Is an adaptive practice app a real alternative to maths tuition?
By The Sophoz Curriculum Team
The short answer. On measured learning outcomes, closer than most parents expect. In the most careful review of the question, human tutoring produced an effect of d = 0.79 and step-based computer tutoring systems d = 0.76, a difference that was not reliably different from zero. But that comparison holds only for good systems against ordinary tutoring on the specific thing both were measured on, which is subject learning. It says nothing about the things tuition is often actually bought for: supervision, motivation, and somebody noticing when a child has stopped trying.
Most writing on this question is either an app selling against tutors or a tuition centre selling against apps. This one is published by an adaptive learning company, which you should factor in, so every claim below is sourced and the places where tuition wins are stated as plainly as the places where it does not.
What the evidence actually says
The headline comparison
Kurt VanLehn's 2011 review in Educational Psychologist compared human tutoring, intelligent tutoring systems and other formats against classroom instruction with no tutoring. The results:
| Format | Effect size vs no tutoring |
|---|---|
| Human tutoring | d = 0.79 |
| Step-based intelligent tutoring systems | d = 0.76 |
| Human tutoring vs tutoring systems | d = 0.21, not reliably different |
The key term is step-based. A system that marks a final answer right or wrong is not what was measured. What was measured is a system that works through a problem step by step, notices where the reasoning went wrong, and responds at that step. That distinction does a great deal of work in this comparison, and most apps on the market are not step-based.
The number you should stop believing
Benjamin Bloom's 1984 claim that one-to-one tutoring produces a two standard deviation improvement is the most-quoted statistic in education technology marketing. It does not hold up.
VanLehn's conclusion, directly: "it seems that human tutoring is not usually 2 sigmas more effective as classroom instruction, as the six studies presented by Bloom (1984) invited us to believe", with the true figure "closer to the mean effect size found here, 0.79".
Two specific problems with the original. Bloom's tutors were undergraduate education majors with minimal training, not the expert tutors the statistic is usually taken to describe. And the tutored group was held to a 90 % mastery threshold against 80 % for the comparison group, which inflated the apparent gap.
This cuts against both sides of the argument. Tuition is not worth two sigma. Neither is any app claiming to deliver "the benefits of one-to-one tutoring", because the benefit being invoked was never that large.
Where the evidence runs out
Three honest limits.
Nobody has run this comparison on Indian board syllabi. The tutoring research is largely American and European, mostly on algebra, physics and programming. Whether it transfers to a class 7 CBSE syllabus with an Indian home tutor is an assumption, not a finding.
No product in the Indian market publishes an audited outcome comparison. Every improvement percentage in this category, including the 28 %-plus that Sophoz reports, is self-reported and uncontrolled. That does not make them false. It means they are marketing, and they should be weighted as marketing.
Behaviour inside software is easier to change than learning. A tutoring system explicitly designed to teach children better help-seeking produced durable behavioural improvement persisting months after the intervention ended, and no improvement in domain learning (Aleven, Roll, McLaren and Koedinger, 2016). Apps are good at changing what happens in the app.
What each one is genuinely better at
Where tuition wins
Noticing what the answers do not show. A tutor sees a child who has stopped trying, who is copying, who is anxious, who had a bad week. A system sees inputs. For a child whose problem is not mathematical, this is the whole game, and no amount of adaptation reaches it.
Accountability that a person creates. A tutor arriving at five o'clock on Tuesday produces work that an app on a phone does not. This is not a learning mechanism, it is a commitment device, and for many families it is the actual product being purchased.
Explaining in a second, third and fourth way, in response to a face. Good tutors read confusion in real time and change register. Systems do this within a fixed repertoire.
Handling the emotional side. This matters more than it sounds. Maths anxiety consumes the working memory a child needs to do mathematics, and the effect is strongest on hard multistep problems under consequential conditions. In one study of 154 young children, anxiety predicted lower achievement only among the children with higher working memory (Ramirez et al., 2013). The capable children are the ones pressure costs most, and a person is better placed to notice and defuse that than a progress bar.
Younger children. For grades 2 to 5, the evidence on problem-solving before instruction turns negative: instruction-first outperformed problem-solving-first for that age group in the meta-analysis of productive failure (Sinha and Kapur, 2021). Younger children generally need more instruction and more presence, not more autonomy.
Where adaptive practice wins
Diagnosis of prerequisites. This is the strongest case and the most specific. When a child keeps failing at equivalent fractions, the cause is often that multiplication as scaling is not secure. A system holding a map of skills and prerequisites checks underneath automatically, every time, across every topic. A tutor does this when they are good and have time, and a tutor working through a syllabus under time pressure often does not.
Volume and immediacy. A tutor is two hours a week. Practice happens on the other five days or it does not happen. Immediate feedback on a Thursday evening is worth more than corrected work on Saturday.
Spacing and mixing, done automatically. These are the two practice structures with the strongest evidence, and both are tedious to run by hand. Interleaved practice produced 61 % on a delayed test against 38 % for blocked practice in a randomised trial with 787 seventh-graders (Rohrer et al., 2020). Spacing improved final performance in 259 of 271 comparisons across 317 experiments (Cepeda et al., 2006). Most tuition, like most workbooks, runs topic by topic, which is blocked practice by construction.
Support that reduces as the child improves. Detailed guidance that helps a beginner becomes redundant and then harmful as expertise grows, because processing it costs attention (Kalyuga et al., 2003). Adjusting continuously is something software does without being asked.
Cost, usually by a wide margin. More on this below.
The cost comparison, honestly
We can only publish figures we verified, so here is what is checkable.
| Option | Published cost |
|---|---|
| Vedantu, grade 5 olympiad course | ₹12,222 for 45 live sessions, about ₹270 per group session |
| Olympiad Success live classes | ₹3,906 for 15-25 sessions, up to ₹26,250 for 120 |
| Cuemath | ₹610-₹800 per class on the annual plan; ₹74,880-₹98,176 a year all-in with 18 % GST |
| MTG workbook set | ₹562 for a seven-book class combination |
| Adaptive practice platforms | Mostly unpublished, including Sophoz's |
Local home tuition rates vary far too widely by city and tutor for us to publish a number we have not verified. Work out your own: divide what you pay a month by the number of hours delivered, then compare against ₹270 per session for group live teaching, which is the cheapest verified live-teaching figure we found.
Two points about that table. Cuemath is the most expensive option in it and is general mathematics tutoring rather than an exam-specific product, which is worth knowing before comparing it to anything exam-specific. And the fact that most adaptive platforms including ours do not publish pricing is a genuine deficiency; it means a parent cannot run the comparison this section is about without making enquiries.
The comparison that actually matters
The question "app or tutor" is the wrong shape. The useful question is what specifically is going wrong, because the answer determines the tool.
| If the problem is | The right tool is |
|---|---|
| Nobody knows why the same error keeps recurring | Adaptive diagnosis. This is what it is for |
| The child does not practise unless somebody is there | A tutor, or a parent. This is a supervision problem |
| The child understands in the lesson and cannot do it alone | Practice volume with immediate feedback, which is the app case |
| The child has decided they are bad at maths | A person. No app addresses this |
| Some topics are two years behind, others are ahead | Adaptive, because any single worksheet is wrong for them twice over |
| The child is in class 3 or 4 and needs teaching, not testing | A person, or a parent working through it alongside |
| Board exams are close and the syllabus is not covered | Structured teaching, tutor or otherwise. Diagnosis is not the bottleneck |
| Money is tight | Workbooks at around ₹562 plus a parent with an index card, which is genuinely defensible |
The honest answer for most CBSE, ICSE and state-board families is not either. It is that tuition is often bought to solve a supervision problem and then blamed for not solving a diagnosis problem, or bought to solve a diagnosis problem when the tutor is in fact re-teaching the chapter in the same order the school did.
Before changing anything, work out which of the rows above describes your household. That question is free and it is more useful than either product.
The thing neither one supplies
Whatever you buy, the practice structure decides most of the outcome, and both tutors and apps routinely get it wrong in the same way.
Blocked practice, meaning one topic at a time, produced 89 % accuracy during practice and 20 % a week later in one study, while mixed practice produced 60 % during and 63 % after. The method that looks better while the child is doing it is the one that fails on the paper.
A ₹98,000 course delivered chapter by chapter is worse, on this evidence, than a ₹562 workbook whose questions someone shuffles across chapters for ten minutes a week. That is not an argument for buying nothing. It is an argument for checking, whichever you buy, whether practice is mixed and spaced, because that is the variable with the largest measured effect and it is the one nobody sells.
Frequently asked questions
Can an app replace maths tuition? On measured subject learning, good step-based tutoring systems perform close to human tutoring: d = 0.76 against d = 0.79, with no reliable difference (VanLehn, 2011). But tuition is frequently bought for supervision, accountability and emotional support rather than for subject learning, and an app replaces none of those.
Is online adaptive practice as effective as a private tutor? For diagnosis and practice volume, often more effective, because it checks prerequisites every time and delivers immediate feedback on the five days a tutor is not there. For a child who is discouraged, distracted or very young, less effective. It depends which problem you are solving.
How much does maths tuition cost compared with an app? Verified published figures include ₹270 per group session for live olympiad teaching, ₹74,880 to ₹98,176 a year for Cuemath's general tutoring all-in with GST, and ₹562 for a full workbook set. Local home tuition rates vary too widely to quote; divide your monthly fee by hours delivered and compare against ₹270 per session.
Do learning apps really improve exam results? The category-level evidence is good: step-based tutoring systems produce roughly the same effect as human tutoring. Individual products' own improvement claims, including ours, are self-reported and uncontrolled, and no product in the Indian market publishes an audited outcome comparison.
Should I stop my child's tuition and use an app instead? Not without first identifying what the tuition is actually solving. If a tutor is re-teaching the chapter in the same order the school did, the money is buying supervision, and that should be compared against cheaper ways of supplying supervision. If the tutor is diagnosing why errors recur, that is the harder thing and worth keeping.
Is an app suitable for a class 3 or class 4 child? With caution. Research on problem solving before instruction finds negative effects for grades 2 to 5, where instruction-first outperformed problem-solving-first. Younger children generally need more teaching and more presence, and an app that drops them into unfamiliar problems and waits is using an approach the evidence does not support at that age.
Sophoz builds adaptive practice and does not currently publish pricing, which is a deficiency noted above. Every research figure here is sourced, including the ones that argue against buying an app.