In a solicitation released June 5, the California Department of Technology (CDT), on behalf of the California Department of Tax and Fee Administration (CDTFA), seeks a “scalable and reliable tax return analysis solution with data analytics” that uses machine learning and artificial intelligence to improve tax return processing accuracy and efficiency, to “provide timely tax compliance advice and actions to better serve the taxpayers.”
The departments refer to the solution as the Data Analytics and Tax Return Processing Solution (DATRPS). The solicitation is a “phased approach” with multiple bidder submissions and evaluations, beginning with a proof of concept (POC). The POC development period will be Nov. 6-Feb. 7, 2024. The state will evaluate POCs from Feb. 16-Feb. 26.
Among the takeaways:
- CDTFA administers “numerous tax and fee programs” that mandate taxpayers submit returns following assorted time frames. The returns are processed via “many complex rules in accordance with specific tax programs” and it can take months for issues to come up, for taxpayers to be notified and for corrections. Among its challenges, the existing solution can’t fully leverage internal data or third-party data sources; can’t get data in a timely manner; can’t verity “business activities, transactions and business revenue”; can’t analyze large volumes of data; can’t fully do taxpayer return analysis; is unable to efficiently create new business models; or to determine anomalies and take timely discovery action. Results include more time needed for analysis; missed opportunities for revenue, customer engagement and education; delayed and inaccurate tax discovery actions and analysis; increased penalties and interest for taxpayers; the inability to identify and confirm business entities and activities and a low frequency and volume of data collection and preparation such as forecasting and trend analysis. Constraints on the current solution include that CDTFA must be able to access and maintain the data analytics models in the solution; there’s no staff data science expertise that would “support new data analytics models and machine learning capabilities;" and the data analytics solution will not procure new data sources.
- DATRPS, plus “appropriate various external data sources procured separately, if needed,” is intended to improve the speed and accuracy of return processing. The department “envisions a multi-wave approach and delivery of a product solution with its enhancements encompassing multiple years” — up to three years. However, it said, “a complete solution can be provided in a single wave or in a mutually agreed time span as well.” The department has planned for up to three waves including a POC. The POC time period will be 90 days and the state may subsidize part of the expense “based on the bidder’s estimate in their Rough Order of Magnitude.” The POC goal will be for the bidder to show its solution and build a “workable solution design” addressing all POC “scenarios and associated requirements.” Among requirements, the solution must use “all available data sources” to clean and “adjudicate erroneous data;" must be able to implement “customer tax return compliancy law” to ascertain the probability of errors and compute under- or over-payment; and use machine learning to identify common return errors.
- The first wave will encompass the contract’s first 12 months with a cost not to exceed $1 million. Its goal is to use the POC to get a solution that solves CDTFA’s “end-to-end business tax compliancy and audit selection process.” The solution must be “enterprise class” and in the cloud; must build and expand on POC scenarios; must provide “further reclassification capabilities for codes beyond the POC; and must have access to machine learning models and to data scientist staff, both for edit and modification.” Deliverables include implementing a data warehouse or data lake. The solicitation’s second and third waves, indicated as “Option Year 1” and “Option Year 2” of the contract, respectively, would swell the contract value to $10 million for waves 1-3. Wave 2’s goal is to expand the “Wave 1” solution and make it do more, including serving as a solution capable of “addressing new machine learning problems;" real-time data visualization; and real-time public dashboards and application programming interfaces (APIs) to CDTFA’s other data sources. Wave 3’s goal is to finish and implement the solution with “final enhancements” including a data scientist as a consultant reviewing “final machine learning models;" APIs for all the department’s data sources; and finishing the data warehouse or data lake.
- Contractor responsibilities include providing “all necessary staffing resources” to complete the POC, its deliverable and milestones along the development timeframe. The company chosen must provide its “own equipment, own data cleansing/matching sources, POC environment and software” needed to implement the POC and develop POC scenarios. Contractors must design and develop the POC remotely at its place of business during the designated period; present weekly to the state its current “system design and implemented POC system;" maintain security for the system and any state data in the POC and throughout its life cycle and “be responsible for the successful development and deployment of a ‘functional, practical, feasible and workable’ POC.” The contractor’s staff must take mandatory CDTFA disclosure and information security training and pass two Franchise Tax Board security questionnaires (standard and cloud computing). Once the POC is complete, contractors must “remove the trial software and purge all CDTFA data.” Respondents must have experience including having completed at least two data analytics projects using machine learning models in the last five years, one of them within the last two years.
- Questions for the bidders’ conference are due by 5 p.m. Monday. RSVPs for the bidders’ conference are due by 5 p.m. Thursday. The bidders’ conference is slated for June 20. Intents to bid are due by July 5. Deadlines for each wave are available here.
- Contract award is slated for June 13, 2024, with the contract starting June 17. The initial solution agreement term is one year, including implementation and production support. The state reserves the right to exercise two optional one-year extensions, making the maximum agreement term three years.
