Skip to content

What ControlAI can do

ControlAI is not a single feature but a whole set of capabilities built around the lesson. This is an overview catalog: what the system sees, measures, and shows. Grouped by meaning.

Some of it already works in the pilot, some is on the roadmap (marked separately, and gathered in the «Roadmap» chapter). What each number actually means is in the «Metrics and what they mean» chapter. Which role sees what is in the «Inside the admin panel» chapter.

1. Lesson speech analysis

  • Talk-time balance — how much time the teacher spoke versus the students.
  • Questions — how many questions the teacher asked.
  • Pauses and silence — how much of the lesson passed in silence and what the longest pause was; whether the lesson "sagged."
  • Target-language ratio — what share of the lesson ran in, say, English (or another target language) and what share ran in the native or another language.
  • Punctuality — whether the lesson started and ended on time (how many minutes later or earlier).
  • How many people spoke — an estimate of the number of distinct voices in the lesson.
  • (roadmap) interaction pace (how many "teacher↔student" exchanges per 10 minutes), distribution of speech among students ("3 students spoke 90% of the time"), the share of native and target language over the course of the lesson, classroom audio-quality monitoring (catches a "dying" microphone before report quality drops), as well as rude or inappropriate language, praise and encouragement of students, addressing students by name, and digressions to off-topic subjects.

2. Lesson video analysis

  • Attendance — how many students were in the classroom (and how many were expected).
  • Late arrivals — how many students arrived after the lesson started and by how much.
  • Empty classroom — the system sees when a lesson did not actually take place.
  • (roadmap) who is distracted by their phone, who is active and who is passive the whole lesson, head on the desk (signs of sleep, fatigue, or boredom in the lesson), and how often students raise their hand.

3. "Lesson DNA"

A visual colored lesson timeline: how time was distributed — where the teacher spoke, where the students did, where there were questions and discussions, where there was group work, and where there was silence. One glance and the "shape" of the lesson is visible: did the teacher monologue the whole lesson or engage the students. Alongside it the system marks the key moments of the lesson (the good ones and the problematic ones) tied to timestamps — for example, "14:18 — 5 minutes of pair work" or "14:27 — 6 minutes of silence."

4. Method compliance ("AI-nazoratchi")

This is the "quality controller" inside ControlAI — it checks the lesson against the standards of your specific center. This is the deepest part of the product.

How it is configured: the center uploads its own rubric (the lesson observation form), its own methodology, and its textbooks/curriculum. From then on ControlAI checks every lesson against exactly these materials — the criteria are not "hard-wired" into the system, they are yours.

What is checked (an example of a typical rubric — about 30 criteria across several blocks; your center will have its own):

  • Lesson organization — seating and layout, classroom discipline, contact with students, starting and ending on time.
  • Lesson structure — whether the lesson went through its stages (warm-up → presentation → practice → independent practice) and whether each stage got adequate time. (some of the stage-level checks — roadmap)
  • Teaching technique — whether the teacher draws out answers instead of "telling it himself" (eliciting), whether he asks concept-checking questions (CCQ), whether he gives instructions clearly and checks that the task was understood (ICQ), whether he monitors students' work, how he gives feedback and corrects mistakes, whether he pitches the language to the group's level, and whether he overuses the native language.

What is produced for each lesson:

  • A filled-in rubric — for each criterion a score, a short evidence quote from the lesson, and a note on how confident the AI is; disputable points it honestly flags as "needs a human eye."
  • An overall methodology score for the lesson (1–10) — not a "bare number," but with a breakdown by criteria and a confidence level.
  • Topic coverage — what of the plan has already been covered, what partially, and what not yet (cumulative per topic: "affirmative forms covered, questions — not yet").
  • Review by the academic manager (methodist) — the center chooses the mode: mandatory human review before publishing, or auto-publishing; either way, the methodist's corrections make the system more accurate.

Disputable and subjective points the AI flags as low-confidence — so that an incorrect assessment of a teacher's work is never presented as fact.

5. Teacher analytics

From the lesson indicators a profile of each teacher is built:

  • Rating — an overall assessment of their work (how it is calculated — in the «How the teacher rating is calculated» chapter);
  • What the rating is made of — a breakdown by factors (talk-time balance, questions, language, punctuality) with clear explanations;
  • Skills radar — strengths and weaknesses across several axes;
  • Key lesson moments — the good and the problematic episodes tied to timestamps;
  • 8-week trend — whether the teacher is improving or slipping;
  • A "Needs attention" feed — the system itself surfaces what matters: a sharp drop in rating, too much silence, a camera problem, a switch to the native language, an outstanding result.

6. Coaching and growth

  • Coaching baskets — the system itself sorts teachers into "recognize," "improve," and "observe" and suggests a concrete action.
  • Goals — the teacher or manager sets a goal (for example, "bring the English share up to 60%"), and progress is calculated automatically.
  • Assessments/reviews — the manager records a review of the teacher's work; the corrections also accumulate and improve the system.

7. Student churn risk

ControlAI shows in advance which groups are at risk of "falling apart": it computes the risk level per group (0–100), tracks attendance and engagement trends, names the likely cause, and recommends an action — so you can intervene before the students leave.

8. Branch comparison

For networks: comparing branches against each other on key indicators (talk-time balance, language, questions, punctuality, retention), a branch ranking, and the best and the lagging ones.

9. Reports and digests

  • Lesson report — metrics plus a short, clear text about how the lesson went.
  • Daily digest — all the lessons of the day in a single list, problems at the top.
  • Personal summary for the teacher — their own statistics and a brief summary of the lesson (in a "help" format, not "judgment from above").
  • History of all lessons — any past lesson can be found by filters (group, teacher, classroom, date) and its result opened.
  • (roadmap) a weekly digest with trends by teacher and group; a lightweight report for parents (attendance + a brief summary).

10. Management and operations

Around the analytics — everything for the center to actually run:

  • Schedule — a weekly grid of lessons (who, when, which group, which classroom); it is exactly what tells the system when to pull recordings. ControlAI warns about conflicts (one teacher or classroom booked twice).
  • Courses and levels — a catalog of courses with levels (for example, on the CEFR scale: A1–C2).
  • People and groups — teachers, groups, students; invitations; consent status.
  • Classrooms and equipment — camera state (ready / needs a microphone / offline), connection check.
  • Billing — hour balance, packages, daily consumption; the bill is computed from the actually processed lesson minutes, and a lesson that did not take place is not billed.
  • Payroll — calculation based on lessons actually delivered (not on the plan) plus a rating bonus; an export for accounting.
  • Operations — a task list (camera offline, no audio, processing failure, lesson did not take place, awaiting consent) and a coverage summary: how many of the scheduled lessons were successfully processed on time.

11. Slices and comparisons

A separate section for analyzing indicators: you can pick a course, a metric, and a period and view a slice — for example, compare teachers by their number of questions over a month — and export the result.

The main takeaway from this chapter: ControlAI analyzes the speech, video, and methodology of a lesson (against the center's own rubric), builds teacher analytics, coaching, and a churn forecast out of it, prepares reports and digests, and around all of that provides everything for running the center — schedule, courses, people, equipment, billing, payroll, and quality control. Some of the advanced signals are still on the roadmap — they are marked.

→ Next: 05. Metrics and what they mean — we will break down each number in plain words and show what a good lesson looks like.