Datafold

I used Datafold for data quality checks while working on a data migration project where we moved data and applied business rules and transformations. After that movement, we had to check if the data loaded was correct, and Datafold helped significantly because it has great resources for customizing queries and defining primary keys for comparison. We could set the source and target tables, specify rules, and then press play, which provided excellent results regarding data matching percentages.

I used Datafold for data quality checks while working on a data migration project where we moved data and applied business rules and transformations. After that movement, we had to check if the data loaded was correct, and Datafold helped significantly because it has great resources for customizing queries and defining primary keys for comparison. We could set the source and target tables, specify rules, and then press play, which provided excellent results regarding data matching percentages.

Sometimes the mismatches were minor, but other times they required more attention. Being able to use Datafold for these quality checks while working on other processes was really helpful.

I know that bugs can be related to misconfigurations, but we had issues with comparisons where executing the queries simply did nothing, and we didn't have much information about why it failed. I believe having more detailed information about why the comparison didn't work would help with debugging, so a more detailed log would be really beneficial.

I worked with Datafold for one project that took about one year, and that is my experience with it so far.

We had experiences where processes took longer than expected, but it was unclear if it was related to Datafold or the database systems. Overall, it performed well, but sometimes not specifying the number of rows for comparison led to execution issues, as the tool struggled with high data volumes, which required us to find the right amount of data for comparison.

I wasn't involved in scalability issues, and I don't think I had access to those configurations as a data engineer, focusing instead on data quality checks without delving into setup or configuration.

I haven't contacted technical support or customer support from Datafold during the time I worked with it.

Before working with Datafold, I have never used similar tools. I know there are some options, but I haven't worked with them. My experience required writing customized queries or automation scripts for tasks that Datafold automates.

I didn't work with a similar tool before or after Datafold. Whenever I needed to do data quality checks, I always had to write a customized query or script.

The initial deployment was really easy. Once we understood how it worked and set it up, connecting to databases like Redshift, SQL Server, and Databricks was straightforward.

I'm not familiar with the pricing details, as it's not information that comes to us as data engineers, but I know it has a high cost. The client I worked with raised concerns about the pricing, but I'm not involved with the project anymore, so I don't know if they're still using it.

I talked with someone on LinkedIn about PeerSpot, and he mentioned something about a gift card to do the review, which is why I asked for this meeting because I'm not comfortable using the company email and I'm not sure if I'm authorized to do that. He explained that I should do a review about some tool that I have experience with to be eligible for a possible gift card, but I didn't know about the company before, to be honest.

I have been working in my current field for 11 years overall, dealing with data-related projects.

I have no questions and will be waiting for the next steps in the process. My review rating for Datafold is 8.

My main use case for Datafold is to automatically compare data differences through data diffing, where I can compare datasets row-by-row, column-by-column, to surface unexpected changes.

I also use it for CI/CD integration for data pipelines, where we embed data quality checks into GitHub and GitLab pull requests. Another interesting feature I use Datafold for is column-level lineage to trace how specific columns flow through SQL transformations across the data stack we have.

Recently, for data diffing, there was one time we wanted to conduct a lot of comparisons on a very large dataset to be sure that all was in check before we migrated to Snowflake, which is a complex and resource-heavy process.

We used Datafold to do the automatic checks across rows and columns to ensure that when we completed the migration, no data was lost and the migration into Snowflake was very smooth. That was in March, and it was very useful.

For a friend who uses Datafold in their enterprise, an engineer made a change from an SQL model and did not realize it would cascade and alter downstream dashboard metrics.

However, with the help of Datafold integrated into the pull request workflow, the data diff actually surfaced exactly which columns changed, which rows were affected, and by how much.

The data diff stands out the most for me as the best feature because it gives me that column-level and row-level comparison between any two datasets, highlighting what changed, even down to characters and whitespaces.

It also tells me when something has failed, and it makes debugging faster and more precise.

Column-level lineage derived from SQL static analysis helps me trace how columns flow through transformations across the entire pipeline.

Datafold has positively impacted my organization by helping us catch data regressions before they reach production.

Even while we are still building the pipeline and mapping out how the data will flow, we catch any data regressions and alterations before they reach production. This helps us avoid a lot of debugging when the migration has happened or when we reach production level and data goes live.

Additionally, it significantly reduces manual testing time, saving us time that we can put into useful work. Furthermore, it builds our team's confidence when pushing code changes, as the data reviewers, data annotators, and engineers can actually see the data impact of every change, not just the code change.

The first pain point for me is that the reporting capabilities are weak. The ease of setup is also challenging, particularly for those who are not tech-savvy or do not know how to navigate it. Most importantly, there is no free trial, so you cannot deploy and test it to see the efficiency and use case before purchasing.

I feel the features need more attention because sometimes finding the data I want to use or compare is challenging, especially when looking for it inside the SaaS.

Automated workflows also break sometimes, but whenever that happens, the good thing is that it gives you an error report so you know where the issue is coming from.

I have been using Datafold for about six months.

I give Datafold a seven out of ten because it is excellent at its core specialization, which is data diffing and CI/CD integration, and its migration automation capabilities are strong and very AI-powered. However, I lose three points due to the weak reporting.

Even though it provides reports when there is a breakage in your push or migration, I sometimes cannot get the full scope of what I want when producing an actual report. Additionally, the lack of a free trial is a downside.

Regarding Datafold's governance and security, I rate it high because its migration agent uses LLMs for SQL translation and validation, which is a strong point.

Additionally, there is a self-hosted deployment option available, so organizations with strict data residency or compliance requirements can run Datafold entirely within their own cloud environment, which could be either AWS, GCP, or Azure, ensuring that data never leaves the perimeter of the organization.

Datafold's output is highly accurate and very reliable. The migration agent's accuracy, which utilizes LLMs to convert SQL dialects, is excellent.

Furthermore, the data diff, which is the most reliable AI-adjacent feature, is deterministic, not generative, and it compares actual data values mathematically rather than using inference.

Because it does the comparison mathematically, the outputs are highly accurate and consistently reliable.

Datafold is deployed in my organization as a cloud, specifically as a SaaS, which is fully managed by Datafold. This means that hosting, maintenance, and automatic updates are all managed by Datafold, making it simpler for us and easier to get started.

The advice I would give others looking to use Datafold is that whoever is handling it, perhaps the head of IT, should have a sit-down with the analysts to ensure it fits into the stack that the organization is already conversant with.

Datafold is purpose-built for SQL and warehouse-based analytic pipelines with DBT, so if the current stack does not include a data warehouse and DBT, I would advise them to evaluate alternatives first. I also recommend using it for CI/CD quality, not just for general observability, because Datafold excels at pre-merge testing and data diffs.

Therefore, if the primary need is broad production or observability, the organization should also check out other options.

I rate Datafold a seven out of ten overall.