Ed.ai raises €5 million to help teachers move every student forward

Ed.ai has raised €5 million from Bpifrance, La Poste Ventures and 50 Partners. The Lyon-based startup is not simply seeking to reduce the time teachers spend grading: its platform analyses handwritten papers, identifies individual and collective errors, then suggests remediation activities that teachers can edit. Having grown, according to the company, from processing 100 papers a week to 2,500 a day, it must now demonstrate that this loop can improve student support before deploying it at scale in France and the United States.

A student paper enters Ed.ai as a photograph or scanned document. It emerges with a suggested grade, detailed comments, an assessment of the skills involved and exercises designed to address the errors identified. In between, the teacher adjusts the marking scheme, reviews the analysis and decides what will be shared with the student.

The Lyon-based startup readily presents this process as a way to reduce the time spent grading, but above all as a means of extending the educational life of each paper. The objective is no longer simply to identify an error, but to determine what it reveals, connect it to a specific skill and prepare the material that needs to be revisited.

The €5 million round announced on 10 September will be used to extend this model to more subjects, institutions and countries. Bpifrance invested through its Digital Venture fund, alongside La Poste Ventures — a fund launched by La Poste Group and managed by XAnge — and 50 Partners. Existing investors AFI Ventures, CentraleSupélec Venture, Ring Capital and Super Capital also reinvested.

The seed round comes seventeen months after an initial €1.7 million funding round, announced in April 2025, and brings the company’s total equity funding to €6.7 million.

Grading is only the first layer

To begin grading, the teacher submits an assessment to Ed.ai. The platform analyses the questions, extracts the criteria and skills being tested, then generates a suggested marking scheme. The teacher can modify the instructions, proficiency levels and allocation of points before submitting the student papers.

This first step is less incidental than it may appear. An AI system can only assess an answer correctly if it understands precisely what the teacher intended to measure. Two lines of reasoning can lead to the same result; an incorrect answer can contain a partially valid method; and an essay can demonstrate the expected knowledge while failing to address the question.

Ed.ai then reads the handwritten papers, compares each answer with the approved criteria and proposes a grade. According to the company’s product presentation, the platform can quote passages from the paper, categorise errors and associate results with the relevant skills. In its US mathematics product, the company also claims it can follow the steps in a student’s reasoning, award partial credit and recognise equations, graphs and geometric constructions.

The technical challenge lies precisely here. Recognising an equation is not the same as understanding a line of reasoning. The system must distinguish between a sign error, an unsuitable method, a poorly understood concept and a simple transcription mistake. It must also handle crossed-out work, annotations and the sometimes circuitous routes by which a student arrives at an answer.

Ed.ai does not publicly detail the model architecture it uses to connect recognition, interpretation and comment generation. The startup nevertheless claims a 95% recognition rate for handwritten mathematics.

The teacher retains the final decision

Ed.ai does not send a grade directly to the student. It prepares a first-pass assessment that the teacher must review, amend where necessary and approve. The teacher can adjust the points awarded, the comments and the skills assessment before any results are returned.

It is therefore more accurate to describe the product as AI-assisted first-pass grading rather than automated grading. The AI absorbs the volume, applies an initial consistent reading and formulates suggestions, while the teacher retains responsibility for the assessment.

This human oversight is one of Ed.ai’s central arguments. It allows the platform to be presented as a tool that augments professional judgement rather than a system designed to replace it. It also reflects a less spectacular reality: in education, a plausible answer is not enough. A grade must be explainable, open to challenge and correctable.

The burden of this review remains to be measured. A platform can generate large quantities of comments in seconds while merely shifting the workload instead of reducing it. If teachers have to revise a substantial share of the grades or rewrite most of the feedback, the time saving shrinks accordingly. Conversely, a high approval rate without modification would provide much stronger evidence of product maturity than the number of papers analysed alone.

Ed.ai currently says that its platform halves grading time and saves teachers four hours each week.

A grade becomes a map of student errors

Once the assessment has been approved, Ed.ai aggregates the results. The platform does not simply show that a student scored 11 out of 20. It seeks to identify the concepts they have mastered, their recurring errors and the stages of reasoning where difficulties arose. At class level, it can group students facing similar problems and highlight the areas the teacher needs to revisit.

This is where the product changes in nature: a stack of papers becomes a map of learning.

The distinction matters because a grade summarises more than it explains. Two students can achieve the same result for very different reasons: one may understand the method but repeatedly make calculation errors, while the other reaches the right number without understanding the principle involved. The appropriate teaching response will not be the same.

This granularity forms Ed.ai’s second value proposition. The first is faster grading. The next is to reveal what the ordinary pace of the classroom can leave hidden: individual misunderstandings concealed behind an acceptable average, errors shared by a group, or concepts that need to be revisited before the class moves on.

The company says that 30% of its team are former teachers and that it develops its product with teacher-authors in each country. This structure reflects the backgrounds of its three co-founders. Jonathan Banon co-founded LeLivreScolaire.fr; Cédric Bignon previously held engineering positions at Microsoft; and Rémi Mazières taught modern literature before working in training and support at Le Choix de l’école. Together, they bring experience in educational publishing, software infrastructure and the classroom.

Remediation is the real value proposition

A diagnosis is only useful if it leads to action. Based on the errors identified, Ed.ai generates personalised remediation activities that the teacher can edit, print or distribute digitally. A student who lacks a prerequisite concept therefore does not necessarily receive the same work as one who understands the method but struggles to apply it.

The platform is attempting to solve one of the oldest paradoxes in differentiated teaching. Adapting work to each student’s needs is educationally desirable, but preparing several sets of exercises, monitoring the results and repeating the process requires precisely the time teachers lack.

Ed.ai wants to reduce this preparation cost: the paper provides the diagnosis, the product generates an initial response and the teacher adjusts it. The expected gain therefore lies not only in the time removed from grading, but in making possible work that would often remain out of reach without automation.

This third layer could make the platform durable. First-pass grading attracts users because it addresses an immediate pain point. The map of acquired skills gives them a view of the class. Remediation is intended to bring them back after each assessment and encourage them to retain a record throughout the school year.

Paper becomes a digital point of entry

The focus on handwritten work gives Ed.ai a distinctive position in an industry that has often confused educational innovation with adding another screen. Students can continue working on paper. Teachers retain their own assessment materials and routines. Digitisation takes place after the assessment, at the point when the workload is concentrated on the teacher.

This continuity reduces adoption friction. It also allows the startup to work with largely unstructured material: the reasoning, hesitation and mistakes that digital platforms struggle to capture when students still produce most of their work by hand.

Student papers cannot, however, be treated as a freely exploitable pool of data. They contain personal information, assessments and sometimes details revealing a minor’s learning difficulties. Ed.ai’s US privacy policy states that student work and data are not used to train either its own models or those of its providers.

Usage validates the entry point, but not yet the impact

In September 2025, the platform was processing around 100 papers a week. By June 2026, that figure had reached 2,500 a day. The startup says it processed more than 100,000 papers over the past six months. In September 2025, according to the company, this measured volume covered French and mathematics; its French product now spans ten subjects across middle and high school. The number of institutions in which the platform is deployed is said to have risen from 120 to 500 in a year.

These figures show that teachers are willing to introduce Ed.ai into their grading process.

Ed.ai prices its institutional product from €6 per student, per year and per subject, with decreasing rates when several subjects are purchased. This model gives the company room to expand within each institution by covering more students, adding subjects and then renewing the licence.

Student progress will be the next test

Ed.ai plans to use part of the funding to conduct studies on student progress and teacher wellbeing. This stage is essential because the value proposition comprises four separate levels that need to be measured independently.

The first is technical: can the platform read student papers correctly? The second concerns workload: does it genuinely save time once the review process is included? The third is educational: does its diagnosis correctly identify the underlying difficulties? The fourth concerns outcomes: do the activities it suggests help students perform better in a subsequent assessment?

The figures released today mainly document usage and satisfaction. They do not yet demonstrate that students learn more or that teachers consistently recover the hours the company claims. This gap is not unusual for such a young product. It becomes significant, however, once the company presents itself not merely as a productivity tool but as a means of individualising teaching.

European regulation will add its own evidentiary requirements. AI systems used to assess learning outcomes or steer the learning process may fall within the high-risk category under the AI Act. The exact classification will depend on the platform’s effective role and the extent to which the teacher retains the decision. From 2 December 2027, the systems concerned will notably be subject to requirements covering data governance, documentation, logging, accuracy and human oversight.

Integrations are now part of the product

An EdTech company does not establish itself in a school simply because its interface works. It must also overcome authentication, data protection, budgeting, public procurement, training and the software already in place. In education, distribution often has as many layers as the product itself.

Ed.ai is available through France’s Gestionnaire d’accès aux ressources, or GAR, as well as through several digital learning environments. The product also integrates with ÉcoleDirecte and Pronote and appears in the catalogues of digital education distributors such as LDE and eMLS. Its website outlines different regional purchasing routes, from Île-de-France and Auvergne-Rhône-Alpes to Brittany, Occitanie and Hauts-de-France.

This presence reduces the number of steps required to move from an individual trial to an institutional deployment. It also explains the strategic relevance of La Poste Ventures in the new round. Following the transaction, Nadia Amal, deputy director of Docaposte’s Education and Youth division, is joining Ed.ai’s board. Docaposte controls Index Éducation, the company behind Pronote.

This does not mean Ed.ai will automatically be distributed to every Pronote user. No exclusive agreement of this kind has been announced.

The United States will test whether the pedagogy travels

Ed.ai began its US expansion with mathematics for middle and high schools. In spring 2026, its product was adapted to the standards of five states. The startup now says it has aligned its content with the curricula of all 50 states and integrated with more than fifteen platforms, including Canvas, Google Classroom, Clever and Schoology.

The startup also plans to add further subjects and launch pilots in higher education. Two trials have been announced with the University of Maryland and Francis Marion University in South Carolina.

The grading market is already occupied

Gradescope, acquired by Turnitin in 2018, claims more than 700 million questions graded, 140,000 instructors and 2,600 universities. Its platform covers paper-based exams, digital assignments and code. Its AI can notably group similar answers, allowing the instructor to apply the same rubric item to several papers.

CoGrader focuses more heavily on essays and open-ended answers in US primary and secondary education. The company says it is used by more than 100,000 teachers across 16,000 schools, with grading frameworks aligned with state standards and examinations. Like Ed.ai, it offers integrations with Google Classroom, Canvas and Schoology, as well as class-level trend analysis.

Graide, now part of Norwegian assessment specialist Inspera, combines grading, feedback and examination workflow management, primarily in higher education. Pensieve has focused on handwritten work in university-level science subjects. Its researchers report that the platform has processed more than 300,000 answers across around twenty institutions and reduced grading time by an average of 65% for predictions with the highest confidence levels.

Ed.ai is therefore not entering an empty market. Its positioning nevertheless combines several choices that are rarely brought together: beginning with paper, covering both scientific and humanities subjects, keeping teachers in control of validation and extending grading into the generation of differentiated activities.

Its advantage cannot rest solely on the use of generative AI, which is now available to all its competitors. It will have to be built across the quality of the entire chain: understanding what the teacher is asking, interpreting what the student has produced, explaining the gap and then suggesting appropriate work.

€5 million to industrialise a loop

The funding will allow Ed.ai to double its workforce, strengthen its teams in France and the United States, add subjects, prepare its entry into higher education and conduct impact studies. The company aims to process two million papers during the 2026-2027 school year, followed by one million a month within two years.

The main risk lies less in the ambition of each project than in their accumulation. Developing multiple subjects in France requires content and teacher-authors. Selling to US school districts demands a commercial team, integrations and local compliance. Entering higher education changes the buyer profile and brings the company up against established competitors. Measuring educational impact, finally, requires time, protocols and independent partners.

Ed.ai has already found a particularly tangible entry point into teachers’ daily work. The real test will therefore not be whether an artificial intelligence can read two million student papers, but whether, after reading them, it gives each teacher a better understanding of what to do next with the class.