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.
.webp)
If the title of this post piqued your interest, you’re probably in one of two places.
You might be an organization months (or years) into an AI product, running into roadblocks. Maybe you’re jumping around models, trying to find the right fit for lacking data. Or maybe you’re constantly adjusting your prompts to account for missing information. It’s understandable, and we’ve all been there.
Or, you’re at the start of your journey and want to make sure you’re setting yourself up for success. Maybe you’ve got an excellent in-house data science practice, and are finally ready to make that data useful. Internally for a team, or externally for a new product category.
Whatever the case may be, we’re here to help. It’s clear now that data is the make-or-break for any successful machine learning project. It’s our number one wish when we start working with clients, and it’s probably what we spend the most time on. Getting this step right makes everything easier - training, fine-tuning, testing, productionizing, securing…we could go on.
Let’s take a look at some different aspects of data maturity and how they contribute to a successful AI project.
The secret is out. You can’t start a robust AI project without data engineering principles, pipelines, and data analytics.
All of these affect the performance of your AI project. It’s almost impossible to have a successful AI feature without quality data, and a scalable way to access it. Let’s see where your practice might lie on the scale.
It’s helpful to see where you fall on the data maturity scale. If you’re not at level 5 or 6, you should consider investing in moving up.
“It’s not just whether or not you have a data scientist. Data maturity really comes down to the state of your data, and how easy it is to gain insights from it. It’s really hard to get insights if you’re still working with disorganized data.” says Jurgis Samaitis, ML Engineer at NineTwoThree.
Ultimately, these levels answer one question: “What can I do with my data without spending too much time on figuring it out?”
You’d be surprised, but a lot of organizations start here. Maybe it’s a ton of papers they need to convert to PDFs, or logbooks they need to digitize.
Obviously, there’s a ton of drawbacks at this stage - zero automation, no redundancy, and next to no searchability.
We can do better.
By now, you’ve graduated to digital storage in a cloud or local hard drive form. These are probably in folders that might have informational labels, but probably don’t.
You might have heard that AI can deal with unstructured data like PDFs and text files well, so being in this state isn’t the worst place to be. But to really start to get insights, you’ll need more structure. There’s no possibility for automation when there’s probably no consistency in structure and naming.
Now we’re getting somewhere. Modern relational databases offer excellent ability to query and modify fields, which is what we’ll need to create a robust data pipeline.
But they weren’t designed with modern data science problems in mind.
To really take your data maturity to the next level, you need to invest in data pipelines and processes.
A dedicated department with engineers can look at things like cleaning and normalizing data, building out an ETL layer, and more. They can proactively ensure data health is high and that data is available 24/7.
There’s still a lot of upkeep around cleaning, updating, and maintaining data though.
Now we’re talking. Data that DOES things. Visualizations, dashboards, basic statistics and analytics. Even basic machine learning like a linear regression or two?
We see this a lot in Ecommerce - data flowing from Google Analytics that shows live sales, conversions, churn, etc. Simple raw data and analysis.
This is technically a product, but usually just used internally as a health check, or in a weekly presentation.
Now we can start thinking about some advanced AI capabilities. Data scientists can start to use better analytics and predictive models to understand their data better.
We’re shifting the focus from data management to predictive and prescriptive analytics – enabling businesses to understand why things happened, and what is likely to happen in the future.
The final boss. Using your data to build actual products that drive the business.
The data is no longer the product of the team - the outcome is. You can scale this rich data out into an AI-powered product and take full advantage.
Keep in mind, this doesn’t have to be an external product, although it often is. If you’re using your data to make your employees’ lives easier, then congrats - you’re at level 5.
All that matters next is how high you can scale.
This is optional, but valuable.
If your data maturity is robust, you can apply it to different use cases, rather than one project. It becomes something everyone can benefit from, even externally.
Few make it this far, but once you do, consider your project skyrocketed.
When data quality is lacking, the first place to turn is an LLM prompt. Imagine writing something like “If you don’t get an answer, rephrase the question and query again” or something similar as an instruction to your LLM.
This is problematic for a few reasons:
LLMs are famously non-deterministic. You can’t guarantee the same answer each time. Now, there are ways to guarantee performance and general accuracy…with great data forming the backbone.
Bad data maturity, bad backbone.
Poor data maturity directly translates to poor performance. And it’s not just worse answers from your AI chatbot - it’s all the annoying things users HATE dealing with.
If you don’t have a robust data pipeline and architecture set up, that means it’ll take longer to query for information. So…expect higher latency for requests.
You can have a state-of-the-art vector database, but poor data quality will drag its performance down as it searches for the right matches.
The first two lead to the most important aspect: security.
It’s not just about making sure the new hire doesn’t drop the production tables. (Although…yes, that’s important!)
A mature data practice places security first and foremost. Especially if you’re dealing with sensitive data.
You need to consider things like:
Mature data practices have answers to all of these questions.
Generative AI is still somewhat of an emerging field, and there are attack vectors that weren’t possible before. It’s easier than ever for bad actors to slip in personal information, and prompting an LLM in the right way can lead it to reveal too much information.
Prompts and data sources need strict access requirements.
As you can see, “the next level” of AI projects isn’t just about performance - it’s about robust security for all the valuable data you’ve been collecting.
This is where a data engineering department, and not just an engineering department, makes a huge difference.
.webp)
I’ll be honest, this term is a little overused.
No one knows EXACTLY what it means. “Governing data” can mean different things to different people.
Ultimately, it comes down to usability. Maintaining the data, its structure, and access.
The idea of a data lake is enticing - one place to access all of your data. These are common in enterprise use cases, and for good reason - unifying data access is helpful when there are hundreds of fragmented teams.
Ultimately, while a useful tool, they can be hard to maintain. There are projects with low data maturity that have a robust data lake, and projects with high data maturity without one.
Most of the time, we recommend taking data from one specific database or source.
A huge part of data maturity is not throwing everything in one huge database, and THEN dealing with latency, failed migrations, and nonsense data.
This last point is one of the best measures of data maturity.
As we talked about earlier, LLMs are non-deterministic. It’s difficult to granularly predict their output, and that has a direct impact on performance.
Similarly, if you don’t invest in data maturity, it’s just as difficult to EXPLAIN the answer - understand the chain of reasoning the LLM followed.
You need to be able to define and explain how a system is working. You can’t just throw data at an LLM and hope it works 8/10 times.
Being able to answer the question of “If I include these three data points, what’s the expected output?” is a seriously underrated part of LLM development.
Because eventually, someone - the CEO, a user, a lawyer - will ask how an LLM got its answer. And if the only response is a shrug, you are not close to the correct level of data maturity.
We’ve just described the foundation of a successful AI project. The foundation affects the stability of every other part - security, stability, performance, and accuracy.
If you don’t fix the foundation, you’ll just be paying for it 10x over in latency and engineering headaches.
This isn’t just a “we have to spend time fixing this” headache - it directly translates to dollars. You’ll pay more for your users and for your requests because the LLM is trying to fix stuff on each query.
“A question I get asked frequently is something along the lines of “ChatGPT took my CSV file and gave me an answer, so why can’t we do that at scale?” But it completely ignores the cost of tokens, the ability to scale that query performantly to 10,000 users, and the guardrails to protect that data. That’s all part of robust data maturity.” says Vitaliius Chernie, ML Ops Engineer at NineTwoThree.
It all adds up.
As an AI studio, we invest most of our time in developing robust data maturity practices for our clients. It pays off, and we’ve seen plenty of AI products and tools pay for themselves.
If you’re ready to learn more, let’s chat.
