From Pandemic Pilot to State Infrastructure
New Mexico made Amira Learning’s reading program mandatory for kindergarten through second grade in 2025-26. By August 26, 2026, an Albuquerque Journal report hosted by Public Services Alliance said 15 of the state’s 89 districts and one charter school had refused it, sometimes temporarily.
The refusals exposed the gap between statewide adoption and the rules for voice data, consent and evaluation. For vendors, the binding work now begins after the model: clearing state approval, working through procurement, meeting privacy rules and finding recurring district funding.
The move from pandemic relief to general budgets makes finance and compliance capacity—not just model development—the part vendors must expand or rebuild. Zero G Talent’s live job board lists a Senior Accountant as Amira Learning’s newest opening, added in the past week. The posting alone cannot show how large the finance function has become, but it is an early signal that regulatory scale, not AI performance, is the binding constraint.
Amira’s path into that infrastructure began in 2020, when Bay City Public Schools in Michigan adopted Amira Learning, using both its assessment and tutoring platforms. Amira listens as a student reads aloud, then identifies gaps and gives targeted feedback. At that scale, vendors must win state review, procurement approval and a place in the budget.
The merger with Istation changed how Amira could win adoption. In June 2024, EdTech Chronicle reported that the companies had merged, with Amira designated as the go-forward business. The combined company expected to serve more than 1,800 school districts by the end of the year. That figure came from the companies, not an independent market count. The merger nevertheless gave Amira an existing network of district and assessment customers and a broader product base from which to seek larger contracts.
The merger also changed what districts could buy. For the 2025-26 school year, Istation began moving into Amira’s Intelligent Growth Engine, a science-based suite combining assessment, instruction and AI tutoring. The combined product connected reading screening, intervention and progress monitoring in one suite rather than leaving districts to assemble separate tools.
State action then moved the suite into the assessment system. In April 2025, Amira announced that the Texas Education Agency renewed Amira as a reading assessment for first- and second-graders. The renewal let all Texas independent school districts keep using Amira Assessment as an approved instrument. It did not require every Texas student to use the tutor; it placed Amira inside the state-sanctioned assessment workflow, where approval, reporting and district purchasing rules applied.
Approval and compulsory use are different stages. Most of the approvals cited here concern screening, dyslexia screening or progress monitoring, not compulsory tutoring. Both stages, however, move the purchasing decision beyond an individual teacher and classroom: a state now sets the rules governing whether and how districts can use the tool.
By February 2026, the Michigan Department of Education had approved Amira for screening and progress monitoring in grades K-3 for the 2026-27 school year. Amira also said its screener was approved for use in 39 states.
Public procurement follows a similar pattern outside the United States. Ontario approved Amira for distribution through a public procurement agency in 2025, the company’s first such approval in Canada. Amira said it was the only AI-powered reading-program and intervention partner in the provincial system. Jordan’s Ministry of Education then launched a national initiative with Amira and other partners that could reach 800,000 students. The program was set to include Arabic localization, locally relevant materials and assessment norms, professional learning and coaching, and pre- and post-program assessments. Amira was one of the technology and implementation partners in that broader public program.
The operating consequence is a longer chain: a classroom intervention now depends on a state or provincial purchasing file, a district contract and a payment trail that can keep it in use.
The ESSER Cliff Rewrites Who Pays
That chain now reaches the district budget.
| Category | Measure | Reported amount | Source |
|---|---|---|---|
| Education funding | Elementary and Secondary School Emergency Relief (ESSER) for K-12 schools | about $190 billion | federal relief laws |
| Education funding | Three ESSER relief packages | about $190.3 billion | Civic IQ’s July 2026 analysis |
| Public program spending | New Mexico annual package | $2.7 million | New Mexico’s Legislative Education Study Committee policy brief |
| Public program spending | New Mexico statewide assessment | $1.4 million | Albuquerque Journal report |
| Company revenue | Duolingo, 2026 projection | $1.2 billion | ValueAddVC |
The final obligation deadline passed on September 30, 2024, and LOOP Methodology says the last meaningful liquidation window closed in early 2025.
Relief money let districts hire reading interventionists, buy tutoring services and purchase educational software without first finding room in their local general funds. Factlen reported that before the pandemic, federal money covered about 8 percent of total K-12 spending. ESSER briefly pushed the federal share above 13 percent.
The timing mattered because districts could commit temporary dollars to recurring costs. A short grace period pushed some spending into 2025, while multiyear contracts and annualized relief purchases stretched commitments further. Once the grant ended, districts still had to fund staff positions, software licenses and services their standard budgets had never carried. Factlen describes the result as a structural deficit created when temporary money financed permanent commitments.
“It amounted to sometimes as much as $5,000 per student or more,” said Marguerite Roza, director of the Edunomics Lab at Georgetown University and research professor at the McCourt School of Public Policy. “But it was one time.”
That pressure is tightening as enrollment falls. Because most state formulas send money to districts per student, losing students removes revenue from the general fund just as relief disappears. Softer state revenue can limit replacement aid. An AI reading subscription now competes with teacher salaries and other recurring costs, not with less visible line items that emergency grants could cover without much internal conflict.
Software was especially exposed. Relief money could buy a subscription quickly, and a district could switch it off later. After the grant vanished, the same subscription became a recurring expense without a home in the operating budget. Vendors that treated ESSER as a route into a district must now persuade buyers to keep the product.
Districts can protect some purchases through budget classification. Loop Methodology found that core curriculum, assessment and multi-tiered systems of support largely escaped the cliff, while supplemental tutoring, summer programs and other one-time purchases faced sharper cuts. Duplicate, low-usage tools are especially vulnerable. A platform approved as an assessment instrument has a stronger claim to be treated as core than an optional classroom app, though that status does not erase the price.
The remaining funding paths are narrower. Title I continues to support schools serving many students from low-income households, and Title III continues to support English-language learners. E-rate covers eligible connectivity costs, not the full cost of an instructional platform. States can backfill lost aid through literacy programs or general aid; districts can seek voter-approved money. Bonds cannot sustain a recurring subscription. The general fund renews annually, making it the recurring source a vendor must win.
State choices determine how painful the transition becomes. Some legislatures replaced lost federal aid with new money for literacy, safety or general operations. Others held funding steady and left districts to absorb the loss. Vendors cannot use one national sales strategy for that variation. A contract that rides on a state appropriation in one market may need a case for a local general-fund purchase in another.
Districts are also deciding earlier. Preliminary budgets take shape in late winter and spring, and most boards adopt final budgets before the July start of the fiscal year. Civic IQ found district board agendas discussing these purchases months before a request for proposals appeared. Waiting for the formal solicitation to explain a tool’s value leaves a vendor little time to strengthen a weak funding case.
Boards are already scrutinizing individual lines. Civic IQ’s May 2026 board-agenda review found officials in Chelsea and Farmington, Michigan, rejecting portions of proposed intermediate-district general-fund budgets. Those disputes put discretionary software under direct line-item scrutiny, ahead of staffing and core instruction. That puts usage data and measured results beside price in a district’s defense of a line; a pilot promise alone is less likely to survive.
Districts are not abandoning technology. Becker Public Schools in Minnesota approved a June 2026 renewal batch worth about $123,111 across more than 10 educational-technology and curriculum vendors, using operating dollars. Loop Methodology described the shift as districts buying differently rather than simply buying less. Civic IQ expects fewer vendors, larger and longer contracts, and evidence before commitment, with duplicate, supplemental and low-usage tools reviewed first.
That change favors vendors that can become part of a core curriculum, assessment or reporting workflow. A single-purpose tool faces greater pressure unless it is bundled into one of those systems or proves distinct value. Vendors must supply usage summaries and outcome evidence before the budget meeting rather than after the renewal notice.
For AI reading vendors, the transition turns each budget meeting into a test of durability: every contract needs a durable funding source and district-specific evidence behind its price. Securing the line item is only half the job. Keeping a public contract paid, documented and renewed is a separate operating discipline.
Compliance Is the New Product
The accounting question is not settled by a paid invoice alone. Amira Learning’s Senior Accountant posting makes the associated work explicit: the role supports revenue recognition and deferred revenue under ASC 606, prepares draft financial statements, tracks contracts and purchase orders, reconciles deferred revenue to the general ledger, and processes credit memos and reversals.
The posting also calls for monthly revenue recognition and related journal entries. Deferred revenue becomes a control problem. If a district pays before Amira delivers the service, the accounting team must track the unearned portion, connect it to delivery and the general ledger, and reconcile it as the company satisfies its obligations. Errors can leave contract liabilities, invoices and actual delivery out of sync.
General-fund contracts also expose renewal risk. A district can end a subscription or decline to renew it at budget cycle. Finance needs to know which commitments recur, how much revenue remains deferred and where a lost renewal would hit cash forecasts.
The same job board also lists Principal Engineer and Machine Learning Engineer / AI Systems Engineer. The postings do not reveal team size, but they place the new role beside continued technical build-out. That mix is consistent with a company managing both the product and the machinery required to sell, bill and support it at public scale.
Tax exposure adds a second layer. The Senior Accountant posting requires the role to prepare and file multistate sales/use, income and franchise tax returns and to support coordination of multistate tax filings and the annual audit. For a company operating across multiple states, the work is jurisdiction-specific rather than one national tax process.
Public purchasing changes the contract itself. State or provincial approvals create distinct channels, and each agreement carries its own terms. Approval opens a channel; districts place and renew contracts; the vendor recognizes revenue under each. The close process must answer three questions: Is the product approved? Has the district bought and paid for it? Has Amira delivered what it owes?
Privacy and research evidence now sit alongside delivery records. New Mexico has set deletion rules for voice recordings, and Boston University’s EVAL Collaborative is designing an independent randomized controlled trial for Amira. Vendors need records that connect those commitments and results to service and renewal documentation. Compliance is not a separate legal notebook. It becomes part of what the company can invoice, deliver and renew.
At that scale, the paperwork is part of the product: the vendor must document delivery, recognize revenue, tax the sale and renew the contract without a compliance failure.
That scale calls for a finance function that treats renewal risk as operational data, reported by jurisdiction, contract type and funding source, rather than as one national subscription book.
Those controls are not abstract. When a child’s voice is at stake, they become a public fight.
The Pushback: Parents, Privacy, and Procurement
A May 2026 brief from New Mexico’s Legislative Education Study Committee found no AI-specific procurement standards, no AI-specific rules for student-data privacy and no formal process for deciding whether tools such as Amira work. New Mexico was one of three states that had legislated a study of AI in education without adopting a formal policy.
An Albuquerque Journal report later said the package covered software, professional development, data analysis and other tutoring and support programs. The purchase funded instruction and support at once, but it did not supply a procurement rulebook, a consent standard or a process for comparing Amira with other methods.
At school board meetings, parents raised concerns that the system was saving and recording student voices and that they did not know where the data was stored or how it was used. Some school leaders said they received no clear explanation before the state introduced the tool.
“My main concern is that my child’s biometric data has been collected for a year without my knowledge or consent, and we have no idea where this data is being stored and what is being done with it,” said Sally DeRadke, a parent, speaking to Education Week.
On August 27, the Albuquerque Journal reported that 35 lawmakers, including four Republicans, had signed a letter to Public Education Secretary Mariana Padilla. Led by state Representative Linda Serrato of Santa Fe, the lawmakers called for stronger parent outreach, consent requirements and safeguards for children’s voice data. The letter also warned that if children’s voices and other real-world data shape a model’s future output, its behavior and performance can change.
Districts found workarounds because waiting carried an immediate cost. Los Alamos Public Schools Superintendent Jennifer Guy asked the education department for guidance in a July letter, while Santa Fe and Los Alamos initially paused use of the requirement. Santa Fe said it would use i-Ready if the state gave no direction by August 14 because that program does not record voices. Las Cruces Public Schools offered an in-person assessment on request; Deputy Superintendent Wendi Miller-Tomlinson said it took about 30 minutes, compared with roughly 10 minutes for Amira. Parents also criticized the growing role of educational technology and screen time in early-grade classrooms, arguing that it detracted from the personal connection between teacher and pupil.
Serrato said stronger guidance for AI would be a priority when the legislature returned in January. That made the dispute broader than one vendor. New Mexico had required the tool before it set standards for every vendor competing for the statewide contract.
The dispute did not settle on test results either. Albuquerque Public Schools reported its first- and second-graders’ highest reading scores since at least 2021, with 34 percent of first-graders and 39 percent of second-graders proficient at the end of May. In Las Cruces, board member Bob Wofford still asked for evidence showing how Amira performed against other programs or teaching methods. Wendi Miller-Tomlinson said little comparative evidence was available because Amira was only in its second year and the first year had been affected by teachers and students learning the system.
A local outcome is not comparative evidence.
Guidance arrived after implementation had begun. The New Mexico Public Education Department issued guidelines requiring recordings to be deleted at least monthly. An Albuquerque Public Schools letter to parents said Amira had certified that voice data from before the current school year had been deleted, that testing data would be deleted within two days and that the tutoring component would score and analyze voices without recording them.
Amira says it keeps student data private and secure, does not create voiceprints or store identifying biometric information and never uses student data to train public or third-party AI models. Mark Angel, Amira’s CEO, said the company collects student data under the education department’s direction so the state can produce reading-performance reports. He also said the program uses student voices to train the system, although an assessment does not change during a school year. Those rules answer some questions about storage and access. They do not resolve the parents’ claim that the initial collection lacked knowledge or consent, and people addressing the Las Cruces board said they did not trust the assurances.
Portland should not be folded into New Mexico’s mandate. Portland Public Schools engaged Amira for a 2025-26 reading pilot, and parents there have objected to the purchase, but the available reporting establishes no school-board vote, petition count or final court ruling. The contrast matters because a vendor can face privacy objections through a district contract even when a state did not order the tool.
A 2026 tracker from Novo Innovative Pathways counted AI-in-education bills in 27 states; seven applied to every district, six through statute and one through a state board rule.
These rules create three mismatches: legislatures can require a tool without evaluation standards, states can issue privacy guidance after implementation begins, and districts can pay for paper alternatives while vendors wait for a replacement contract.
The next test in New Mexico is the planned request for proposals for a new K-2 reading assessment. District leader Wendi Miller-Tomlinson did not expect Amira to be selected again. For the incumbent, that bid would put privacy and evaluation evidence at the center of procurement. A reading tool can perform well and still lose the contract if it cannot show a parent, school board or state auditor what happens to a child’s voice. The next bid will make that proof—or its absence—visible across the field.
Competitors Consolidate Around Evidence
A June 24, 2026 Brief Glance account said Lexia already serves one in three U.S. school districts. That reach matters as ESSER funding ends and districts consolidate onto fewer contracts covering core instruction, assessment and related systems. Buyers increasingly pull login and engagement reports and ask whether students and teachers are using a product and whether it moved a metric they care about.
Lexia packages implementation as evidence
Lexia’s Science of Reading Classroom strategy links professional learning, instruction and reporting rather than selling them as separate purchases. LETRS Insights embeds guidance from the Language Essentials for Teachers of Reading and Spelling program inside Core5 Reading. Lexia Assignments automates targeted practice, while enhanced Skill Check reporting gives educators a clearer view of student progress.
That makes implementation part of the product. For districts, that means one evidence file must cover teacher support, product use and student performance. The August 2025 Core5 analysis reported that students who used Core5 more frequently achieved higher scores on the Smarter Balanced English Language Arts assessment. The strongest gains appeared among students whose teachers completed more LETRS training.
The finding links product use, teacher preparation and student performance, but it does not prove that Core5 caused the gains. It does give procurement teams a more useful case than a product demonstration: software use and staff implementation require one evidence record.
Vendor selection is becoming more evidence-driven, but the available record is product- and evidence-based rather than a head-to-head district switch. It does not document a shared-district Lexia and Imagine Learning decision with comparable cost and outcome data. It supports a narrower conclusion: software use and staff implementation belong in the same evidence file.
Imagine Learning competes by widening the bundle
Imagine Learning is pursuing consolidation through portfolio breadth. The company says it serves students in more than half of U.S. districts and offers core curriculum, courseware, supplemental programs, assessment and school services. For a district trying to reduce purchasing and implementation work, that range offers a plausible case for using one supplier for several instructional jobs.
Its April 16, 2026 product update pushed that strategy further. Dragonfly is a next-generation K-5 literacy program grounded in the science of reading and knowledge building. ClassMate puts Curriculum-Informed AI inside the Imagine IM classroom and aligns it with instructional intent and pacing. Imagine+ Math will launch this fall with practice, intervention and enrichment for grades K-8.
Lexia’s pitch is a tighter implementation story around literacy; Imagine’s is a wider catalog that puts AI, curriculum and school services under one roof. Breadth can simplify consolidation, but it also raises the burden of proof. Buyers still need evidence for each module they add to a contract. A large portfolio cannot substitute for program-level results.
Renaissance makes data continuity the pitch
Renaissance’s back-to-school materials present a portfolio spanning assessment, practice and instruction alongside a new Renaissance Intelligence system.
The available material does not explain what that system contains. It provides no components, adoption count, pricing or independent efficacy findings. Renaissance can therefore point to continuity across its product portfolio, but a district cannot use the material to compare its implementation with Lexia’s or Imagine’s. The missing details leave buyers without the evidence they need.
Khan Academy and Duolingo show scale without a district case
ValueAddVC reported that Khanmigo grew from 40,000 to 700,000 student users in one school year. The figures show audience and business scale, but the research gives no district adoption rate or independent outcome data for either service. It also does not establish either company as a provider of state-approved reading assessment, research-based teacher professional learning or a unified district literacy suite. On this record, their advantage is distribution rather than a documented case for replacing an integrated literacy platform. They compete for learner attention, not yet for the core district literacy contract.
Evidence becomes the entry requirement
Boston University’s EVAL Collaborative framed the procurement problem plainly: generative AI has entered classrooms faster than independent validation. The collaborative reported that fewer than 10 percent of AI tools used in classrooms have undergone independent validation. Amira will undergo an independent randomized controlled trial that the university will co-design, but the trial cannot supply the evidence a purchasing official needs today.
The field drew 32 applicants across literacy and math platforms, tutoring, assessment and classroom support. The research team is co-designing the evaluation with expertise in computing, AI, biostatistics, economics, special education and learning sciences.
The competitive field now offers implementation, portfolio breadth, data continuity or user reach. What remains scarce is comparable evidence. Under tighter budgets, an evidence package now does work that a product demo used to do.
The next procurement cycle should demand product-level evidence tied to curriculum, staff implementation, privacy terms and recurring cost. Vendors should fund independent studies before renewal season and publish failures as well as wins. A proposal built around an AI feature list is not ready for a general-fund vote.
At the next budget meeting, an AI reading tutor will arrive with more than a model card: a price, a funding source, a privacy schedule, an outcome record and a renewal date. If those records hold, a district can keep the tool. If they do not, the line item disappears—even if the reading app works.
Working in frontier tech? Zero G Talent tracks the openings: see every open Amira Learning role, browse frontier tech jobs, the companies hiring, and the people building the field.