Data-Informed Teaching: How to Spot the Learning Story Your Marks Miss

You mark thirty scripts on a Sunday evening. You total them, enter them into the markbook, and the class average climbs four points. It feels like good news. Then on Monday, the same three students who have been quiet all term sit just as quiet, and something in you says the number did not tell the whole truth.

That instinct is worth trusting. It is the starting point of data-informed teaching: the habit of treating every mark as a clue about learning, never as a final verdict on it. A score tells you what happened. It rarely tells you why. And the why is where your next lesson lives.

This article draws on a Suraasa webinar led by Manish Pandey, Teacher Career Mentor and Associate Director at Suraasa. Pandey makes a clear case: as AI takes over the job of supplying answers, a teacher's real professional value moves to diagnosis. Reading the learner. Spotting the pattern. Choosing the precise next step. Below, you will find his core ideas, a practical framework you can use this week, and worked examples that show how to separate a deep misunderstanding from a simple language gap.

Watch Manish Pandey walk through the full classroom case study and the Evidence-to-Action Loop in his own words, so you can see the reasoning behind each step.

The Shifting Role of the Educator in the Age of AI

For most of the history of schooling, the teacher held the knowledge and the student came to receive it. That arrangement is ending. A Grade 8 student can now ask a chatbot to explain photosynthesis, solve a quadratic equation, or summarise a chapter of history in under ten seconds.

Manish Pandey argues that this changes what teachers are for. In his framing, content delivery is depreciating, and diagnostic judgment is becoming the premium skill. Facts and answers will reach students from many directions. What they cannot get from a search bar is someone who understands their particular path through a subject.

"For teachers it will not be about knowing answers. For teachers it will be about knowing the learner, knowing the human being." (Manish Pandey)

Read that carefully, because it does not diminish teaching. It raises the bar. Anyone can tell a student that 3/4 is larger than 2/3. It takes a teacher to notice that a student keeps comparing numerators only, to understand why, and to know which representation will finally make the idea click for that child.

This is not a new insight in education research. The OECD, through its long-running work on the teaching profession, has consistently described teaching as knowledge-rich professional work rather than simple content transmission. What AI changes is the urgency. When the content layer becomes free and instant, the diagnostic layer becomes the whole job.

If you want a wider view of how these tools are reshaping classrooms, our guide to AI in education and what teachers should do about it sets out the bigger picture. For now, hold on to one idea: your value lives in what you notice about learners, and data is only useful if it sharpens that noticing.

Why Markbook Averages Deceive: Lessons from Classroom Case Studies

Averages feel objective. They compress thirty students into a single tidy number you can report upward. That compression is exactly the problem.

Pandey illustrates this with a comparison between two classes. Class A raised its average score by 15 points. Class B raised its average by 10 points. On paper, Class A looks like the clear success story, and most review meetings would treat it that way.

Then you look underneath the number.

In Class A, five high-improving students pulled the average up while 25 of their classmates made almost no gain. In Class B, the improvement spread across nearly the entire room. If your goal is learning for every student, Class B tells the stronger story, even though its headline number is smaller.

"The figures were correct, but the conclusion that was there was incorrect or incomplete." (Manish Pandey)

That sentence captures the whole trap. Nobody made an arithmetic error. The teacher in Class A did not fake anything. The problem lay in the conclusion people drew from a correct figure.

Here is why this matters in your own markbook. An average hides at least three different realities:

• A few soaring, most static. The Class A pattern. A small group races ahead and masks a stalled majority.
• Steady, broad gains. The Class B pattern. Smaller headline growth, but genuinely shared progress.
• A split class. One group improves while another goes backwards, and the average shows almost no change at all.

Three classrooms with three very different needs can produce the same average. If you plan your next unit from that average alone, you are planning for a class that does not exist.

Try this: before your next data review, sort your students into three columns based on change since the last assessment: improved clearly, stayed roughly the same, dropped. Count each column. That simple spread will tell you more than the average ever will, and it takes ten minutes.

Data-Driven vs Data-Informed: Looking Through the Windshield, Not the Rear-View Mirror

The two phrases sound alike. They describe very different ways of working.

Data-driven practice lets the number make the decision. The score dropped, so we reteach. The average rose, so we move on. The data sits in the driver's seat.

Data-informed practice keeps you in the driver's seat. The data is evidence you weigh alongside what you saw in the room, what you know about each student, and what the task actually demanded. You decide. The data advises.

Pandey uses a driving image that sticks. Formative classroom evidence should work like a car's windshield, not its rear-view mirror. A rear-view mirror shows you where you have been. That has its uses, but nobody drives forward by staring into it. The windshield shows you the road ahead and the turn you need to make next.

In practice, this means every piece of classroom evidence should answer one forward-looking question: what do I need to adjust in the very next lesson? If a quiz result cannot change what you do tomorrow, it is a record, not a tool.

This aligns closely with the distinction between assessment for learning and assessment of learning. End-of-term exams look backward and certify. Exit tickets, hinge questions, and quick checks look forward and steer. Our guide to assessment for learning vs assessment of learning explains how to balance both.

The research base supports this forward focus. The Education Endowment Foundation in England, which reviews evidence on classroom approaches, consistently highlights well-delivered feedback as one of the more powerful levers available to teachers. Feedback only works when it changes what happens next, for the student and for the teacher. A windshield, in other words.

The Evidence-to-Action Loop: A Four-Step Framework for the Classroom

Knowing you should act on evidence is easy. Knowing how is harder, especially at the end of a long teaching day. Pandey offers a simple structure he calls the Evidence-to-Action Loop. It runs on four questions, asked in order:

1. What did learners show? Describe the evidence plainly. What did students write, say, draw, or do? No interpretation yet.
2. What pattern might it suggest? Look across responses. Which errors repeat? Which students cluster together? Hold this as a hypothesis, not a fact.
3. What will I adjust? Choose one focused change to your teaching. Not a whole new unit. One precise move.
4. What will I check next? Decide how you will find out whether the adjustment worked. This check becomes the evidence for the next turn of the loop.

Notice the careful language in step two: might suggest. The loop builds humility into the process. Your first reading of a pattern can be wrong, and step four exists to catch that.

A worked example

Here is how the loop might run in an illustrative Grade 6 science lesson on states of matter. The teacher ends class with a two-question exit ticket. Question one asks students to explain why a puddle disappears on a hot day.

This entire loop fits on a sticky note. It takes perhaps fifteen minutes after class. And it produces something a class average never could: a plan aimed at specific students for specific reasons.

If you need quick ways to gather evidence for step one, our list of 25 classroom assessment techniques to check understanding every day gives you options that take under five minutes.

Diagnosing Learning Barriers: Conceptual Misconceptions vs Explanation Gaps

Step two of the loop asks you to find a pattern. The most useful pattern you can find is the type of barrier behind a wrong answer. Two students can get the same question wrong for completely different reasons, and they need completely different help.

Pandey uses a fraction task to make this concrete. Some students chose 1/8 over 1/5 as the larger fraction. The classic thinking behind that choice is familiar to every maths teacher: 8 is bigger than 5, so 1/8 must be bigger than 1/5. That is a conceptual misconception. The student has built a rule that feels logical but is wrong. Pandey's response for this group is visual modelling, so students can see that splitting a whole into eight parts makes each part smaller.

A second group chose the correct answer. On a simple marking scheme they look fine. But when asked to explain why, they wrote nothing or could not put their reasoning into words. Pandey identifies this as a language barrier, an explanation gap. The understanding may well be there. The words are not. These students need sentence starters, not another lesson on fractions.

Treat these two groups the same way and you fail both. Give the misconception group sentence starters and they will fluently explain a wrong idea. Give the explanation-gap group more fraction strips and they will be bored by something they already grasp.

The practical takeaways from the webinar add a third category worth checking: procedural fluency. Some students understand the concept and can explain it but make errors under time pressure because the steps are not yet automatic.

The explanation-gap category deserves special attention in multilingual classrooms. A student learning in English as a second or third language may understand photosynthesis perfectly in their home language and still produce a thin written answer. If you read that answer as a conceptual failure, you underestimate the student and reteach the wrong thing. Diagnosis protects students from being mislabelled.

Try this on your next test: pick the one question with the most wrong answers. Sort the responses into the three columns above. You will often find that a single "problem question" holds two or three different problems inside it.

The Pitfalls of Blanket Reteaching and How to Target Micro-Interventions

When test scores drop, the reflex in many schools is to reteach the whole topic to the whole class. It feels responsible. It looks thorough. Pandey argues it is often inefficient and counterproductive.

His reasoning has two parts. First, full-class reteaching bores the students who already understood. They sit through the same explanation again, learn nothing new, and start to disengage. Second, it fails to target the distinct causes behind the errors. If one group holds a misconception and another has an explanation gap, the same repeated lesson cannot fix both. It may fix neither.

Reteaching in a louder voice is still the same lesson. Students who did not get it the first time often need a different route in, not the same route repeated.

The alternative is the micro-intervention: a short, precise response aimed at one barrier for one group. Here is how a single follow-up lesson on the fraction task might look when you organise it by diagnosis rather than by topic:

• Misconception group (around 8 students): ten minutes at a table with fraction strips or paper folding. Fold one strip into fifths and another into eighths. Compare the size of one part. Let the physical evidence break the rule.
• Explanation-gap group (around 6 students): work with a sentence frame such as "1/5 is larger than 1/8 because when a whole is divided into more parts, each part is..." Pairs rehearse orally, then write.
• Secure group (the rest): an extension task. Order five fractions with different denominators, or write a short explanation a younger student could understand.

Everyone works on something they need. No one sits through a lesson pitched at someone else. And because each group has a clear target, your follow-up check in step four of the loop becomes simple to design.

Micro-interventions do not require elaborate planning. Most take one focused activity and a few minutes of your time with a small group while others work independently. If you want to structure the independent work well, our guide to cooperative learning strategies that actually produce learning can help you keep the secure group productive while you work with others.

Using AI to Surface Student Work Patterns Without Losing Teacher Agency

Sorting thirty or forty written responses into categories is the most time-consuming part of diagnosis. This is where Pandey sees a genuine role for AI, with firm limits.

His position is precise. AI should act as an evidence analysis assistant, not a decision-maker. It can group anonymised student responses and count how often each pattern appears. It should never label students, infer anything about their intelligence, or choose your teaching interventions.

"AI will organize possibilities for you. As a teacher you will interpret, decide and check." (Manish Pandey)

That division of labour maps neatly onto the Evidence-to-Action Loop. AI can speed up step one and offer candidates for step two. Steps three and four stay with you.

What a guarded prompt looks like

Here is an illustrative prompt built on the principles Pandey describes. Replace student names with numbers before you paste anything.

Task: Below are 30 anonymised student responses to the question "Which is larger, 1/5 or 1/8? Explain why." Each response begins with a student number.

Do: Group the responses by the type of reasoning shown. For each group, give the student numbers, a count, and one short example quote. Note any responses that give a correct answer with no explanation as a separate group.

Do not: Label or rank any student. Do not comment on ability or intelligence. Do not suggest teaching strategies or interventions. Only describe patterns in the responses.

The "do not" section is the heart of it. Without those guardrails, many AI tools will cheerfully describe students as "struggling learners" or recommend a lesson plan. Both cross the line Pandey draws. A label generated by a machine can follow a child for years. A teaching choice made by a machine ignores everything you know about the room.

Once you have the groups, your judgment takes over. Does the AI's grouping match what you saw in class? Did it lump together two different errors? Did it miss the student who wrote a correct answer in their home language? You check, you correct, you decide.

Data privacy matters here too. Anonymise every response, follow your school's policy on AI tools, and never paste personal details. UNESCO's guidance on artificial intelligence in education places a human-centred approach and data protection at the core of responsible classroom use, which sits comfortably alongside Pandey's guardrails.

For more practical prompt structures you can adapt across planning and assessment, see our guide to ChatGPT for teachers, and for building tasks that generate better diagnostic evidence in the first place, our article on the AI assignment generator.

From Informal Noticing to Demonstrating Verifiable Learning Impact

Good teachers have always noticed things. The puzzled look. The hand that half rises and drops. The student who finishes too quickly. The difference between an experienced teacher and a growing one is rarely whether they notice. It is what they do next, and whether they can show it worked.

Pandey describes professional growth across three stages:

1. Notice. You pick up learning cues informally in the room. You sense something is off, even if you cannot yet name it.
2. Respond. You adapt your teaching based on evidence. You move from a hunch to a deliberate adjustment, like the micro-interventions above.
3. Demonstrate. You articulate and document the measurable impact of your adaptation. You can show a school leader or appraisal panel what changed, for whom, and how you know.

Many teachers live in stages one and two for years. They do excellent responsive work, and almost none of it becomes visible. When appraisal time arrives, they describe their teaching in general terms because they never captured the specifics.

Stage three changes that. The webinar's practical takeaways recommend structured before-and-after statements. Here is an illustrative template:

Before: On the fractions exit ticket, [number] of [total] students chose the smaller fraction as larger, and [number] gave correct answers without any explanation.

What I changed: I ran a ten-minute fraction-strip activity for the misconception group and gave the explanation group a sentence frame.

After: On the follow-up check, [number] of the misconception group answered correctly, and [number] of the explanation group wrote a complete reason.

That is three sentences. It names the evidence, the action, and the result. It is honest about scale. And it is far more persuasive to a head of department than "I differentiate regularly for my learners."

These statements also become the raw material for your professional development record. If you are planning how to log and evidence your growth across a year, our guide to CPD for teachers and how to plan your year shows how to build that habit, and our piece on bridging the gap between professional development and real classroom impact explores how reflection turns into visible results.

One caution: demonstrating impact does not mean manufacturing perfect numbers. Sometimes the follow-up check shows your adjustment did not work. Record that too. A teacher who writes "the visual model helped five of eight students, so I am trying a different representation with the other three" shows more professional maturity than one who only reports wins.

Developing Teacher Judgment as an Irreplaceable Professional Competency

Everything in this article comes back to a single capability: judgment. The ability to look at evidence, weigh it against what you know about your students, and make a sound decision about what to do next.

"Data-informed teaching is not about collecting more information. It is about making better decisions from the right information." (Manish Pandey)

This line is worth pinning above your desk, because the opposite instinct runs deep in many schools. More tests. More trackers. More spreadsheets. More colour-coded dashboards. None of it helps if nobody asks what the information means for tomorrow's lesson. A single well-designed exit ticket, read carefully and acted on, beats a term of unread data.

Judgment is not a fixed trait. You build it, like any professional skill, through deliberate practice. Here is what that practice looks like week to week:

• Run the loop once a week. Choose one class and one piece of evidence. Ask the four questions. Write the answers down.
• Sort one question by barrier type. Misconception, explanation gap, or fluency. Over time, you will start seeing these categories instantly as you mark.
• Look at spread, not just averages. Every time you see a class average, ask how many students moved and how many did not.
• Use AI for counting, not for deciding. Let it tally patterns. Keep every interpretation and intervention in your own hands.
• Write one before-and-after statement a month. By the end of the year, you will have a portfolio of real, specific impact.

None of this requires new software, a bigger budget, or a change in school policy. It requires a shift in how you read what is already in front of you.

There is a larger stake here too. When teachers practise this kind of judgment, they make the strongest possible case for the profession itself. A machine can generate a worksheet. It cannot sit beside a child who believes 1/8 is bigger than 1/5, understand exactly why that belief makes sense to them, and choose the one activity that will change their mind. That is teaching. It always was.

If you are thinking about how to build this kind of evidence-based practice into a longer professional journey, our comparison of the best teaching certifications for career growth can help you see which paths develop classroom judgment rather than simply certify attendance.

Your Next Step

You do not need to overhaul your practice to start. Pick one class this week. Give a two-question exit ticket. Before you total the marks, sort the responses: who understood, who holds a misconception, who knows it but cannot yet say it. Then plan one small move for each group.

That is data-informed teaching in its simplest form. Marks as clues. Patterns as hypotheses. Your judgment at the centre. And a check that tells you whether it worked.

The numbers in your markbook are true. The story behind them is yours to read.

For the Love of Teaching.