This page is a tangible walkthrough of how we build new value from our customer's existing historical data, diving into one application of our insights as a key example of how we can build value. This is not our entire platform, all of its use cases or capabilities. Instead, it is a targeted example: how can we extract the top customer complaints and their corresponding resolutions from unstructured textual service orders?
The demonstration below is built from mocked up data, inspired by public webforums and created by Predii to resemble automotive service orders.
Information Extraction From Historical Servicing Data
Service Employee Augmentation With Prescriptive Insights
Workforce and Market Performance Executive Analytics
Prebuilt Models Built for Rapidly Processing Aftersales Data
This page has three sections:
At Predii, we believe that our results speak for themselves. This Results section includes three graphs.
First, a set of mocked-up unstructured data similar to the customer service data we work with every day. None of the IDs, names, complaints, or resolutions are real, but they are familiar to anyone who has worked in automotive service. Importantly, this includes the complaint that the customer communicated, and the technician's notes regarding resolving the issue. Example Predii Insights are included in the Predii Discovered Symptom and Predii Discovered Repair columns.
Second, the symptoms detected by Predii NLP within the "Customer Complaint" column of the raw data.
Third, the corresponding repairs detected by Predii NLP within the "Technician Notes" column.
A technician reviewing these graphs takes the customer's complaint and immediately knows the proper repair to resolve that issue. This data can also be embedded into tools and other enterprise workflows.
Predii processing involves a highly advanced suite of Natural Language Processing (NLP) componentry, build to understand the true meaning locked within unstructured text. This is far more than standard text analytics – these NLP features replicate "understanding" of the true intent of servicing data. They build the context of how words are used within a sentence, understanding the true meaning of the data recorded, allowing the enterprise to create a true-to-life intelligence platform to power their operations and products.
Components are the cornerstone of asset repair and maintenance. They also represent some of the most variegated data in this ecosystem – each enterprise has their own terminology, naming patterns and shorthand for the thousands of parts that make up mission-critical assets. Predii Component Discovery’s ability to correctly identify the component being addressed in the data unlocks the rest of servicing analytics. Included in Predii’s capabilities are longest matching string detection, negative list filtering, part of speech and phrase identification, synonym discovery, deep learning techniques and more.
“Found water pump leaking. Removed and replaced it, topped off coolant”
water pump, coolant
Every domain has various concepts that are particular to the use case. For example, in warranty fulfilment, fraud is a key concept. In dispatching and call center support, customer sentiment is a key indicator of performance. In this example, warranty and call center analytics might be examining the same set of phone calls – with completely different key indicators and desired outcomes.
Predii Concept Discovery flexibly adapts to the natural shape of the concepts particular to the use case. In our automotive repair and maintenance example, Predii’s customer required symptom discovery – a symptom usually being what a customer can hear, feel, see, or smell. This framework is further detailed by spatial, temporal, fluids and component modifiers.
Saw a leak in the fuel tank. When the car is low on fuel, there is no leak. But when the fuel tank is full, there is a leak.
fuel tank leaks from the top when full
Component: fuel tank , symptom: leaks, temporal: when full, spatial: from the top
What does “it” mean? Well, it depends. Where words are placed within a sentence, and within a larger collection of sentences and expressions of natural language, completely controls the meaning imparted. The Predii Platform will discover all relationships between various aspects of your data – but only certain relationships are relevant or useful.
Predii Proximity Detection identifies certain expressions of those relationships, and detects whether they occur within a significant enough proximity that it can be inferred to be an accurate finding. Proximity detection can be the key to the machine reading correctly identifying whether two or more concepts are complimentary, or distinct.
“Battery is 6 years old and shows signs of wear. Tried recharging it, but it will die soon again. Recommend replacement as soon as possible.”
Failure: battery wear.
Resolution: replace battery
Text: Frequently the patterns of natural text will include information that is less relevant or completely unrelated to the issue at hand. Predii Precedence Discovery enables the platform to correctly extract the key insights from the data, and reduce noise.
When it comes to automotive repair and maintenance, the data will also frequently have key indicators that are more telling to the true nature of the problem than more vague inputs. Predii is able to distinguish which concept takes precedence, and correctly identify the more important insight.
“check engine light is on and I hear a grinding sound from the engine”
Symptom: “grinding sound from engine” , “check engine light on”
Subject matter expertise allows knowledge workers to record their data in such a way that their implications and references add up together to communicate a coherent insight. For example, a report might be five sentences long, and the information included in the first sentence unlocks the meaning in all of the following.
Predii’s Implication Discovery is able to connect the dots and understand the implications present in the data – just as if a subject matter expert was reading the input themselves. In the example below, Predii is able to correctly identify which particular type of hose and clamp the technician noted as replaced.
No issue with battery. Hose clamp is broken. It was loosing coolant. Replaced hose and clamp, topped off coolant
Coolant hose, coolant hose clamp
The human mind can combine chains of events into one coherent concept. In order to replicate the interpretations an expert might make from data, Predii Hierarchal Discovery can infer relevant data from other sources to arrive at the true insight.
An easy example is data where the source involves multiple inputs and selections, like surveys that include a combination of drop down questions and free input boxes. The concepts expressed in one input change the meaning of the others, and thus must be disambiguated in order to achieve accuracy. In the example below, Predii has discovered the context (ENTERTAINMENT) that explains the second input (“my husband can’t get it to pair”), as well as synonyms for components and symptoms.
Column A: ENTERTAINMENT SYSTEM
Column B: “my husband can’t get it to connect”
“husband” = phone, Bluetooth. “pair” = connect.
Symptom = BLUETOOTH – NOT PAIRING