Data hub Product cost B2B usage Teacher usage

Top users overview

Who uses FLYER most, and who drives AI cost. Last 12 months, from the production read replica. Generated 29/08/2026.

Teachers assigning work
2.188
of 21.280 accounts — 10%
Students active
285.399
of 1.687.013 profiles — 17%
AI grading calls
325.529
speaking + writing + IELTS
AI chat sessions
601.765
Bingo conversation
All three rankings concentrate the same way. Roughly a fifth of each group accounts for 70–83% of everything. That means a short call list reaches most of the usage, and most of the cost. It also means the averages you see elsewhere describe almost nobody.

The 20/80 picture

RankingPopulationTop 20% They account for 
Teachers by FLYER usage (study hours generated) 1.18423681%
Teachers by AI usage 4869783%
Students by AI usage 64.66312.93270%

Teacher AI usage is the most concentrated of the three. 97 teachers drive 37.246 of 45.136 AI grading calls. If AI spend needs controlling, that is the entire list.

1. Teachers using FLYER the most

Ranked by study hours their assigned work generated. This measures what students actually did with the assignments, not how many were handed out. Sample assignments excluded.

#TeacherCentreStudy hours SubmissionsStudentsHours / student
1Loan ĐỗLoan Đỗ13.88538.02834540,2
2Mrs NgọcMrs Ngọc5.10236.95416530,9
3Hải Yến NguyễnAmity English3.98762.12029713,4
4MinhMinh3.1115.69110429,9
5Pham Hong VyHoàng Gia3.0018.61512723,6
6Nguyễn Thị Thúy HằngMoonLink2.90416.9574886,0
7yen leyen le2.62622.83513519,5
8Ms ThuyMs Thuy1.93932.2602856,8
9Như Anh DangNhư Anh Dang1.83332.5722806,5
10Hương PhạmHương Phạm1.77346.4172696,6
11Nguyễn Trần Anh ThưNguyễn Trần Anh Thư1.6646.6928519,6
12Anh Ngữ LisaAnh Ngữ Lisa1.44327.0328117,8
Loan Đỗ is an outlier, not a leader by a nose. 13.885 hours is 2,7× the second place and 40 hours per student per year — roughly one hour a week, every week. Worth understanding how that class runs before assuming it can be copied.

Note how weakly submissions predict hours. Hải Yến Nguyễn handed out 62.120 submissions, the most on this list, yet generated less than a third of Loan Đỗ's hours. Counting assignments handed out tells you about the teacher's habit; counting hours tells you whether students engaged.

2. Teachers driving the most AI usage

AI grading calls triggered by their assignments. This is the list that maps to AI spend.

#TeacherCentreAI calls SubmissionsStudentsAI per student
1Pham Hong VyHoàng Gia4.4038.61512734,7
2Nguyễn Ngọc TiếnNguyễn Ngọc Tiến2.1395.1565024,3
3Hương PhạmHương Phạm1.20746.4172694,5
4Hồ Văn LinhHồ Văn Linh1.1823.6911866,4
5Nguyễn Thị Cẩm HồngUK Academy Bà Rịa1.0243.5241208,5
6Phuong Lam DangPhuong Lam Dang1.0164.6325119,9
7Nguyễn Trần Anh ThưNguyễn Trần Anh Thư8716.6928510,2
8Trang LuTrang Lu7595.5586511,7
9Hường FlyerHường Flyer74725.6471814,1
10Anh Văn Cô Liên Thạc SĩAnh Văn Cô Liên Thạc Sĩ72924.3383362,2
11Mrs. Patricia ReyesSkyHigh Flyers6814.1466410,6
12Ann CPiSchool Cẩm Phả6084.0071185,2
One teacher accounts for 10% of all AI grading. Pham Hong Vy triggers 4.403 calls from just 127 students — 34,7 per student, eight times the rate of the teachers below her. She is also the reason Hoàng Gia ranks second among centres.

The contrast with Hương Phạm is the useful part: same tier of overall usage, 46.417 submissions against 8.615, yet a quarter of the AI calls. Two teachers working equally hard, one leaning on AI and one not. Understanding that difference is what makes AI spend predictable.

3. Students using AI the most

AI grading calls plus Bingo chat sessions, per student. Internal and test accounts removed.

The raw ranking is unusable without filtering. Before cleaning, the top spot was the FLYER BINGO internal account with 3.760 events, followed by consecutive IDs 1453127 to 1453135 with near identical counts — test accounts. The table below excludes them. Names are withheld: these are children, and this page has no login.
#Student IDAI gradedChat sessionsTotalPattern
1325217266612878mostly chat
26379354949598mostly graded work
34215915477554graded work only
4153759347490537chat only
5299188137392529mostly chat
6310358139497mixed
78225211480491chat only
8141768842929458mostly graded work

Two distinct behaviours sit side by side here, and they cost different amounts. Chat-heavy students consume Inworld conversation minutes; grading-heavy students consume Azure and Speechace calls. Any per-student cost model needs to treat them separately.

What to do with this

To grow usage: the top 20% is already saturated. The 948 teachers outside it generate 19% of hours between them. Moving even a fraction of that group toward the habits of the top list moves the total far more than pushing the top list higher.

To control AI cost: 97 teachers is a manageable list. Start with the ones whose AI-per-student ratio is far above their peers, since that ratio, not class size, is what drives spend.

To learn what works: Loan Đỗ on hours and Pham Hong Vy on AI are both extreme outliers within their own tier. Two conversations would explain more than another month of dashboards.
Source: production read replica, 12 months to 29/08/2026.
Teacher usage attributed through assign_tests.teacher_profile_id; system-generated sample assignments excluded. AI grading counted from request_ai_review and request_ai_azure; chat sessions from ai_conversation_sessions. Student list filters out internal and test accounts.
Internal page. Contains teacher names and centre names — do not share outside the company.