting reports for common use cases, such as reporting on shipments that exceed specific dollar thresholds for carrier rates. However, these reports usually lack the intelligence to identify profoundly important and interesting insights for users. When these reports are auto-delivered, recipients often feel that they are being spammed with unnecessary information. Also, these reports can be highly manual to set up and maintain, and tedious to customize for a large number of slight differences (e.g., for many different thresholds of carrier rates varying by combinations of lane-mode-service-equipment). 3PLs need more advanced and automated systems to improve anomaly detection, where human setup is minimized and where the system auto-identifies what is "normal" (for differing sets of shipments) and auto-adjusts itself as averages vary over time.
LLMs
Large language models (LLMs) such as Chat-GPT are still a topic of great interest, although much of the discussion continues to be speculative. (See articles like 'Large Language Model Impacts On Supply Chain' and 'Using Large Language Models To Revolutionize Supply Chain'. )
A subset of LLM features and applications are related to analytics. While LLMs do offer the ability to summarize data, and that functionality seems to be advancing fast, there are still issues (such as widely-reported 'hallucination' effects) that discourage deployment for data analysis. That may change as LLM features evolve quickly with each subsequent release.
One feature that is slowly gaining traction, offered by some LLM-enabled software, is to allow users to ask data-related questions as typed text and then receive back a specific analysis on a web page (or in an app). This works similarly to how LLMs can be used to create simple programming code, or to write articles... the engine writes a (complicated) technical query on behalf of the user. Power BI, Tableau and other vendors are currently working to improve features in this area. However, the implied use case here is that: a) the user has both the time and an important-enough question to go to an interactive website / app; and b) the user's new, unique, ad-hoc question was not previously thought of by any BI and dashboard designers who create the ready-and-waiting reports. From this perspective, the feature seems most useful to 'data mining' staff that do data investigation, rather than the broader user audience. These use-case limitations arguably discourage a large expenditure for this feature/service, and it is not yet widely accepted as 'table-stakes' functionality that 3PLs must adopt.
Comentarios