# Predictive Maintenance to Revolutionize Fleet Management

Predictive Maintenance to Revolutionize Fleet Management - YouTube

Tap to unmute

[Predictive Maintenance to Revolutionize Fleet Management](https://www.youtube.com/watch?v=VRupuk72fYk) [Uptake](https://www.youtube.com/channel/UCn312Qjz_J_WxY6BHN1y-1g)

Uptake857 subscribers

The presentation focuses on AI and data strategies for fleet management, exploring why AI is revolutionizing fleets, key use cases (with a focus on predictive maintenance), and practical takeaways for implementing AI in fleet operations.

---

**Transcript of Adam McElhinney’s Talk – Uptake (uptake.com)**

_Topic: AI and Data Strategy for Fleets_

Hi everyone, thanks for coming to my talk—especially late on the last day of this great conference!

My name is **Adam McElhinney**, and I’m the CEO of a company called **Uptake**. Today, we’re going to cover a lot of ground about **fleets and fleet data**, with a particular focus on **AI strategy**.

Here’s a quick overview of what I’ll be discussing:

- Background on myself
- Why AI is transforming fleets _right now_
- Key use cases of AI in fleet management
- A checklist to evaluate AI use cases
- A deep dive into **predictive maintenance**
- Concrete takeaways for your organization
- Time for Q&A (hopefully!)

---

### A Little About Me

By day, I’m the CEO of Uptake, where we help customers **reduce unplanned downtime**.

By night, I serve on:

- The Advisory Board of the **University of Illinois Chicago College of Engineering**
- Multiple startup boards
- I also teach **Computer Science and Mathematics** at the **Illinois Institute of Technology**

I hold **19 patents** in machine learning and AI—mostly in industrial applications—so this is something I’m deeply passionate about.

---

### Why AI Is Revolutionizing Fleets

Let’s talk about what’s changed to make AI _practically useful_ in the past five years:

1. **Theoretical Breakthrough**  
   In 2017, Google released a groundbreaking paper: _“Attention Is All You Need”_. It introduced the **Transformer** model—the “T” in GPT (as in ChatGPT).

Most of the authors have since left Google to start AI companies, showing how commercial this space is becoming.

2. **Computing Power**  
   New neural networks scale almost linearly with computational resources. Unlike older models, you don’t hit diminishing returns as quickly. That’s why we’re seeing huge investments—up to **$500 billion**—in computing infrastructure.

3. **Data Availability**  
   Models are being trained on vast datasets like **Common Crawl** (a publicly available index of the web). Companies are now signing **exclusive data licensing deals**—Google and Reddit, for example—to feed proprietary AI models.

---

### Two Key Axes for Evaluating AI Use Cases

When thinking about where to apply AI in your fleet, consider these two dimensions:

1. **Impact** – _If fully automated, how much would this transform your fleet?_  
   (0 = no impact, 10 = game-changing)

2. **Viability** – _Can this realistically be automated using today’s tools?_  
   (0 = not feasible, 10 = plug-and-play)

---

### AI Use Cases in Fleet Management

| Use Case | Impact | Viability |
| --- | --- | --- |
| **Autonomous Driving** | 10 | 2 |
| **Predictive Maintenance** | 7–8 | 8–9 |
| **Routing Optimization** | 6–7 | 7 |
| **Customer Service AI** | 4 | 9 |
| **Driver Behavior Monitoring** | 6–7 | 6–7 |
| **Warranty Processing** | 5–6 | 6 |
| **Inventory Management** | 6 | 6 |

---

### The Predictive Maintenance Deep Dive

**Objective**: Detect and alert on potential failures _before_ they happen.

Current standard is **preventive maintenance (PMs)**—based on miles or hours.

Predictive maintenance takes it further by using **fine-grained data** from telematics, fault codes, and sensor readings.

---

#### Why Fault Codes Alone Fall Short:

- **Extremely noisy**: A single truck might generate **300–400 fault codes/month**
- **Hard to interpret**: Many techs just clear them without action
- **Not always predictive**: Same fault codes don’t always mean the same outcome

---

### Uptake’s Predictive Approach

We use:

1. **Work order data** to identify historical failures
2. **Telematics & fault codes** to match events leading to those failures
3. **Sensor data** (especially from **Geotab**, a partner of ours) for real-time insights

Think of it like Netflix:

> “If these combinations of fault codes and signals appear, that likely means X failure is coming.”

This leads to **longer lead times**, **fewer breakdowns**, and **higher fleet availability**.

---

### AI Use Case Checklist

Use this framework before launching any AI project:

1. **Can you quantify the current cost?**  
   If not, you can’t measure improvement.

2. **Is the business impact material?**  
   Don’t spend AI resources on $10K/year problems.

3. **Do you have lots of data?**  
   Rare events are tough for AI.

4. **Can you validate outputs?**  
   Always back-test with historical data before deploying.

5. **Can a human do it at small scale?**  
   If no one can do it manually, AI probably can’t either.

---

### Final Recommendations for Your Fleet’s Data Strategy

1. **Standardize Your Metrics**  
   Agree on how you measure things like _downtime_. Definitions should be consistent across departments.

2. **Accept Imperfect Data—but Improve It**  
   Everyone has dirty data. Use scorecards. Check if devices are on, properly assigned, and if technicians are entering codes meaningfully.

3. **Write a Data Strategy**  
   Align your data approach with your business goals. Data should be treated as a **horizontal asset**, not a siloed IT concern.

---

### Closing & Q&A

That wraps up my talk. I’d love to take any questions—whether it’s about AI, fleet data, predictive maintenance, or anything else.

If not, thank you so much for your time.