1
The rich text element allows you to create and format headings, paragraphs, blockquotes, images, and video all in one place instead of having to add and format them individually. Just double-click and easily create content.
A rich text element can be used with static or dynamic content. For static content, just drop it into any page and begin editing. For dynamic content, add a rich text field to any collection and then connect a rich text element to that field in the settings panel. Voila!
Headings, paragraphs, blockquotes, figures, images, and figure captions can all be styled after a class is added to the rich text element using the "When inside of" nested selector system.
Every other post on LinkedIn talks about how machine learning and AI can “10x your organization’s output” without going into any real specifics.
So, let’s get into specifics - here’s how we worked with Amerit Fleet Solutions to create a ML solution that will probably return their investment over 10x.
Amerit is one of the country’s largest trucking fleet repair services. Every day, they do a few thousand maintenance jobs on delivery trucks from Amazon, AT&T, and more.
That means lots of mechanics, and lots of repair jobs. 15,000 separate repair codes, to be exact!
While they’re an efficient operation, they noticed one specific bottleneck: errors. Sometimes, their mechanics make honest mistakes. Incorrect billings, wrong part numbers, false readings, and forgotten repairs. They have a team of a few dozen quality control people who manually review each work order (a few thousand a day!) to ensure they’re correct.
Plus, incorrect repair orders can be a costly mistake. Investigation can cost many multiples of the price of the original repair, as well as the time of the team. An incorrect order means complaining customers, means talking to customer support, talking to procurement, going to billing, back to procurement. An $80 repair bill can easily become $600 worth of extra work. Not to mention losing their reputation. Flagging errors that mechanics make can make investigation quicker, more efficient, and ultimately save money.
The only problem? They needed a team to help execute.
NineTwoThree was selected to help work together with Amerit to make the automated system a reality. With our experience in building cutting-edge machine learning solutions, we were selected to cut through the mountains of data and find the insights. We combined our extensive data science expertise with Amerit’s incredible talent and data to find a solution.
NineTwoThree’s approach to solving data problems was critical for this project. Our team of data scientists worked closely with the customer, without the need for an intermediary PM. This allowed us to translate data science concepts into business value for the customer.
Once we started working with Amerit, we identified several key challenges standing in the way of our goal. Chief amongst them was the data problem.
A little aside about data: it’s important to not just “jump in” to using some expensive generative AI solution to solve a problem. Especially given how expensive these projects can be. You might not even NEED GenAI - which is exactly what we found out by following our proven process.
Understanding the problem, and what approaches might work, is the first step. So, we first set out to understand the problem, and only after we knew it was a data science problem, what type of data science solution we could develop. Then, once we had an understanding, we could determine how the project could move forward.
After analyzing the data (the 300k+ mechanic’s comments), we were able to develop a POC model that showed the data retained a promising level of predictive power. We were confident that future efforts would work how we wanted the system to work initially. Importantly, our POC worked WITHOUT spending money on modern infrastructure. We validated first.
Then, we started investing in setting up a modern data pipeline.
We’ve talked extensively about how far data quality goes in delivering a successful data science project, and this was no different.
The current process was broken for a few reasons:
To tackle these challenges, we continued to refine the model using a state-of-the-art machine learning concept called gradient boosting (CatBoost). Note this is NOT using generative AI, GPT, or any other fancy (potentially expensive) solution. Some prediction use cases are better suited for machine learning techniques.
The system we developed was astonishingly effective in comparison.
We developed a modern data pipeline to extract the data from the SQL database, clean the data, and load the data into our model.
To combat the lack of version history, we started by taking multiple daily snapshots to mimic a history (as the repair orders would be updated during the day.) This, combined with the answers from the quality control team, allowed us to create an accurate set of labeled training data.
Our goal with the model was to predict two things:
Our phase 2 model predicted that there would be no error with 66% accuracy, and that there would be error with 72% accuracy.
This was completely transformative.
Building AI and ML solutions for real-world problems means you’re combining UX and data insights. For Amerit, this was critical. Telling a mechanic “this might be wrong, please double check” would significantly reduce the changes they’d actually check.
Thanks to our model, we were able to tell them “This is wrong, please fix” with a high degree of confidence. This meant our data insights wouldn’t fall on deaf ears.
This is where the real value came in. Not only were we able to deliver on our promise, we were able to unlock a new capability for the team - prioritization. Thanks to our model, we were able to weigh the expected cost of a repair order vs the impact of addressing it. Instead of the previous first-in-first-out method, they could focus on higher priority orders.
And that’s not all. That same cost-benefit structure allowed us to automatically categorize 1,000+ repair orders as “unneeded for manual analysis.” They had a high likelihood of zero errors, and the amount of actual missed errors (6 in total) would cost less than investigating all of them.
Further feature enhancements (noticing a price discrepancy between the part listed and the charge for the repair order) and natural language explanations for flagging (e.g. “This is why we flagged this repair order”) will further improve the system.
Thanks to our ML model, we estimate Amerit will be able to save at least six-figures annually, and dozens of hours of labor each day.
With the right ML solution, these results aren’t outliers.
Not only that, but their recurring commitment is measured in tens of thousands of dollars, instead of paying a team of data scientists 7-figures to maintain the system.
This was only possible through close communication with the domain experts. Finding the right business fit for the data was our number one priority, and combining our expertise with Amerit’s extensive domain knowledge was a winning formula.
Ask yourself these questions when thinking about AI, and you’ll get a better idea of what the ROI could be.
1. Would replacing this process with AI save you time? some text
2. Would replacing this process with AI save you money?some text
3. Would replacing this process with AI reduce mistakes?
4. Would replacing this process with AI lower the chances of reputational damage and risk?
5. Would replacing this process with AI make your product better, measurably via metrics like engagement?
6. Would replacing this process with AI make your employees work faster, during processes like onboarding?some text
7. Would replacing this process with AI let your employees spend their time focusing on higher-value activities?
If you answered “yes” to more than one of these, the ROI could be worth it.
Amerit was kind enough to leave a killer review for us on Clutch. Judge for yourself!
Here are some highlights:
“What I found to be unique about 923 was their willingness to work with us while we built our historical data. It started with manual data queries and uploading to a shared file before automating this process. Even during the early days of the project they were able to deliver fantastic results and learn more about our business. Because of the process we implemented with 923 we have set ourselves for future work in the AI space.”
“The 923 team was fantastic to work with and taught me a lot about ML and AI. The team was patient while we worked through solutions to help them and always delivered on time. This was our first step into AI/ML and we have learned a lot to set us up in the future. Our biggest issue that 923 was able to overcome was data. We did not have the historical records to be able to supply for the model learning. Even with a small data set 923 was able to deliver great results. Rather than just taking our requirements and building something they took the time to learn our internal processes and make suggestions on how we could implement the solutions they've developed. “
“The 923 team always met timelines for deliverables and quickly responded to my teams needs. Because they took the time to get to know our business and learn our process they were not only able to adjust to changes in scope but were also able to recommend ways to implement solutions.”
“One of my favorite things about them was their ability to share the results of their work in a clear way that could be understood even if you are unfamiliar with AI/ML. This was my team's first time working with AI/ML and they were able to eliminate any confusion we had on the process/results which can be very complex.”
With the right team, this doesn’t just have to be a dream. Reach out to us if you want to make data insights a reality.
