How does GMAT Data Insights reward reasoning over arithmetic?

GMAT Data Insights is the newest section on the GMAT (formerly the GMAT Focus Edition) and, for most candidates, the most unfamiliar. It replaces the old Integrated Reasoning section almost item-for-item, but the scoring scale, the on-screen toolkit, and the weight the section carries in your overall 205-to-805 profile have all been rebalanced. The section runs 20 questions in 45 minutes, and every question is built around a small piece of data: a chart, a table, a passage, an exchange rate, a two-email thread, or a partially-revealed spreadsheet. What you actually get tested on, beyond reading numbers off a screen, is the discipline of asking the right question before you touch the answer choices. That single habit is the difference between a 75th-percentile performance and a 95th-percentile one.

This article walks through the five item families that the section draws from, the reasoning moves each one rewards, the scoring mechanics, and the preparation strategy that produces consistent gains. The aim is practical: by the end, you should know what each question type is asking of you, how long to budget for it, and which habits silently drag the score down.

The five item families inside GMAT Data Insights

The section draws on a defined pool of item types, and although the visual formats vary, the underlying reasoning is built from five families. Knowing which family you are looking at is the first triage decision, and it usually happens within the first 15 to 20 seconds of reading the prompt.

Data Sufficiency (DS) in the Data Insights context

The Data Insights section still contains standalone Data Sufficiency items, the same question shape that appears in the Quant section. The format is unchanged: a question stem followed by two statements, with five fixed answer choices that always ask whether the statements, individually or together, are sufficient. What shifts in Data Insights is the type of question being asked. Instead of "what is the value of x", you are more likely to see "is the average order value greater than \$50?" or "did revenue exceed budget in the third quarter?". The arithmetic is light; the modelling is everything. Candidates who treat DS as a memorised five-choice pattern tend to over-invest in statement combinations when the real move is to recognise the question type first: value question, yes/no question, or comparison question. Each responds to a different elimination path.

Multi-Source Reasoning (MSR)

MSR presents two or three tabs of information, usually a short scenario plus an email chain, a report, and a chart. You click between tabs to integrate the data, then answer two or three questions about the bundle. Time is the enemy here, not difficulty. Most candidates over-read on the first pass and run out of time on the second or third sub-question. The tactical fix is a 90-second skim of every tab before you read the first question, and a hard rule that you never re-read a tab you have already parsed.

Table Analysis (TA)

Table Analysis gives you a sortable spreadsheet and asks whether you can determine a specific fact, often a total, a ratio, or a condition that holds across rows. The tool is genuinely useful, and most candidates underuse it. The right move is to sort by the column referenced in the stem before you evaluate any answer, because the question is usually framed so that one sort order makes the answer obvious and another hides it.

Graphics Interpretation (GI)

GI pairs a single chart with two statements and asks you to mark each as true or false. The trap is in the wording: statements are written to be partially correct, so you have to verify each clause. The 50/50 framing also makes this the easiest family on which to leave points on the table, because careless reads on a true/false pair can flip both answers.

Two-Part Analysis (TPA)

TPA presents a scenario plus a question with two linked answer choices, often expressed as a pair (X, Y) drawn from two columns. The linking is the whole point: the same trade-off or relationship has to hold for both answers, so a candidate who solves for one part mechanically often gets the second part wrong by ignoring the constraint. TPA items reward candidates who sketch the relationship before touching the answer grid.

Across all five families, the consistent lesson is the same: triage the family, decide whether the question is value, yes/no, or comparison in form, then act. Candidates who skip that first 20-second read and jump straight to the data tend to lose 30 to 45 seconds per item, which compounds into two to three lost questions over a 45-minute section.

What the section actually rewards: reasoning over arithmetic

GMAT Data Insights is not a numeracy test. The arithmetic on display is almost always single-digit or simple two-digit, with no quadratic factoring, no trigonometric identities, and no probability chains. The numbers are there to be read, not to be computed. What separates strong performers from the rest is the quality of the question they ask themselves before they touch the answer choices.

Consider a typical Data Sufficiency item in this section. The stem might read: "A company's monthly revenue in March was greater than its monthly revenue in February. Was the average monthly revenue for the three-month period from January through March at least \$20,000?" The two statements then give you fragments of data, such as January revenue and the percentage change from January to March. The arithmetic you might eventually need is division by three. The reasoning you need immediately is to identify the question as a yes/no DS item with an averaging structure, then ask: do I need the actual average, or do I need a lower bound on the sum? The moment you name the question type, the sufficiency check becomes mechanical. Most candidates miss this because they start computing before they have classified the prompt.

The same principle applies to Multi-Source Reasoning. An MSR bundle might contain an internal memo, a sales report, and a customer-survey chart. The questions attached to it often ask which statement is supported, which is contradicted, or what additional piece of data would resolve a specific disagreement. A candidate who reads the questions first and only then returns to the tabs to harvest the relevant line saves roughly 90 seconds per bundle. That is the difference between finishing MSR sub-questions calmly and guessing the last one because the timer ran out.

Graphics Interpretation makes the reasoning emphasis even clearer. The chart will display two quantities, often with overlapping error bars or stacked categories. The two statements usually combine a true observation with a false inference, and the false clause is hidden in a quantifier ("most", "all", "some") or in a causal claim that the chart does not actually support. The arithmetic is trivial. The intellectual work is in refusing to accept a statement as true until each clause has been checked against the data.

Two-Part Analysis items often look like the hardest of the five because they ask you to fill two blanks at once, but they are usually the most constraint-rich, which means they are also the most tractable once you have sketched the relationship. If the scenario describes a trade-off between cost and delivery time, the two parts of the answer are not independent: optimising one usually punishes the other, and the answer pair reflects that trade-off. Candidates who try to solve each part in isolation end up selecting an answer that satisfies one part but contradicts the other.

Pacing the 45-minute section: minute-per-question budgets

Pacing is where most candidates leave points on the floor, not content knowledge. With 20 questions in 45 minutes, the average budget is 2 minutes 15 seconds per question, but that average hides a wide spread. Data Sufficiency items can be solved in 75 to 90 seconds once you have classified the question. MSR sub-questions often run 100 to 120 seconds, but the first question of a bundle is cheaper than the second or third. TPA items routinely take 150 to 180 seconds because the constraint must be modelled before you select.

The reasonable budget looks like this in practice. For a 20-item section, plan to spend roughly 18 to 20 minutes on the DS-style items, 12 to 14 minutes on MSR, 4 to 5 minutes on TA, 4 to 5 minutes on GI, and 6 to 8 minutes on TPA. That leaves about 2 minutes of buffer, which you should treat as insurance for the single hardest item in the section, not as slack to be burned on the first ten questions.

Two tactical rules follow. First, never let a single item eat more than 3 minutes 30 seconds of clock. Beyond that, your expected return on the time invested drops below the expected return on a fresh question you can solve cleanly. Mark, move, and return only if the section is light. Second, do not save all TPA items for the end. The reasoning load is high, and a tired brain at minute 40 will misread the constraint. Interleave the families instead of letting one pile up.

Common pitfalls and how to avoid them

The most expensive mistake in Data Insights is reading a chart's title and assuming you know the units. A chart labelled "Average revenue per customer (USD)" looks identical to one labelled "Average revenue per customer (EUR, thousands)" until the scale matters. Always locate the unit label before you evaluate a statement. The second most expensive mistake is misreading the conditional in a DS item. A stem that asks "Was x greater than y?" is a yes/no question; a stem that asks "What is the value of x?" is a value question. The sufficiency paths are different, and confusing them leads to picking statement combinations that look right but answer the wrong question. Finally, in MSR bundles, candidates often answer a sub-question from the wrong tab. The tabs are designed to look similar. Tag each tab in your head the first time you open it, and re-check the tag before you commit to an answer.

Scoring mechanics: how the 60-to-90 scale interacts with the 205-to-805 profile

Data Insights is scored on a 60-to-90 scale, the same range as Quant and Verbal, and the three sectional scores combine with roughly equal weight to produce the overall 205-to-805 total. The exact weighting is not published as a single formula, but the practical effect is clear: a strong Data Insights score can lift an uneven profile, and a weak one can drag down a strong Quant or Verbal. Because the section has only 20 items, every question carries more weight per item than Quant or Verbal, where each section has more than 20 items. A single careless answer in Data Insights is roughly twice as expensive as a single careless answer in Quant.

The enhanced score report that arrives with your results breaks the section down by content area, and the categories used are the five item families themselves, plus a cross-cutting "reasoning" category. For most candidates, the report is most useful as a triage tool: it tells you which family contributed the most missed items, and that is where the next round of preparation should focus. A candidate who scores well on DS in Data Insights but stumbles on TPA has a much narrower fix to make than a candidate who scatters misses across all five families.

How Data Insights interacts with the rest of the profile

Admissions committees rarely look at sectional scores in isolation; they look at the overall total first, then at the pattern. A balanced 79 Quant, 78 Verbal, 80 Data Insights reads as a strong generalist. A 85 Quant, 70 Verbal, 65 Data Insights reads as a candidate who can handle quantitative work but is shaky on data-driven reasoning, which is exactly the skill profile that business school case work and the MBA core demand. A useful internal rule: aim for sectional scores within 5 points of each other, and treat any gap larger than 8 points as a signal that the weaker section deserves focused preparation, not just extra practice tests.

Preparation strategy: the four-week skill loop

Most candidates who improve meaningfully on Data Insights do so through a four-week skill loop, not through more full-length practice tests. The loop has four phases, each about a week long, and each phase targets a specific failure mode.

Week one is the classification phase. Take a single untimed set of 20 mixed items and, for each one, name the family out loud, identify the question shape (value, yes/no, comparison), and write down the first reasoning move you would make. The point is not to score well; the point is to make the triage step automatic. Candidates who skip this phase tend to stay slow on the real exam because they keep re-discovering the family on every item.

Week two is the tool phase. Spend deliberate time on the on-screen toolkit: the sort feature in Table Analysis, the tab-switching in MSR, the answer grid in TPA. Practice each tool in isolation until it feels like an extension of the cursor. The toolkit is a force multiplier, and most candidates use about 40% of its capability on the first pass.

Week three is the timing phase. Take timed sets of five items, with a hard cap of 11 minutes. The cap forces triage: you either classify the question and move, or you stall. After each set, log the family of every item you missed or had to mark, and look for patterns. A candidate who misses three TPA items in a row has a different fix to make than one who misses one DS, one MSR, and one GI.

Week four is the integration phase. Take full-length timed sections, but treat them as rehearsal, not as score events. After each section, do a 15-minute debrief: which items took too long, which families dragged, where the timer surprises came from. The aim of the integration phase is to lock in a per-item budget that holds under pressure.

Diagnostic question set for self-assessment

Before launching into the four-week loop, run a 10-item diagnostic that draws from at least three families. The point is to get a baseline read on which families feel natural and which feel opaque. A useful diagnostic set is one DS item, two MSR sub-questions, two GI items, two TPA items, one TA item, and two DS items from the Data Insights context. If you score evenly across the families, the four-week loop is the right next step. If one family collapses, spend the first week on that family alone.

Item-family comparison: which family rewards which skill

The five families look superficially similar, but the skills they reward are different enough that preparation can be targeted rather than generic. A useful way to read the section is to think of each family as a particular lens on the same underlying data.

Reading the table, the right diagnostic question is not "which family is hardest for me" but "which family do I lose the most time on, even when I get the right answer?". Time loss is a leading indicator of accuracy loss, and it is much easier to fix before the score collapses.

Common item patterns you can train for in advance

Although the section is designed to resist rote memorisation, certain patterns recur often enough that pattern recognition is a legitimate part of preparation. Three patterns in particular are worth drilling.

First, the conditional yes/no DS pattern. The stem asks whether a condition holds, and the two statements each give a piece of the underlying data. The trap is that statement (1) often makes the condition true and statement (2) often makes it false, which is exactly the kind of asymmetric answer that catches candidates who have only drilled value-style DS items. Train for this by deliberately solving at least one yes/no DS item per practice set, and by writing down the answer in plain language ("yes, the condition holds; no, the condition fails") before you touch the fixed-choice grid.

Second, the supported-versus-contradicted MSR pattern. The question asks which of five statements is supported, which is contradicted, and which cannot be determined. The efficient move is to read each statement and tag it in your head as S (supported), C (contradicted), or U (undetermined) before you look at the answer choices. The tagging collapses a 30-second decision into a 5-second lookup.

Third, the trade-off TPA pattern. The two answer parts are linked by a cost-versus-benefit, speed-versus-accuracy, or risk-versus-return relationship. The efficient move is to identify the trade-off before you look at the answer grid, sketch the curve in your head, and then select the pair that sits on the trade-off line. Candidates who try to evaluate each column independently often pick a pair that is internally inconsistent, which is the most common cause of TPA errors.

What a strong preparation timeline looks like

A reasonable timeline for a candidate whose baseline Data Insights score sits in the low 70s and whose target is the low 80s runs 8 to 10 weeks, with 6 to 8 hours of focused work per week. The first two weeks go to the classification phase, the next two to the tool phase, the next two to the timing phase, and the final two to integration and full-length rehearsal. The mistake most candidates make is to skip straight to full-length practice tests, which is a measurement activity, not a training activity, and produces a stream of scores that do not improve.

A second mistake is to over-invest in mock exams at the expense of item-by-item analysis. A single 20-item set, fully debriefed, is worth three to four untimed sets. The debrief is where the learning happens. A useful debrief template is: which item took too long, which item I got right for the wrong reason, which item I misread, and which item I would solve differently on a second pass. Filling that template for 10 items takes 45 to 60 minutes and is the single highest-return activity in the whole preparation cycle.

How to read your progress without overreacting to noise

Because Data Insights has only 20 items, score variance from one practice set to the next is high. A swing of three to four scaled points from one week to the next is normal, and it is noise, not signal. The right way to read progress is to look at the rolling average over four to five sets, not the latest set in isolation. If the rolling average is flat after three weeks of focused work, the preparation is misaligned and needs to be retargeted, not intensified.

Putting it together: a candidate's working routine for the final fortnight

In the two weeks before the exam, the routine should narrow. Drop the broad family-by-family drills and run two full-length timed sections per week, each followed by a 60-minute debrief. Keep a "three-strikes" log of every item type that produced an error or a forced guess, and spend 20 minutes per day on the worst-performing family from the log. The aim of the final fortnight is to consolidate, not to expand, and the right score gain at this stage is usually one to two scaled points from reduced time loss, not from learning new content.

Two more rules. Do not introduce a new item type in the final week. The cognitive load of pattern-matching a fresh family is not worth the marginal coverage. And do not cram the night before. Sleep is a higher-yield preparation activity than any practice set, and the section rewards a clear head more than a saturated one.

Conclusion and next steps

GMAT Data Insights rewards a particular kind of discipline: classify the family, name the question shape, model the constraint, then act. The arithmetic is light, the toolkit is generous, and the section is short enough that pacing is the dominant lever. Candidates who train the triage step explicitly, who use the on-screen tools deliberately, and who debrief every practice set in full will see steady gains across the eight-to-ten-week timeline. A useful next move is to spend a single 60-minute session auditing your performance across the five item families and selecting the weakest one for focused work. TestPrep Europe's targeted drills on the family you identify are a natural starting point for candidates building a sharper preparation plan around Data Sufficiency versus Multi-Source Reasoning triage.

Frequently asked questions