Skip to content
math revision

Product roadmap · updated 24 Oct 2026

Prove calm practice teaches more, then charge, then add subjects

PSLE Maths for P3 to P6, sold to parents. We win by being the calm, honest option in a market built on pressure.

About this page

This is the working plan for Math Revision, an adaptive PSLE Maths practice service built to the EduAI SG brief (v4.5). It is written for the team, pilot partners and anyone advising us. It says what we are building, in what order, and why.

The overview below gives the answer first. Every section after it is the evidence: who our customers are and how they split into segments, Gibson Biddle's strategy model, an opportunity solution tree, the staged roadmap with MoSCoW priorities, RICE scores, a BELT check and the list of things we have ruled out. Each section opens with a short note on why we use that tool. The plan changes as the pilot teaches us things; the date at the top says when it last did.

Overview

Situation
Singapore parents already spend heavily on PSLE Maths: assessment books, tuition and apps, from P3 onwards.
Complication
Most apps compete on time spent, with streaks and rewards that cause fights at home. Parents still can't tell what is working, and we have no evidence yet that ours teaches more.
Question
What should we build, and in what order, to earn families who pay and stay?
Answer
Prove in an 8-week pilot that calm, personalised practice teaches more. Then charge. Then add subjects.

Governing thought

Earn the right to charge by proving the learning, not by growing time on app

Now

The MVP already does the core job without pressure

  • 25 features built: timed sessions, a 12-question check-in, adaptive question choice and spaced review.
  • Parents see progress with how sure we are, and planned against done time.
  • No streaks, points or nudges. Sessions end when they said they would.
See what is built
Next

The pilot must show it teaches more, safely

  • 60 children, 8 weeks: half get personalised questions, half a fixed order.
  • The verdict is a test on unseen questions two weeks after practice.
  • Late-night use, overruns and parent-reported conflict must stay low.
  • First: Singapore hosting with tested restores, rights records, named reviewers.
See the opportunity tree
Later

Only then do we charge and widen

  • SGD monthly subscription per child, after the pilot evidence.
  • Exam technique coaching for P5 and P6.
  • English, Chinese and Science, each behind its own content gate.
See the full roadmap

Ideal customer profiles

The parent buys; the child uses. Every segment describes both.

Our ICP in one sentence

A Singapore parent of a P3 to P6 child at a local primary school who already pays for maths help, sits with their child for homework on some evenings, and worries about PSLE Maths but is tired of the arguments it causes.

The buyer
The parent who runs homework in the house. They choose, pay, set the limits and read the weekly report.
The user
A 9 to 12-year-old who does not choose to practise maths. They need short sessions, quick wins and a clear end.
The household
Middle-income, often dual-income, in an HDB flat or condo. Weekday evenings are the only study time and they are already crowded.
What they already pay for
Assessment books, a tuition class or tutor, and often a practice app. Maths tuition is the anchor price we are compared with.
The moment they look for help
A disappointing result, the step up into P5, or a week where every homework session ends in tears.

Four segments, in priority order

Primary

The worn-out supporter

Pain
Willing to pay
Easy to reach
Fits a calm product

30 of 60 pilot places

 Secondary

The early planner

Pain
Willing to pay
Easy to reach
Fits a calm product

20 of 60 pilot places

 Secondary

The reluctant learner's parent

Pain
Willing to pay
Easy to reach
Fits a calm product

10 of 60 pilot places

 Not now

The stretch seeker

Pain
Willing to pay
Easy to reach
Fits a calm product

0 of 60 pilot places

Pilot recruitment mix

30
20
10

30 worn-out supporter, 20 early planner, 10 reluctant learner's parent. Weighted to the segment with the most pain and the easiest reach, with enough of the others to learn from.

Each segment in depth

Primary 30 pilot places

The worn-out supporter

"PSLE is next year and every maths session ends in a fight. I don't even know which topics are the problem."

Open

Who they are

  • Parent of a P5 or P6 child in the middle of the class for maths (roughly AL4 to AL6).
  • Already pays for a tuition class or tutor, plus assessment books.
  • Checks homework most nights and knows the routine is not working.

The child

Feels behind, avoids word problems, rushes to get it over with. Wants to stop being nagged and to feel they are getting somewhere.

What sends them looking

  • A weak end-of-year result in P4 or P5.
  • The jump in difficulty in P5: fractions, ratio, percentage and long word problems.
  • A tutor or teacher saying "they need more practice" without saying what.

Jobs to be done

  • Show me exactly which topics are costing marks, so practice goes where it counts.
  • Let someone else be the one who sets the questions, so I can stop being the enemy.
  • Tell me honestly whether it is working before I spend more.

Pains

  • Nightly arguments that leave parent and child drained.
  • Tuition is generic: the class moves on whether their child got it or not.
  • Assessment books mean marking, explaining and chasing, all by the parent.
  • Paying a lot without seeing what changed.

What they use now, and why they'd switch

Group tuition class:
Paced for the class, not the child. Parents get little feedback.
Private tutor:
Works, but costs several times our price and depends on one person.
Assessment books:
Cheap, but the parent does the marking and the nagging.
Practice apps with rewards:
Children chase coins and streaks, not understanding. Parents distrust the progress numbers.

What they value

A calm eveningA specific list of weak topicsEvidence they can trustLess parent effort

Willingness to pay

High. They already spend on tuition, so a monthly fee well below one tuition class is easy to justify if the report shows progress.

Where we reach them

  • KiasuParents forum threads on P5 and P6 maths.
  • School parent support groups and class chat groups, through pilot parents.
  • Referrals: "it stopped the fights" is the story they pass on.

Signs of a good fit

  • Says "we argue about maths" without being asked.
  • Already pays for maths help.
  • Wants a weekly summary more than a daily dashboard.

Red flags

  • Wants hours of drilling every night.
  • Wants to monitor the child minute by minute.

What we say

Short maths practice that ends on time, picks the topics your child needs, and shows you what changed each week.

Needs they bring (opportunity tree)

Risk

They are under the most time pressure, so they may judge us in two weeks. The first report has to show something specific.

Secondary 20 pilot places

The early planner

"She's doing fine now, but I've heard P5 is a big jump. I want the habit in place before it matters."

Open

Who they are

  • Parent of a P3 or P4 child who is doing reasonably well.
  • Often has an older child who found P5 and P6 hard.
  • Spends less today but plans to spend more later.

The child

Still fairly relaxed about maths. Likes seeing progress and doing things on their own. Can be put off for years by pressure that starts too early.

What sends them looking

  • An older sibling's PSLE year.
  • Friends' children starting tuition in P3.
  • The first school assessments that come with marks.

Jobs to be done

  • Build a steady, small habit now, so P5 and P6 are not a scramble.
  • Catch gaps in the basics before they snowball.
  • Keep maths from becoming something my child dreads.

Pains

  • Worry about falling behind peers who already have tuition.
  • Not wanting to start the pressure too young.
  • No clear view of whether the basics are really secure.

What they use now, and why they'd switch

Starting tuition early:
Expensive for a child who is coping, and it adds hours to the week.
Assessment books:
Fine for a while, but they become a chore for the parent.
Doing nothing yet:
The default. They will act if something easy and calm is offered.

What they value

A small routineSpotting gaps earlyA child who still likes maths

Willingness to pay

Medium now, but the longest lifetime: up to four years as a customer if we keep their trust.

Where we reach them

  • Parents of older children in our pilot (siblings).
  • Parent talks at schools and libraries on "starting PSLE prep without the stress".
  • Search for P3 and P4 maths topics, answered with free worked examples.

Signs of a good fit

  • Asks "how much is enough?" rather than "how much more?"
  • Happy with 15 or 20-minute sessions.

Red flags

  • Wants P1 or P2 content (not built yet).
  • Wants advanced or competition maths.

What we say

Fifteen minutes, a few times a week, so P5 doesn't come as a shock.

Needs they bring (opportunity tree)

Risk

Low urgency. They may sign up and drift. Spaced review and the weekly report have to make the habit easy to keep.

Secondary 10 pilot places

The reluctant learner's parent

"He just won't start. The moment I mention maths he shuts down."

Open

Who they are

  • Parent of a child at any level from P3 to P6 who avoids maths.
  • Has tried tuition, rewards, or taking away screen time, and it has not lasted.
  • Worried about the relationship as much as the grades.

The child

Has failed enough to expect to fail again. Hides it as boredom or anger. Needs a first win in minutes, a clear end, and no audience for mistakes.

What sends them looking

  • A run of refused or abandoned homework sessions.
  • A teacher raising concern about effort or confidence.
  • The parent noticing the child calls themselves "bad at maths".

Jobs to be done

  • Get my child to start, without me forcing it.
  • Give them a few wins so they believe they can do it.
  • Tell me what to do, and what to stop doing.

Pains

  • Every approach so far has ended in a standoff.
  • Apps with timers and leaderboards make it worse.
  • Guilt and worry about pushing too hard or not enough.

What they use now, and why they'd switch

One-to-one tutor:
Can work, but is expensive and the child may refuse that too.
Rewards and bribes:
Work for a week, then the price goes up.
Backing off:
Lowers the conflict but the gaps grow.

What they value

A first winNo pressurePractical advice for the parent

Willingness to pay

Medium. Very willing if it works, quick to cancel if the child refuses in week one.

Where we reach them

  • Pilot parents who see it work and tell friends in the same situation.
  • School counsellors and teachers, once we have pilot evidence.

Signs of a good fit

  • Open to starting with 7 to 10-minute sessions.
  • Wants the child to have a say in when and how long.

Red flags

  • Child needs specialist learning support. We refer, we don't replace it.
  • Parent wants us to enforce practice the child refuses.

What we say

A short first session your child can finish, with no timer, no score and no audience.

Needs they bring (opportunity tree)

Risk

Hardest to win, and the brief's Activation Engine is built for them. Brief 3.5 sets a stop rule: after two or three failed tries we stop and hand back to the parent.

Not now 0 pilot places

The stretch seeker

"She's already AL1. We want her ready for the hardest questions and maybe DSA."

Open

Who they are

  • Parent of a high-achieving P5 or P6 child.
  • Already pays for premium tuition or enrichment.
  • Measures success by top grades and selective school places.

The child

Confident, quick, can get bored by questions that are too easy.

What sends them looking

  • DSA applications.
  • Competition season.

Jobs to be done

  • Push my child to the hardest questions.
  • Prove they are ahead.

Pains

  • Generic practice is too easy.
  • Wants rankings against peers.

What they use now, and why they'd switch

Premium tuition and enrichment:
Already well served, at prices we will not compete with.

What they value

Hard problemsRankingsMore hours

Willingness to pay

High, but for things we have ruled out: rankings, more hours, competition maths.

Where we reach them

  • Not targeted.

Signs of a good fit

  • Welcome if they come, served by the same calm product.

Red flags

  • Wants leaderboards or daily hour targets.

What we say

Not marketed to.

Needs they bring (opportunity tree)

Risk

Chasing them would pull the product towards rankings and hours, which the brief bans and which would cost us the other three segments.

Not for us, for now

P1 and P2 parents
Not built yet. We offer to email them when it is, and never make up a plan for a level we have not written.
Parents who want maximum hours
Sessions end on time and there is a daily cap. We would disappoint them, and they would push us to break our rules.
Parents who want to watch every minute
We report weekly progress and effort, not live tracking. The brief rules out surveillance.
Schools, teachers and tuition centres
Parents are the only buyer for the first 12 months (brief 1.3).
International-curriculum schools
Our items follow the MOE primary maths syllabus, not IB or Cambridge.
Children who need specialist support
We can sit alongside it, but we are not a substitute. We point parents to the right help.

How we talk to them

Say

ends on timethe topics your child needswhat changed this weekno fightsshort sessionsmade for the MOE syllabus

✕ Don't say

AI-poweredaddictiveguaranteed AL1gamifiedunlimited practicebeat your classmates

Strategy

Gibson Biddle's DHM model: how we delight parents and children in hard-to-copy, margin-enhancing ways.

Delight
  • Sessions end when they said they would. No fights about "one more".
  • A real win inside the first 10 minutes.
  • Parents see evidence, with how sure we are, not a vanity score.

Why: parents keep what lowers stress at home. A calm evening is the reason they renew, more than any feature.

Hard to copy
  • An original, calibrated P3-P6 item bank, tagged to the syllabus.
  • Every answer stored as an event, so mastery data compounds.
  • Pilot evidence of learning gain against a fixed sequence.
  • A "no addiction" stance competitors can't adopt without losing their numbers.

Why: any app can add more questions in a month. Calibrated items, years of answer data and trial evidence take years to build, and they get better with every child.

Margin-enhancing
  • Deterministic marking: no LLM cost per answer.
  • We own the items, so no licence fees.
  • One monthly SGD subscription per child.
  • Parents change settings themselves, which keeps support light.

Why: serving one more child should cost almost nothing. That lets us charge far less than tuition and still pay for good content.

GEM order for the next two quarters

  1. 1Engagement. Do pilot children finish planned study and come back by choice?
  2. 2Growth. Do pilot parents tell other parents?
  3. 3Monetisation. Paid launch only after the pilot shows safe use and learning gain.

Strategies, proxy metrics, tactics

Finish without a fight

Proxy: 70% of planned weekly minutes completed

Tactics: Bounded blocks, break prompts, Activation Engine

Practise the right thing

Proxy: Adaptive arm +0.3 SD on the 2-week unseen test

Tactics: Cold-start check, BKT, recommender, spaced review

Show parents it works

Proxy: 80% rate the weekly report useful; 4 in 5 stay past month 3

Tactics: Dashboard evidence levels, weekly report

Fewer marks lost to exam technique

Proxy: Command-word errors halve between mini-mocks

Tactics: Exam coach, timed mini-mocks with debrief

Earn trust first

Proxy: Zero privacy incidents; 100% consent recorded

Tactics: MFA, export, deletion, audit, Singapore hosting

Opportunity solution tree

Outcome

More children finish their planned study each week and come back by choice

Measured as planned minutes completed and a voluntary second session within 7 days. Never session length or app opens.

Parent and child

"I don't know what my child should practise next."

Solutions

Test: Adaptive arm beats the fixed-sequence arm on the unseen-item test.

Parent

"I can't tell if it is working."

Solutions

Test: 8 in 10 pilot parents rate the weekly report useful.

Parent

"Can I trust this with my child's data and my money?"

Solutions

Test: Zero privacy incidents; every pilot family completes consent and MFA.

Roadmap

MoSCoW within each stage: Must ships or the stage doesn't end. Inside a priority, highest RICE first. Gate means the brief requires it before release, whatever it scores.

Now: MVP

Built Oct 2026

A safe, complete maths loop that a handful of invited families can use.

  • Must Gate Built RICE 200

    Immutable answer events and idempotent submit

  • Must Built RICE 160

    Short cold-start check (12 items, broad sampling)

  • Must Built RICE 140

    Spaced-review due dates

  • Must Built RICE 106.7

    BKT mastery with 5-attempt rolling score and gain

  • Must Built RICE 100

    Parent registration with consent record

  • Must Built RICE 90

    Bounded study block: question, mark, feedback, worked solution, reflection

  • Must Built RICE 90

    Momentum controls: 15/20/30/45 min blocks, break prompt, pause, daily cap, quiet hours

  • Must Built RICE 80

    P3-P6 curriculum graph and original item bank

  • Must Built RICE 80

    Parent dashboard: progress with evidence level, planned vs done time, settings

  • Must Built RICE 70

    Constrained greedy recommender with stored rationale

  • Must Gate Built RICE 9

    Admin: content review, suppression, policy versions, audit, rationale view

  • Should Built RICE 18

    Readability and slop checks in the question editor

  • Could Built RICE 33.3

    Learning Support Profile (minimal, with evidence level)

Next: closed pilot

Nov 2026 to Feb 2027 (brief Phase 1 exit)

60 learners, 30 per arm, 8 weeks. Prove the workflow, the data and the burden limits. Hosting, content rights and reviewer gates come first.

  • Must Gate Built RICE 200

    Pilot protocol frozen before enrolment; check-up items never seen

  • Must Built RICE 100

    Consent step after MFA, with terms and privacy versions

  • Must Built RICE 93.3

    Delayed unseen-item test two weeks after practice

  • Must Gate Built RICE 80

    Pilot arms: adaptive vs fixed sequence, assigned after baseline and stored immutably

  • Must Gate Built RICE 60

    Content rights record for every question, image and rubric

  • Must Gate Built RICE 50

    Email verification and parent MFA

  • Must Built RICE 46.7

    Weekly parent report: planned vs done, workload, pauses

  • Must Gate RICE 30

    Singapore hosting: multi-AZ, backups with a tested restore, 1,000-session load test

  • Must Gate RICE 20

    Item bank to 2,000 reviewed questions with provenance

  • Must Built RICE 12

    Parent data deletion (share revoke comes with sharing)

  • Must Gate Built RICE 12

    Named staff accounts with least-privilege roles; a different reviewer approves content

  • Should Built RICE 33.3

    Full Learning Support Profile

  • Should Built RICE 24

    Offline answer retry on reconnect

  • Should RICE 16

    Exam technique coach and timed mini-mocks with debrief

  • Should Built RICE 15

    Activation Engine: friction check and a 7-10 minute first win (built; off until Sponsor approval and a closed beta)

Later: paid launch

Mar to Jun 2027

Parents pay SGD monthly for one child. Hosted in Singapore.

  • Must RICE 26.7

    Billing: SGD subscription, one child per entitlement, 7-day trial

  • Should RICE 36

    Exam coach for all P5-P6 learners

  • Should RICE 16.7

    Calibrated difficulty (Rasch/Elo) and item exposure limits

Then: more subjects

Mid to late 2027 (brief Phase 2, months 8-12)

English, Chinese, Science and Bahasa one at a time, each behind its own release gate. AI-drafted content after 2,000 reviewed items.

  • Could Gate RICE 3.6

    English, Chinese, Science and Bahasa, one subject at a time

  • Could Gate RICE 2.5

    AI-drafted content proposals, human reviewed (after 2,000 reviewed items)

RICE ranking

Reach × Impact × Confidence ÷ Effort. Reach is pilot families out of 100 per quarter (internal tools use the team, 10). Impact: 3 massive, 2 high, 1 medium, 0.5 low. Effort in person-weeks.

  1. Immutable answer events and idempotent submit

    200

    Must MVP R 100 · I 2 · C 100% · E 1

  2. Pilot protocol frozen before enrolment; check-up items never seen

    200

    Must closed pilot R 100 · I 1 · C 100% · E 0.5

  3. Short cold-start check (12 items, broad sampling)

    160

    Must MVP R 100 · I 2 · C 80% · E 1

  4. Spaced-review due dates

    140

    Must MVP R 100 · I 2 · C 70% · E 1

  5. BKT mastery with 5-attempt rolling score and gain

    106.7

    Must MVP R 100 · I 2 · C 80% · E 1.5

  6. Parent registration with consent record

    100

    Must MVP R 100 · I 1 · C 100% · E 1

  7. Consent step after MFA, with terms and privacy versions

    100

    Must closed pilot R 100 · I 0.5 · C 100% · E 0.5

  8. Delayed unseen-item test two weeks after practice

    93.3

    Must closed pilot R 100 · I 2 · C 70% · E 1.5

  9. Bounded study block: question, mark, feedback, worked solution, reflection

    90

    Must MVP R 100 · I 3 · C 90% · E 3

  10. Momentum controls: 15/20/30/45 min blocks, break prompt, pause, daily cap, quiet hours

    90

    Must MVP R 100 · I 1 · C 90% · E 1

  11. P3-P6 curriculum graph and original item bank

    80

    Must MVP R 100 · I 3 · C 80% · E 3

  12. Parent dashboard: progress with evidence level, planned vs done time, settings

    80

    Must MVP R 100 · I 2 · C 80% · E 2

  13. Pilot arms: adaptive vs fixed sequence, assigned after baseline and stored immutably

    80

    Must closed pilot R 100 · I 2 · C 80% · E 2

  14. Constrained greedy recommender with stored rationale

    70

    Must MVP R 100 · I 2 · C 70% · E 2

  15. Content rights record for every question, image and rubric

    60

    Must closed pilot R 100 · I 1 · C 90% · E 1.5

  16. Email verification and parent MFA

    50

    Must closed pilot R 100 · I 0.5 · C 100% · E 1

  17. Weekly parent report: planned vs done, workload, pauses

    46.7

    Must closed pilot R 100 · I 1 · C 70% · E 1.5

  18. Exam coach for all P5-P6 learners

    36

    Should paid launch R 60 · I 2 · C 60% · E 2

  19. Full Learning Support Profile

    33.3

    Should closed pilot R 100 · I 1 · C 50% · E 1.5

  20. Learning Support Profile (minimal, with evidence level)

    33.3

    Could MVP R 100 · I 1 · C 50% · E 1.5

  21. Singapore hosting: multi-AZ, backups with a tested restore, 1,000-session load test

    30

    Must closed pilot R 100 · I 1 · C 90% · E 3

  22. Billing: SGD subscription, one child per entitlement, 7-day trial

    26.7

    Must paid launch R 100 · I 1 · C 80% · E 3

  23. Offline answer retry on reconnect

    24

    Should closed pilot R 30 · I 1 · C 80% · E 1

  24. Item bank to 2,000 reviewed questions with provenance

    20

    Must closed pilot R 100 · I 3 · C 80% · E 12

  25. Readability and slop checks in the question editor

    18

    Should MVP R 10 · I 1 · C 90% · E 0.5

  26. Calibrated difficulty (Rasch/Elo) and item exposure limits

    16.7

    Should paid launch R 100 · I 1 · C 50% · E 3

  27. Exam technique coach and timed mini-mocks with debrief

    16

    Should closed pilot R 40 · I 2 · C 60% · E 3

  28. Activation Engine: friction check and a 7-10 minute first win (built; off until Sponsor approval and a closed beta)

    15

    Should closed pilot R 30 · I 2 · C 50% · E 2

  29. Parent data deletion (share revoke comes with sharing)

    12

    Must closed pilot R 20 · I 1 · C 90% · E 1.5

  30. Named staff accounts with least-privilege roles; a different reviewer approves content

    12

    Must closed pilot R 10 · I 2 · C 90% · E 1.5

  31. Admin: content review, suppression, policy versions, audit, rationale view

    9

    Must MVP R 10 · I 3 · C 90% · E 3

  32. English, Chinese, Science and Bahasa, one subject at a time

    3.6

    Could more subjects R 60 · I 3 · C 40% · E 20

  33. AI-drafted content proposals, human reviewed (after 2,000 reviewed items)

    2.5

    Could more subjects R 10 · I 2 · C 50% · E 4

BELT check

Behaviour

Singapore parents already buy assessment books and tuition, and already sit beside their child for homework. We replace the assessment book. We don't invent a new ritual.

Enduring problem

PSLE comes every year, every topic builds on the last, and children forget. The need for the right next question never goes away from P3 to P6.

Lock-ins

The child's mastery history, a schedule built around the family's week, and the weekly report parents rely on. Trust, not streaks.

Transient distractions

Points and leaderboards, AI chat-tutor hype, more subjects before maths is proven, school portals, and any nudge that grows time on app.

Won't do in the next 12 months

Streaks, points, random rewards, endless feeds, urgency timers
Banned by brief 3.2. They grow time on app, not learning, and they cost parent trust.
An LLM in the answer path
Marking and feedback stay deterministic and human-written (brief 10.1). Cheaper, auditable, no surprises.
Teacher and school portals
Parents are the only buyer for the first 12 months (brief 1.3).
O and A level maths
Brief Phase 3 (months 12-18), and only once the first four subjects show good learning, burden and content results.
Family discounts
Deferred until retention and support data justify them (brief 17.1).
Re-engagement nudges based on inactivity
Brief 3.5 forbids child-targeted escalation. Missed sessions create no debt.