Metrics and what they mean

In the chapter «What it can do» we listed what ControlAI measures. Here is what each number means, how to tell whether it is good or not, and why it matters.

Important from the start. The numbers are a signal, not a verdict. «Good» values depend on the type of lesson (speaking, grammar, children), the level of the group, and the lesson format. The benchmarks below are a starting point, not a rigid standard; each center tunes them for itself. More on this at the end of the chapter.

Speech: what can be heard in the lesson

Talk-time balance (students / teacher)

• What it is. What share of the time students spoke versus the teacher.
• Benchmark. For a conversational lesson it is good when students speak 45–55% of the time. If the teacher takes up more than 70%, that is a reason to review the recording: most likely there was a monologue.
• Why it matters. A language (and not only a language) is learned when the student speaks themselves, rather than listening. A skew toward the teacher is a common cause of weak progress.
• Context. During a lecture or an explanation of new grammar, the teacher naturally speaks more — and that is normal.

Teacher questions

• What it is. How many questions the teacher asked the class during the lesson.
• Benchmark. About 15–20 questions per ninety-minute lesson is a healthy level of engagement. Fewer than ~10 means the lesson most likely ran in a «told them and left» mode.
• Why it matters. Questions engage students, check understanding, and hold attention.

Silence and pauses

• What it is. What share of the lesson passed in silence and what the longest pause was.
• Benchmark. Silence up to ~15% is the norm. Short pauses are useful (students think), but long gaps (for example, 6 minutes in a row) are a signal that the lesson «sagged».
• Why it matters. A lot of silence is lost paid time and, often, confusion or poor organization.

Target-language ratio

• What it is. What part of the lesson ran in the target language (for example, English), and what part ran in the native language.
• Benchmark. For a language center, the higher the target-language ratio, the better the «immersion»; a typical quality bar is 70% and up. But for beginner levels the native language is naturally more present.
• Why it matters. Language immersion is one of the main reasons parents pay for courses.

How many people spoke

• What it is. An estimate of the number of distinct voices heard during the lesson.
• Why it matters. An indirect check of engagement: if almost only the teacher was heard throughout the lesson, few students spoke — even if the time «balance» came out reasonable.

Punctuality

• What it is. How well the lesson started and ended on time relative to the schedule.
• Benchmark. It is good when more than 90% of lessons start on time.
• Why it matters. Late starts and early endings are both disrespectful to students and paid-for but unworked time.

Video: who was in the classroom

Attendance

• What it is. How many students were in the classroom and how many were expected.
• Why it matters. Group fill rate is directly tied to money and to retention: a drop in attendance is an early sign that a group is «falling apart».

Latecomers

• What it is. How many students arrived after the lesson started and by how many minutes.
• Why it matters. Regular late arrivals are a signal about discipline or that something is wrong with the group/schedule.

Summary scores

On top of these metrics there are two «top-level» scores — each has its own chapter:

• Teacher rating — an overall assessment of the teacher's work, built from the metrics of their lessons; how exactly and why it is fair is in the chapter «How the teacher rating is calculated».
• Method-compliance lesson score (1–10) — how well the lesson matched the center's rubric; it is computed not from the metrics above but from checking the lesson against the rubric — see the «Method compliance» section in the chapter «What it can do».

A picture of a good lesson

In summary — what a healthy conversational lesson looks like by the numbers (benchmarks, not a standard):

Metric	Good benchmark	Warning signal
Student talk-time	45–55%	teacher > 70%
Teacher questions	15–20 per lesson	fewer than ~10
Silence	up to ~15%	long gaps, > 15%
Target-language ratio	high (≈ 70% and up)	drops sharply to native
Punctuality	on time	lessons regularly starting late
Attendance	stable / full	falls lesson after lesson

The main point: the numbers are a signal, not a verdict

Three rules for reading metrics:

1. Context decides. A lecture-style lesson, a conversation club, and a class with young children look different by the numbers. The same talk-time balance can be excellent in one case and poor in another.
2. The trend matters more than a single lesson. Everyone has one «off» lesson now and then. What is worth looking at is the dynamics — three weak lessons in a row mean more than one.
3. A number shows where to look — a person decides. ControlAI does not pass a verdict. It highlights where attention is warranted; the conclusion is drawn by the academic manager (methodist) or the owner, opening the recording or transcript if needed.

---

Key takeaway from this chapter: each metric is a clear signal about how the lesson went, with its own «good / warning» benchmarks. But they must be read with regard to context and over time: ControlAI shows where to look, while the decision always remains with a person.

→ Next: 06. How the teacher rating is calculated — what makes up the overall score and why it is fair, not a «black box».