P3M.AI, Observation Intelligence
Watching a
lesson is easy.
Changing one is not.
IMPACT Observations turns a classroom visit into a rated, evidenced record — then turns the weak parameters into assigned tasks with owners and dates, and tracks whether anything actually moved by the next visit.
Most observations
end in a drawer.
Schools already observe. The visit happens, a form is filled, a conversation occurs, and then the next term begins with nothing carried forward.
A rating without a reason
A 3 out of 4 tells a teacher nothing they can act on. Without the expectation statement and the evidence behind it, the number is a verdict rather than a description.
No line to the next visit
The same weakness appears in three consecutive observations because nothing between them was assigned, owned or checked. Observation becomes an annual ritual.
Counting is done badly
Nobody in the room can accurately track talk share, wait time or who was actually asked a question, while also watching thirty-six children. Those are the numbers that matter most.
Two ways to observe
A person in the room.
Or a camera at the back.
Most schools use both. The observer judges what only a professional can judge. The video audit counts what no professional can count while teaching attention is elsewhere.
The observer decides
A structured form on phone or laptop. Each criterion is rated by selecting the expectation statement that matches what was seen, with a comment and evidence attached. Works offline in a classroom with weak signal and syncs later.
- 14 criteria across 6 parameters
- Expectation-mapped ratings, not bare numbers
- Comment required on any rating below expectation
- Photo, document and audio evidence attached inline
- Closing questions the rubric cannot capture
The recording measures
A single fixed camera at the back of the room. The analysis returns a minute-by-minute timeline, talk share, wait time per question, seat-level question distribution and aggregate on-task rate — with a confidence score on every finding.
- Lesson timeline classified minute by minute
- Flagged moments with timecodes and confidence
- Seat positions only — no facial identification
- Machine ratings stay provisional until an observer confirms
- Divergence from the observer is logged, not hidden
Counting. Talk ratio, wait time, question spread, transition length, on-task trend — measured exactly, every lesson, without observer fatigue.
Judgement. Whether an explanation was well pitched, whether a question was worth asking, whether a child was quietly lost. A camera cannot see understanding.
A machine rating never enters a teacher's record unconfirmed. Where the two disagree, the observer's rating stands and the gap is logged to improve the model.
Four observation types.
One shared rubric.
The same 14 criteria are used by every type, so a self-rating, a peer note and an external audit can sit on one chart and be compared honestly. The differences are who observes and what the result is used for.
External
Conducted by a P3M.AI auditor or an empanelled external expert. Used for benchmarking, accreditation evidence and an unfamiliar pair of eyes on established habits.
Internal
Run by the academic coordinator, head of department or principal. The backbone of the cycle — this is where support decisions and development plans are grounded.
Peer
One teacher observes another, both directions, with cover provided. The cheapest intervention a school has, and the one teachers request most often in feedback forms.
Self
The teacher rates their own lesson on the same rubric, before seeing anyone else's rating. The gap between self and observer is often the most useful thing on the profile.
Why the self-rating matters. A teacher who rates themselves 3.6 where the observer sees 2.4 needs a different conversation from one who rates themselves 2.2 where the observer sees 3.0. The first is a calibration problem; the second is a confidence problem. Both are invisible without the self view.
Inside the form
Context in.
Better output
out.
A rating on its own is thin data. What makes the generated report specific is everything attached to it — the comment explaining the rating, the photograph of the board, the lesson plan, the sample of student work. The engine writes from the evidence, so the more context an observer captures in the room, the less generic the output.
Comments carry the reasoning
Required on any rating below expectation. This is what a teacher reads first and what protects the rating if it is later contested.
Evidence carries the proof
Board photographs, lesson plans, student work samples, short audio clips, worksheets. Attached to the specific criterion, not dumped at the end.
Closing questions carry the rest
Was the objective met? Did anything happen the rubric cannot capture? Is immediate support needed? Free text, and often the most valuable field in the form.
One audit,
or a whole year
of them.
Some schools want a single external read before an accreditation visit. Others want a running record for every teacher. Both run on the same instrument, so a school can start with the first and move to the second without redoing anything.
One-time audit
A defined window — a week, a department, a grade band. External observers, a full set of individual reports, and one consolidated findings document for the leadership team.
- Scoped and priced per engagement
- Individual report per teacher observed
- Consolidated school findings and priorities
- No platform commitment required
Active tracking
A live dashboard across the year. Every observation, of every type, lands on the same teacher profile — so the third visit can open with what the second one asked for.
- Live dashboard for coordinators and heads
- Individual profile per teacher, across years
- Self, peer, internal and external on one chart
- Tasks, due dates and completion tracked throughout
From finding to follow-through
A low score
should create
a task.
This is the part most observation systems are missing. When a parameter falls below its threshold, IMPACT proposes a task tied to that specific criterion — with an owner, a cadence and a due date — and carries it into the next observation as something to re-check.
Threshold breached
A criterion is rated below expectation, or a parameter mean falls under the school's set threshold. The trigger is a rule the school configures, not a black box.
Task recommended
The engine proposes actions drawn from that criterion's playbook, shaped by the observer's comment and evidence. The coordinator accepts, edits or dismisses each one.
Owned and scheduled
Every accepted task gets an owner, a cadence and a due date. Some sit with the teacher, some with the coordinator, some with the department or the school.
Re-checked, not repeated
The next observation opens with the open tasks and re-checks only the criteria they address. Progress is measured against the specific thing that was asked for.
Recommended
Print the class list, tick names as asked, review weekly.
Trigger · P4 ≤ 2.0Observe Ms. Iyer, cover arranged by coordinator.
Requested in feedbackIn progress
Six minutes of written response before discussion, every lesson.
Trigger · P5 ≤ 2.5Done
Confirmed at the 29 Jul observation. Rated Exemplary.
Closed · from Mar auditRules the school sets
Thresholds, who owns which task type, and how long a task may stay open before it escalates. A school that observes lightly and one that observes intensively need different rules, so they are configuration rather than product.
Recommendations, not instructions
Every proposed task is reviewed by a human before it is assigned. The engine is good at noticing that a criterion has fallen three times; it is not the right thing to decide what a specific teacher needs.
Closed means verified
A task closes when the next observation confirms the change, not when someone ticks a box. That single rule is the difference between a development record and a compliance log.
What comes
back out.
Four report types, all built on the same rubric, all written in plain language for the person receiving them.
Teacher report
Per observation. Ratings with evidence, strengths, development areas, tasks and sign-off.
Video audit report
Timeline, talk metrics, question distribution, flagged timecodes and stated limitations.
Department rollup
Parameter means by department, weakest criterion, and where practice varies within a team.
School annual
Year-wide teaching profile alongside student outcomes, with priorities by owner.
Trust is the product
An observation system
teachers don't trust
produces compliance.
Every audited teacher receives a feedback form on the process itself. If a rating is contested with evidence, it can be revised — and the revision is recorded, not quietly overwritten.
Auditee rates fairness, observer conduct and whether feedback was usable.
Any rating can be challenged with evidence. Revisions are logged with reasons.
Post-observation conversation target, tracked as a process measure.
Video audits run with written consent, seat-level anonymity and scheduled deletion.
What schools
ask first.
Is this going to be used to appraise or rank our teachers?
Not by design. Observations are written as development records — the teacher's own feedback is attached to the report, contested ratings are visible, and tasks close on verified change rather than on a manager's sign-off. If a school chooses to use the data in appraisal, that is a school policy decision and we would ask you to state it to your staff before the first cycle rather than after.
Do we need cameras in every classroom?
No. Most schools run manual observations only, and add video audit for a small pilot group or for specific development cases. A single portable camera moved between rooms covers a full department. Video audit requires written consent and carries a deletion schedule.
Who can see a teacher's observation history?
The teacher, their observer, and named leadership roles. Peers see only the observation they conducted. There is no school-wide leaderboard, and department rollups report parameter means rather than named individuals.
What if a teacher disagrees with a rating?
They contest it through the feedback form with their evidence. The observer either revises the rating or explains in writing why it stands, and both appear in the final report. In our reference school, 11 of 96 ratings were contested last year and four were revised — which is the process working, not failing.
How long does one observation take?
The lesson itself plus roughly ten minutes on the form, most of which is the observer's comments. The report is generated within a day. The post-observation conversation is the part that matters and should happen within 48 hours, which the platform tracks.
Can we run a one-time audit before committing?
Yes, and it is the usual starting point. A single window — one department or one grade band — produces individual reports and a consolidated findings document with no platform commitment. If you move to active tracking afterwards, that first audit becomes the baseline rather than being discarded.
Start with
one department.
One window, real reports, your own rubric. Then decide whether it deserves a year.
A 30-minute call with your academic head. We bring a sample observation report built on your rubric.