Questions & Responses RFI 2223-36.pdf

Vendor Questions & Department Responses RFI 22/23-36, FDOR Solution for Fuzzy Matching Data

(Business Record Linkage)

Good Morning:

Questions received timely and the Department response are below:

1. Question: In order to inform development of relevant solutions, can the Department provide the approximate volume of data/transactions that the requested solution will be applied to?

Department Response:

The current system maintained by the FDOR comprises a total of approximately 294 million records. The average volume of third-party records processed on a regular basis is estimated to be up to 250K, which are then subjected to fuzzy matching with 14 million records. On a monthly basis, the FDOR imports approximately 30 main sources of third-party files, however, additional files may be received periodically throughout the year in accordance with changes in taxable transactions established by the Florida legislature. The third-party data files are received on a daily basis by the FDOR and consist of a mixture of individual and business entities.

2. Question: Can the Department provide context on whether any software solution would have to be housed on Department of Revenue platforms or if a cloud based/third party hosted solution would be acceptable?

Our preferred option for hosting the ultimate solution is on the FDOR cloud. If the solution is hosted on a cloud outside of the FDOR cloud, it would need to comply with IRS Publication 1075 2.E.6.1 Cloud Computing and 3.3.1 Cloud Computing and we would need confirmation that the vendor is FedRAMP certified.

In addition to the cloud hosting options, we would greatly prefer a secured REST API to call from our existing SSIS packages we currently have established.

Regardless of whether the selected vendor will process data on our behalf or host on a cloud service, the vendor will need to follow Publication 1075, Exhibit 7 Safeguarding Contract Language. Also see all sections related to contractor/sub-contractor facilities (if the data was sent offsite) and AC-20: Use of External Systems.

As the Florida Department of Revenue, we take the protection and confidentiality of our data very seriously. All information contained within our databases is deemed "protected" and "confidential," and cannot be disclosed to the general public. This is to ensure that sensitive and personal information is kept secure and not misused. The FDOR is committed to maintaining the highest level of data privacy and security to ensure that all data remains protected and confidential at all times.

3. Question: Please confirm due date. Section 6. Response Submissions states

"Noon Monday, February 27, 2023" and also states "no later than 12:00 p.m. Eastern Time, Friday, February 27, 2023."

Response due date is Noon on February 27, 2023.

4. Question: Can the various flat files of data be sent to third party organizations for the matching (in a secure fashion) or is FDOR solely looking for an internal solution that would be integrated to their existing systems? The reason we ask is Marketing Systems Group (MSG) has sophisticated database matching processes that can be adjusted internally based on the format and structure of input data to improve match rates, rather than providing a “canned” solution.

We are in search of a solution that can be housed on our internal systems or cloud-based platforms. Our data is considered "protected" and "confidential", and as such, we must ensure that all data is kept secure and is not shared with the general public. With that being said, we are open to exploring various solutions that meet these requirements, including a so called “reported sophisticated database matching process like the one offered by Marketing Systems Group (MSG)”, as long as the format and structure of the input data can be adjusted internally to improve match rates and the solution is able to be integrated into our existing systems.

Our preferred option for hosting the ultimate solution is on the FDOR cloud. If the solution is hosted on a cloud outside of the FDOR cloud, it would need to comply with IRS Publication 1075 2.E.6.1 Cloud Computing and 3.3.1 Cloud Computing and we would need confirmation that the vendor is FedRAMP certified.

5. Question: Do the flat files for fuzzy matching under this RFI only contain the residual data that did not initially match based on SSN or FEI?

Department Response:

The Department is faced with a challenge in its current process when the imported records are lacking a FEI or SSN. In such cases, the Department is required to resort to "fuzzy matching" based solely on the individual's name and address. Regrettably, these records often lack any additional "unique" identifiers that could assist in identifying potential matches. The flat files received from various external organizations to the Department contain data fields such as name, address, FEI, and SSN, but it is possible that some of these records may not contain a FEI or SSN, resulting in the need for fuzzy matching based purely on name and address.

6. Question: What is the expected frequency (i.e., daily, weekly, monthly, etc.) for performing the fuzzy matching?

The current system contains a total of 294 million records. The normal volume of 3rd party records to process up to 250K and these records are then fuzzy matched to 14 million records. The FDOR usually imports about 30 different main sources of 3rd party files on a regular monthly basis, and the 3rd party files are received daily by the FDOR.

7. Question: Can you please detail the number and allocation of records which are persons (individuals) or businesses (non-person entities) which the MDM solution should address with matching and outcome management? The mix of persons vs.

non person entities has an impact on our solution.

The third-party data files received by the FDOR contain a mixture of individual entities and business entities, such as corporations and partnerships, depending on the file source. The current system contains a total of 294 million records. The normal volume of 3rd party records to process is up to 250K and these records are then fuzzy matched to 14 million records. The FDOR usually imports about 30 different main sources of 3rd party files on a regular monthly basis, and the 3rd party files are received daily by the

FDOR.

8. Question: What's the current match accuracy for data that doesn’t have an identifier (e.g., SSN)?

Department Response:

We have not implemented a systematic method to measure the match accuracy for data that doesn't have an identifier such as an SSN. Given the large volume of data we receive, consisting of 294 million records, and the diversity of sources, which can number up to 30 different main sources on a monthly basis, it would be challenging to accurately determine the match accuracy for data without an identifier. Additionally, the data received from these sources may also vary widely in terms of quality and completeness, making it even more difficult to assess the match accuracy.

9. Question: How many people would be using the tool? Are they members of a specific team(s) or is this broadly available?

Department Response:

The allocation and usage of the tool would depend on the specific needs and requirements of different teams and departments within the organization. The tool would be made available to all relevant personnel who need to perform data matching tasks, as well as any other stakeholders who may require access to the data. The exact number of users would be determined based on the specific implementation and deployment of the solution. We expect there would be 5-6 users with a possibility of additional users in the future.

10. Question: Can they share some obfuscated / sanitized examples of the data they’re working with?

Yes, though confidentiality requirements prevent us from providing live data records, we can provide a test set of records that is similar to the data sets we normally match.

11. Question: What’s the number of records that the tool would be expected to process? What’s the cadence (e.g., real-time, on-demand daily, weekly, etc.)

The current system contains a total of 294 million records. The normal volume of 3rd party records to process is up to 250K and these records are then fuzzy matched to 14 million records. The FDOR usually imports about 30 different main sources of 3rd party files on a regular monthly basis, and the 3rd party files are received daily by the FDOR.

12. Question: Would a flexible solution like Alteryx be something they’re interested in

(have experience/license using)?

Department Response:

As the Florida Department of Revenue, we are always open to considering different solutions that can improve our current processes and meet our requirements. Whether we have experience or a license using Alteryx would be evaluated based on its capabilities and compatibility with our current systems. Ultimately, our primary focus is to find a solution that can effectively support our record matching process, and we are open to exploring different options to achieve that goal.

Our preferred option for hosting the ultimate solution is on the FDOR cloud. If the solution is hosted on a cloud outside of the FDOR cloud, it would need to comply with IRS Publication 1075 2.E.6.1 Cloud Computing and 3.3.1 Cloud Computing and we would need confirmation that the vendor is FedRAMP certified.

13. Question: The RFI mentions the current process as, “The Florida Department of

Revenue - Lead Development Process receives files from various other agencies or third-party data providers. These files are imported into an internal application and the file records are matched to existing records in the database.” Am I correct in understanding that the matching process is all managed by internal employees or are there any other vendors or organizations contracted for support that would be replaced in the event of a solicitation for this desired solution? If all internal, do you have any estimate of approximately how many employees currently manage this process?

As the Florida Department of Revenue, the matching process is managed by internal employees. There are no other contracted vendors or organizations that are involved in the process. The data in our databases as it is considered to be of high sensitivity and value. The data includes personal and financial information of individuals and businesses, and it is our duty to ensure that it is not disclosed or made accessible to unauthorized parties.

The number of employees managing the current data matching process may fluctuate in accordance with the volume of incoming data and the prevailing workload. However, the process is typically managed by a dedicated team of 5-6 employees within the Department.

14. Question: What is the typical volume of records that need to be matched in any given period?

Department Response:

The current volume of records within the FDOR system is approximately 294 million. The FDOR typically processes up to 250K third-party records, which are then subjected to fuzzy matching with 14 million existing records within the database. The FDOR receives an average of 30 different primary sources of third-party data files, with daily imports and a varying frequency of additional files throughout the year, dependent upon changes in taxable transactions as outlined by the Florida legislature. The third-party data files consist of a mixture of individuals and non-person entities, such as corporations and partnerships, and are usually presented in the form of text files for ingestion into the FDOR database. The match accuracy of the third-party data varies widely, depending on the source of the data.

15. Question: Can you provide a few sample records that are representative of the data sources that will need to be matched?

Yes, though confidentiality requirements prevent us from providing live data records, we can provide a test set of records that is similar to the data sets we normally match.

16. Question: What organizations/sources outside the Department of Revenue send flat files?

I'm sorry, but as all data in the FDOR databases are considered protected and confidential, it would not be appropriate to publicly disclose the organizations or sources that send flat files to the Department of Revenue.

17. Question: What are the flat file formats that are used by outside organizations?

Department Response:

The format of the flat files received from outside organizations vary, but typically they are presented as text files. To maintain the confidentiality of the information contained within these files, the Department does not publicly disclose the specific file formats used by outside organizations.

18. Question: Can attributes other than name and address be used to link records when no FEI or SSN is available?

Department Response:

If a FEI or SSN is not available, the usual practice is to link records based solely on the attributes of name and address. No other qualifiers are typically present in the data.

19. Question: Can you please clarify the requirement “Must be able to link multiple data sets (rather than a single pair) and are ideal in circumstances where no set of data is a superset of another”?

The requirement for the ability to link multiple data sets refers to the capability of the proposed solution to match and associate multiple sets of data, rather than just a single pair. This feature is particularly important in scenarios where none of the data sets is a complete representation of another set. It is essential that the solution can identify connections between multiple data sets, rather than just matching pairs of data.

20. Question: Can you please quantify the requirement “Ability to process vast amounts of data in short time frames”, particularly in terms of defining “vast” and “short”?

Based on the current system, the FDOR processes a vast amount of data, with a total of 294 million records in its databases and a normal volume of approximately up to 250K third-party records to be processed and fuzzy matched to 14 million records. The FDOR receives around 30 different main sources of third-party files on a regular monthly basis, with additional files received throughout the year, and the data files are received and processed through nightly batch jobs. The third-party files contain a mixture of entities, including individuals and business entities, and are normally presented as text files that are ingested into the FDOR's database. The match accuracy for third-party data varies widely depending on the data source.

21. Question: What is the success rate for matching files records to existing records using the current process?

Department Response:

The match accuracy for third-party data processed by the Florida Department of Revenue (FDOR) varies widely depending on the data source. This implies that the success rate of matching file records to existing records in the FDOR database is subject to significant fluctuations, dependent upon the particular origin of the third-party data being processed. This highlights the need for a more robust and reliable system for matching records, which can ensure a higher degree of accuracy and consistency across all data sources.

22. Question: What is the false positive rate when matching files records to existing records using the current process?

Department Response:

The Florida Department of Revenue has not implemented a systematic method to measure the false positive rate for record matching processes from the various sources of data it receives. Given the sheer volume of data, consisting of 294 million records, and the diversity of sources, which can number up to 30 different main sources on a monthly basis, it would be challenging to accurately determine the false positive rate for the current matching process. Additionally, the data received from these sources may also vary widely in terms of quality and completeness, making it even more difficult to assess the false positive rate.

23. Question: What is the current process for identifying false positive linked records?

The Florida Department of Revenue faces the challenge of accurately identifying false positive linked records in its current process. The sheer volume of data received from various sources makes it difficult to consistently measure and detect false positive matches.

However, our team of data analysts and lead development personnel are equipped to detect false positives on the front end. Additionally, our auditors and caseworkers are responsible for reviewing and verifying matches on the back end.

When a false positive is detected, steps are taken to improve the detection process and enhance the overall accuracy of the data matching process.

24. Question: What is the business consequence of falsely linking records?

Department Response:

Falsely linking records can lead to a number of unintended consequences for the Department, including errors in data analysis, incorrect information being shared, and incorrect decisions being made based on that information.

Additionally, it can harm the reputation of the Department by spreading false information or making incorrect decisions that negatively impact stakeholders. In some cases, it may also result in legal or financial consequences. It is crucial for the Department to maintain accurate data and ensure that any linked records are accurately matched to prevent these consequences.

25. Question: How many source systems and how many master data files are you currently performing fuzzy matching on?

As the FDOR, we receive and process data from various sources on a regular basis. The exact number of source systems and master data files that we perform fuzzy matching on can vary, but typically we import about 30 different main sources of 3rd party files on a regular monthly basis. These files contain a mixture of entities that are individuals and business entities, and we aim to match these records to our internal database of approximately 14 million records. The volume of 3rd party records to process is usually up to 250K. However, the exact volume and number of source systems can change depending on changes in taxable transactions laid out by the Florida legislature, which can then result in the ingestion of related additional databases into our main database.

26. Question: Could you please provide some high level estimates on record counts for datasets used in the matching process?

We currently import and process approximately 30 main sources of third-party data files on a regular monthly basis, with the normal volume of records to process being up to 250K. These third-party records are then fuzzy matched to 14 million records in our internal database. The records that are processed contain a mixture of individuals and business entities, and the data files are received daily. The total number of records in our internal database is 294 million.

27. Question: Outside of FEI, SSN name and address, what other data attributes are available for each individual’s records?

Department Response:

The Florida Department of Revenue typically only has access to name and address information for individuals if FEI or SSN is not available. In such cases, other data attributes beyond these identifiers are generally not available for linking records.

28. Question: Are you looking to enhance your current solution or are you open to other tools and technology?

Department Response:

The Florida Department of Revenue is open to exploring a range of solutions to improve its current process of fuzzy matching records, including the enhancement of its current solution as well as the examination of other tools and technology. This includes the potential incorporation of artificial intelligence and modeling techniques that can be utilized in similar fuzzy matching algorithms. The Department is dedicated to ensuring that it adopts the most effective and efficient methodologies to achieve its desired outcomes, and is open to considering a broad spectrum of possibilities.

Our preferred option for hosting the ultimate solution is on the FDOR cloud. If the solution is hosted on a cloud outside of the FDOR cloud, it would need to comply with IRS Publication 1075 2.E.6.1 Cloud Computing and 3.3.1 Cloud Computing and we would need confirmation that the vendor is FedRAMP certified.

29. Question: Are you open to acquiring additional third party data to further supplement your current internal data?

As the Department of Revenue, we understand the significance of acquiring additional third-party data to supplement our current internal data. This approach is commonly used by organizations to improve the accuracy and robustness of their data-driven decision making processes. The acquisition of third-party data can provide valuable insights and perspectives that are not available in the current internal data. This can help to improve the accuracy of data matching processes, reduce the risk of false positive matches, and ensure that our data-driven decision making is based on the most up-to-date and comprehensive information.

However, the acquisition of additional third-party data is not a decision that should be taken lightly. The Department of Revenue will carefully evaluate the benefits and costs involved before making a decision on whether to acquire additional third-party data. This will include a thorough analysis of the quality and relevance of the data, the costs involved in acquiring and integrating the data into our existing systems, and the potential impact on our data privacy and security measures. Ultimately, the decision to acquire additional third-party data will be made by the relevant stakeholders after careful consideration of all the factors involved.

30. Question: Are you looking for a fully automated process or ok with manual review at established checkpoints, if it means increased accuracy?

Department Response:

As the Department of Revenue, we are open to both fully automated processes as well as limited manual review at established checkpoints, as long as the solution provides increased accuracy in the data matching process. The specific approach that is chosen will depend on the results of our evaluation of the available options and their ability to meet our specific needs and requirements. Ultimately, the goal is to achieve the highest level of accuracy in the data matching process, and we are willing to consider both fully automated and manual review approaches to achieve this objective.

Thank you for your interest in doing business with the Florida Department of Revenue.

Sincerely:

Ed Valla, FCCN, FCCM, FCPM, CPPB Purchasing Analyst EXE Program/Office of Financial Management/Purchasing Florida Department of Revenue

(850) 717-7567 edwin.valla@floridarevenue.com mailto:edwin.valla@floridarevenue.com