7 GMAT Data Insights errors that quietly cap scores in the high 70s

GMAT Data Insights is the youngest section of the GMAT Focus Edition, and it punishes a particular kind of careless reader. Candidates who sail through Verbal and Quant sometimes bleed points in Data Insights not because the maths is hard, but because the section rewards a specific discipline: it punishes sloppy reading, premature calculation, and a refusal to label what kind of item the test is actually asking. This article walks through the seven error patterns I see most often when a candidate plateaus in the high 70s, and it offers concrete reading and pacing moves for each one. By the end, the goal is not just to know the question types — Graphics Interpretation, Table Analysis, Data Sufficiency, Multi-Source Reasoning, Two-Part Analysis, and the integrated reasoning hybrids — but to recognise the precise moment inside the test when the careless move happens, and to install a replacement habit before that moment arrives.

Error 1 - Treating every Data Insights prompt as a maths problem

The single most common mistake I see is a candidate reading a Data Insights stem as if it were a Quant problem. The student jumps straight to the chart, starts scanning numbers, and only later checks what the question is actually asking. On the GMAT Focus, Data Insights is built to reward the opposite order. The stem almost always contains the instruction that turns the visual into a question, and the instruction is what decides the work the candidate has to do.

Consider a typical Graphics Interpretation item. The chart shows quarterly revenue across four product lines, and the prompt reads something like: "In the quarter in which Division B's revenue was closest to Division A's revenue, approximately what was the per-unit gross margin for Division D, given that Division D sold 18,400 units?" A student who rushes to the chart first will read the bars, estimate the gap between A and B, and then stare at Division D's axis trying to retrieve a margin figure that simply does not exist on the chart. The margin has to be derived from the chart's secondary axis. The reading order is the entire game.

The fix is mechanical and unglamorous. Read the stem first, fully, twice if necessary. Underline the verb ("closest to", "exceeded", "fell below", "at least", "at most"). Then translate the verb into a numeric shape — closest to means a minimum absolute difference; exceeded means a strict greater-than; at least means an inclusive inequality. Only after the stem is parsed should the candidate touch the visual. This single reordering saves between 30 and 60 seconds per item and prevents the most common class of misread answers.

For most candidates reading this, the change feels artificial at first. The habit of going to the visual first is so ingrained that the new order feels slower. In my experience, after 40 to 60 practice items, the new order stops feeling slow and starts feeling safer. The candidates who resist the reordering tend to be the ones who plateau at 78.

Error 2 - Skimming the chart's axes, units, and scale before answering

Skim-reading a chart is the second-most expensive habit. Candidates read the bar heights, the line slopes, or the table entries, and they answer the question with the numbers they think they saw. The problem is that GMAT visuals are built to be misread. The y-axis is often non-zero. The units switch mid-chart. The legend uses a colour the candidate is reading on a poorly calibrated monitor. Two of the five answer choices will be quantitatively defensible if the candidate misreads the scale — and that is not an accident.

The reading move that closes this hole is a 10-second axis audit before any calculation. The candidate should name out loud or write down: the x-axis label and unit, the y-axis label and unit, the starting value of each axis, the tick interval, and the legend mapping. On a Table Analysis item, the same audit applies to the column headers, the row labels, and any footnote symbols. The footnote is where the GMAT hides the unit conversion that breaks the candidate who did not look.

Concrete example. A line chart shows revenue in millions of dollars on the left axis and units sold in thousands on the right axis. The stem asks for revenue. The candidate who is not auditing will read the right axis, double it, and pick an answer that is off by a factor of a thousand. The candidate who audits sees two axes, names both, and answers in the correct unit. The test-makers know that a chart with two axes is a deliberate trap, and the test-makers are correct that most candidates will not catch the trap on a first read.

The unit audit also applies to percentage charts. A stacked bar showing market share can be read as either percentage of total or percentage of category, and the chart's footnote will specify. The candidate who skips the footnote answers a question that is technically true but answers a different question from the one on the screen. The 10-second investment is not optional.

Error 3 - Doing arithmetic the question does not require

The third pattern is arithmetic over-reach. Data Insights is the section of the GMAT Focus where the test-makers most aggressively reward a candidate who refuses to compute. Several answer choices can be eliminated by inspection. One or two can be confirmed by inspection. Only the rare item requires a full pencil-and-paper calculation, and even then the calculation is usually a single step rather than a chain.

The pattern of over-reach looks like this. The candidate sees a stem asking which of five answer choices is closest to the average growth rate across four quarters. The candidate takes out the scratch paper, computes the quarter-by-quarter growth, sums them, divides by four, and arrives at a number. The number, of course, is wrong — not because the maths was wrong, but because the stem said "closest to" and the candidate needed an estimate, not a value. The test-makers built the distractor answers to be exactly close enough to the true value that the candidate's rounding error is enough to land on a wrong choice.

The fix is a discipline of asking, before calculating, whether the answer can be ranked by inspection. For an average growth rate, the four growth values cluster around a band; the average lives inside the band. The candidate who identifies the band — for example, all four values between 3% and 7% — can eliminate any answer outside the band, and then pick the middle of the band. No arithmetic needed. The candidate who computes the precise average falls into the rounding trap.

For Two-Part Analysis items, the same logic applies. The candidate is asked to pick two values, one from each column, that jointly satisfy a condition. The temptation is to enumerate. The discipline is to test the two most likely candidates first, eliminate the column on the basis of a single test, and then narrow the other column. Two-Part Analysis items almost always give the candidate enough information to do a binary elimination rather than a full enumeration. The candidate who enumerates spends 3 to 4 minutes; the candidate who triages spends 90 seconds.

A simple triage framework for "compute or not?"

When a stem lands, the candidate can run a four-step triage in roughly 20 seconds. Step one: read the verb. If the verb is "closest to", "approximately", or "is best described as", the answer is an estimate. Step two: check the answer choices. If the choices are wide apart — say, a factor of five or more — the answer can be ranked by inspection. Step three: check the chart's tick spacing. If the ticks are coarse, the chart itself is an estimate, and the answer should match the chart's resolution. Step four: only then decide whether to reach for the scratch pad.

This triage is the difference between a 78 and an 84 on Data Insights. The 78 scorer computes when the question wants an estimate; the 84 scorer estimates when the question wants an estimate, and computes only when the question forces precision. The test-makers calibrate the answer choices to reward this judgement.

Error 4 - Reading a Data Sufficiency stem like a Quant stem

Data Sufficiency is the oldest item family in the GMAT lineage, and it is the one that Data Insights inherits in its purest form. Candidates who come from a heavy Quant background tend to over-solve. They read the stem, identify what the question is asking, and then attack statement 1 with a full calculation. The question, however, is not "what is the value" — the question is "is the value determinable, and is statement 1 enough on its own?"

The classic trap is the statement-1-always-sufficient bias. A candidate sees statement 1 and computes the answer, marks statement 1 as sufficient, and moves on without checking statement 2. Sometimes the answer to statement 1 is correct, and statement 2 is redundant. But on the harder items, statement 1 is sufficient for a different value, or statement 1 is sufficient only under an assumption the candidate has imported from the stem. The test-makers design the distractor to be sufficient-looking and insufficient in fact.

The reading move that prevents this is a strict protocol: read the stem, identify the target value, identify the constraint, then test each statement in isolation. The test for sufficiency is not "can I compute an answer" but "does this statement, taken alone, fix the target value to a single number, a single yes, or a single no". If the statement leaves the value ambiguous, it is insufficient, no matter how confidently the candidate can crunch a number from it.

For most candidates, the hardest shift is accepting that an answer of "insufficient" can be correct for both statements. The stem may genuinely be unanswerable from the given data, and that is the test's design. The candidate who insists on a numeric answer is the candidate who will pick the wrong "sufficient" choice. In my experience, this is the single most common reason a Verbal-strong candidate caps out in Data Insights — the candidate treats Data Sufficiency as a Quant sub-test and reaches for numbers that the prompt is not asking for.

Error 5 - Letting a Table Analysis question become a re-read of the entire table

Table Analysis is the item family that punishes skim-reading and over-reading in equal measure. The tables are wide — sometimes 8 to 12 columns — and the sortable feature lets a candidate reorder the rows by any column. The stem usually asks a question that requires two columns to be read together, and the candidate who reads the table sequentially often runs out of time before isolating the relevant rows.

The reading move that fixes this is column-first, not row-first. The candidate should read the stem, name the two columns that are relevant, and then sort the table by one of them. Sorting reveals the extreme values and the cluster. The candidate who sorts can answer most Table Analysis items by inspecting the top three or bottom three rows, and never has to read the table in full.

Concrete example. A table shows 11 rows of supplier data with columns for region, lead time, defect rate, and unit cost. The stem asks which supplier has the lowest defect rate among those with a lead time under 10 days. The candidate who reads row-by-row will need 90 seconds. The candidate who sorts by defect rate first will see the lowest-defect supplier at the top, check its lead time, and either confirm or move to the next row. Total time: 25 seconds.

The same logic applies to the "which column is sortable but irrelevant" trap. The table often includes a column that the candidate is tempted to sort by, but the stem's condition is on a different column. The candidate who sorts by the wrong column spends 60 seconds confirming that the sorted data does not contain the answer, and then has to re-sort. The triage protocol — name the relevant columns before touching the sort — eliminates the wasted sort.

Common pitfalls and how to avoid them

Across the seven error patterns, three pitfalls repeat often enough to deserve a named fix. First, the candidate who reads the stem and then forgets the verb by the time the chart is being inspected. The fix is to write the verb above the scratch pad before the chart is read. Second, the candidate who answers a previous item while mentally still on the current item — that is, the candidate marks an answer for item N when the work on the screen is for item N+1. The fix is a strict 5-second pause at the end of each item to confirm the item number. Third, the candidate who leaves a hard item blank instead of marking a best guess. The GMAT Focus does not penalise wrong answers, so a guess is strictly better than a blank. The fix is a rule: if the 90-second budget is exceeded, mark the most defensible answer and move on.

Error 6 - Letting Multi-Source Reasoning tabs become a re-read of the prompt

Multi-Source Reasoning presents two or three tabs of information — typically a short scenario, a chart or table, and a follow-up email or memo. Each tab has its own questions, and the candidate is expected to integrate information across tabs. The trap is re-reading the scenario tab on every question. The scenario tab is, in effect, a passage, and the questions are passage-based. Re-reading the passage on every question is the single most expensive habit a candidate can bring from the Reading Comprehension section.

The reading move is to read each tab once, in full, and to build a mental map of where each piece of information lives. The map is a single line per tab: "Tab 1: scenario, vendor X awarded the contract in March. Tab 2: cost table, vendor X is the third column. Tab 3: email, vendor X's delivery was delayed in May." With that map, the candidate can answer any question by retrieving the relevant tab without re-reading. Total time saved per question: 20 to 30 seconds, which compounds across the four to five questions that follow a Multi-Source set.

For the harder questions — the ones that ask what a reader of the email would infer, or which conclusion is supported by combining the table with the email — the map is what makes the question answerable. The candidate who re-reads the email will arrive at the right answer, but the candidate who retrieves the email from memory will arrive at the right answer in half the time. The test-makers calibrate the section's pacing to a candidate who builds a map, not a candidate who re-reads.

Error 7 - Mismanaging pacing because the section is "only 20 questions"

The final error pattern is structural rather than tactical. Data Insights on the GMAT Focus is 20 questions in 45 minutes, which works out to 2 minutes and 15 seconds per question including the review. Candidates who treat this as a relaxed pace often run out of time on the last three items. The section is not a sprint, but it is also not a stroll. The 2:15 average masks a real distribution: easy items run 60 to 90 seconds, medium items run 2 to 2.5 minutes, and hard items can run 3 to 4 minutes. The candidate who spends 3 minutes on every medium item will not finish.

The pacing protocol is to bank time on the easy items and spend the banked time on the hard ones. If the candidate finishes an easy Graphics Interpretation in 70 seconds, the banked 65 seconds is available for the next Data Sufficiency stem. The candidate who refuses to skip — who insists on completing every item to full confidence — will run out of time at item 18 and either rush the last two or guess on the last two. Both outcomes cost more points than skipping one medium item and coming back to it with a clear head.

For most candidates reading this, the hardest pacing change is the willingness to flag and move. The GMAT Focus allows the candidate to mark an item for review and return to it later in the section. The flag is a tool, not a confession of weakness. In my experience, the candidates who plateau in the high 70s are the candidates who do not use the flag. The candidates who move into the 80s are the ones who flag, move, and return.

A comparative read of the five item families and their dominant error patterns

The error patterns cluster differently across the five item families, and recognising the cluster is what makes the section feel predictable. The table below maps each item family to its dominant error pattern and the reading move that closes the gap.

The pattern across the families is consistent. The error is almost never a maths error. The error is a reading error that the candidate then tries to compensate for with more arithmetic, which compounds the original mistake. The test-makers design the answer choices to exploit exactly this compounding.

Building a preparation plan that targets these seven errors

A preparation plan that targets these errors looks different from a plan that targets content review. The candidate does not need a second pass through percentages or a refresher on standard deviation. The candidate needs a set of 60 to 80 practice items, divided evenly across the five families, worked under strict timing, with an error log that names the error pattern by number. The log is what makes the practice productive. The candidate who logs "got it wrong" learns nothing. The candidate who logs "Error 3 — computed when the stem asked for an estimate" can target that pattern in the next session.

The recommended cadence is four to five sessions per week, with 15 to 20 items per session. After each session, the candidate reviews only the wrong and the flagged items, names the error pattern, and writes a one-sentence rule that the candidate will follow on the next session. After two weeks, the error log will reveal a dominant pattern — usually Error 1 or Error 3 — and the candidate can spend the third week on a single pattern until the rate of recurrence drops below one in ten.

The plan should also include at least two full-length timed sections. The 45-minute pacing is felt, not calculated, and the candidate who has never sat for the full 45 minutes will discover on test day that the last five items feel very different from the first five. The full sections should be taken with the official interface if possible, because the click patterns and the flag-and-return feature behave differently in different platforms.

Conclusion and next steps

Data Insights on the GMAT Focus is a reading section that uses data as its text. The seven error patterns above are the predictable failure modes of a candidate who treats the section as a maths section, and the reading moves that close each gap are mechanical, learnable, and repeatable. The candidates who break into the 80s on Data Insights are almost always the candidates who have stopped trying to compute their way out of a misread and have installed a disciplined reading order in its place. The next concrete step is a single timed set of 20 official-style items, taken under the 45-minute clock, with an error log that uses the seven error numbers above. TestPrep İstanbul's diagnostic assessment is a natural starting point for candidates who want a sharper readout of which of the seven patterns is doing the most damage to their current Data Insights score.

Frequently asked questions