That's why Michigan is taking an infrastructure-first approach. While AI remains part of the state's long-term aspirations for its MiGreatDataLake — an educational data ecosystem in the making — its leaders told the Center for Digital Education this week that the immediate challenge is not deploying AI itself, but building the secure, interoperable infrastructure necessary to support future data analytics.
Last August, the initiative’s Project Development Director Tammy Evans said the goal was to resolve systemic inefficiencies, including disconnected student records, incompatible district platforms, redundant database systems and excessive administrative friction.
This week, Project Director Doug Leisenring said MiGreatDataLake is evolving into a long-term modernization effort where, because AI capabilities for the data lake remain largely future-facing, the state is treating technical and human infrastructure as strict prerequisites. He described the project as deliberately paced, noting that statewide educational data systems cannot effectively scale without established trust and governance.
“What we're trying to do is design a program that will support districts across the state who all will use different tools, who all have different staffing levels, who all have different levels of access to resources," he said. "We're trying to design a tool that will meet the needs of all educators in Michigan, whether you live in River Rouge or you live in Rapid River."
Leisenring said the initiative is still in a proof-of-concept phase that builds on a broader framework established in August 2025, when project leaders first discussed the vision for a statewide interoperable educational data system. That initial framework emphasized holistic student data designed to reduce barriers to actionable information and AI-assisted insights.
At that stage, however, operational agreements with Amazon Web Services, the main collaborator in the initiative, and other technical partners were still under negotiation. Leisenring clarified that public discussion of the initiative preceded its actual execution.
“We didn't sign that contract and get started until January [2025],” he said.
Moreover, school districts are not yet broadly using data from the lake. Instead, Leisenring said, the proof-of-concept phase has focused on validating the data pipeline: ingesting structured educational data from student information systems, moving it through the lake's architecture, and integrating it with two pilot applications. One of these, MiEWIMS, is an early warning and intervention system; the other, MiRead, is a literacy intervention planning tool designed to help districts draft reading plans for individual students performing below grade level.
“Our first step was, let's make sure we can move that structured data from the student information systems and their assessment tools into those applications," he said. "We're taking data from our data hub system and putting it into those tools."
Architecturally, Leisenring said the lake organizes data through a multistage pipeline known as a medallion pipeline: Raw educational data is first ingested into the "bronze layer," before being cleaned and standardized in the "silver layer," then finally aggregated in the "gold and platinum layers," where it becomes ready to feed targeted tools like MiEWIMS and MiRead.
“We know we have built an operational data lake, because we're able to take raw data, put it in through the medallion process, move it from the raw bronze layer ... standardize it in the silver layer, and then aggregate it for those tools in the gold and platinum layers,” Leisenring said. "We're able to move data through those layers. Now, the next challenge is to start doing this on a more broad basis across the state."
Leisenring added that as of now, his staff are manually moving data through each medallion layer. The next step will be to create an automated process where data transitions between each layer effectively without direct human involvement.
“Over the summer, we will start onboarding districts that will get their data for those two tools, MiEWIMS and MiRead, from the MiGreatDataLake, and then from there we’re also working on our road map over the next three to five years on what other data sets are we going to ingest into the lake, to help teachers and principals make decisions on how to improve instruction,” he said. “Then at every level of that road map is going to be more training and more support, not only for their data people, but also for their teachers and their administration on how to use those tools to help their kids.”
Leisenring said the deliberate pace of the rollout reflects a consensus among project leaders that a single high-profile privacy or security failure — not unlike the recent Instructure breach — could permanently undermine district participation.
“We're trying to come up with the common recommendations for good data practices that will help data flow into the lake to get them better outputs,” Leisenring said.
This challenge is reinforced by Michigan's structure as a local control state for education, he continued. Because local school boards independently set directives, districts across the state operate with completely different software tools, varying IT staffing structures, disparate governance models and unequal resource allocations. Consequently, deploying the data lake is not merely a technical installation — it also requires extensive coordination between statewide systems.
“We're all upgrading our systems,” he said of different state-level departments. “We need to connect the data and open up actionable insights for educators, and I know that's going to improve instruction in Michigan and outcomes for kids.”
To establish formal oversight over the statewide educational data system, Leisenring said this summer project leaders will form the Michigan Data Lake Data Trust, a governance body that will include representatives from different regions of the state, officials from the Michigan Department of Education and independent privacy experts.
By treating governance as operational infrastructure, Leisenring said, the state is signaling that it will not scale data collection without established rules.
"We have to have the data trust board in place before we inject any other data sets,” he said.
