A numerical reasoning solver needs more than a calculator. It must find the right figures, understand the question and show a calculation that leads to the answer. TestSolve reads that information off your screen and returns the worked answer to your phone, so you do not have to type a data table into a chat window.
The distinction becomes useful when the calculation is short but the display is dense. A table can contain several periods, departments and units, while the question depends on only two cells. Sending the original screen preserves the labels that tell the solver which two cells matter.
You can try the TestSolve numerical reasoning workflow free on your first question. It works through desktop apps for Windows and macOS or through browser screen sharing, with no account needed for the free question.
A useful solver turns a numerical display and a question into a checkable answer. That means selecting the relevant data, performing the right operation and interpreting the result in the requested form. Each stage can fail independently.
For example, a question about percentage growth needs a starting value and an ending value. An answer that merely subtracts the two has calculated the absolute change. The arithmetic can be correct while the response still answers the wrong question.
The worked explanation should make the chosen values visible to you. It should also make clear whether the requested result is a quantity, ratio, percentage or judgement about a statement. That is what separates a useful numerical reasoning answer from an unexplained number.
Aon's candidate information lists 37 tasks in 12 minutes for scales numerical. Dividing the available time across all 37 gives roughly 19 seconds per task. That is an average pace for attempting the whole set, not a separate timer imposed on each item.
Aon also explains that candidates are not expected to complete every question. The practical aim is to produce correct responses within the available time. A solver should therefore be assessed on the complete process of obtaining and checking an answer, rather than on an impressive-looking response alone.
These timings describe the named Aon format. Do not apply them to every numerical assessment, even if an invitation uses similar wording. The actual module instructions remain the reference for your test.
In Aon's published numerical guidance, statements refer to information held on different data sheets. Each statement relates to one relevant sheet, which the candidate must locate. The task therefore includes finding evidence before evaluating it.
That structure creates a specific limit for a screen reader. If the relevant sheet is hidden behind another tab, the visible screen may not contain enough information to answer. Reading the current display should not be confused with automatically opening or searching every sheet.
Once the correct sheet is visible, screen reading can remove the need to copy its contents manually. The solver can then work from the labels, values and question together. This is an important distinction between reducing input effort and claiming to automate the entire assessment interface.
Tables encode meaning through position. A figure under “forecast” is different from the same figure under “actual”, and a total row is different from one division's result. Pasting a run of numbers into a chat can strip away those relationships.
Manual transcription also creates opportunities to omit a negative sign, decimal place or unit. A value labelled as thousands should not be treated as a count of individual units. Screen reading keeps that surrounding context available to the AI.
It does not prevent the AI from choosing the wrong cell. The benefit is that you can keep the complete table as the source and check the explanation against it. A useful response should name the relevant category and period, rather than simply presenting arithmetic with unexplained inputs.
For repeated table questions, the screen reading numerical solver is worth assessing on that exact point. Can you recognise where its figures came from without reconstructing the question yourself? That is a more practical test than whether the response sounds sophisticated.
Image capable AI can attempt chart questions, including graphs and other numerical displays. The chart must contain enough legible information to support the requested comparison. A screenshot with a missing legend may be as incomplete as a text message with a missing sentence.
Charts create distinct interpretation risks. A bar's height, a plotted point and a labelled value are not always interchangeable, especially when the axes start above zero or the graph provides only approximate readings. The answer should respect the precision the display actually supports.
Keep an eye on the question wording too. “Largest increase” and “largest percentage increase” can select different categories. No input method removes the need to interpret that distinction correctly.
Choose the simplest tool that receives the complete question and produces useful working. Paying for screen reading is easier to justify when transferring the display is the main obstacle. It is harder to justify when all you need is a short multiplication.
| Tool or approach | Good fit | Limitation to consider |
|---|---|---|
| Calculator | A calculation you have already identified | You must select the data and operation |
| ChatGPT | Text problems and uploaded tables or charts | Input transfer and visual interpretation still need attention |
| Aptitude Solver custom GPTs | Task focused chat assistance | The name does not demonstrate better numerical performance |
| aichatonline.org or yeschat.ai | Trying a free or trial solver | Check actual image input and usage limits |
| TestSolve | Complete questions already on a computer screen | It cannot reason from a data sheet that has not been shown |
| Microsoft Math Solver | Historical example of focused arithmetic and algebra support | Microsoft announced retirement for July 2025 |
| Prep articles and directories | Understanding formats or finding products | Reading about a solver does not itself solve the screen question |
Microsoft Math Solver's mathematical focus was useful for equation-based problems. That should be acknowledged without presenting a retired product as a current download recommendation. For a simple expression today, an available free maths tool or calculator can still be enough.
The explanation should begin with the relevant data, not skip straight to an option letter. You need to know which period, category and unit were used. Otherwise, you cannot tell whether the tool read the question correctly.
Next comes the operation. For percentage change, the change is divided by the original value; for a share of a total, the part is divided by the relevant total. Those can involve similar-looking numbers but answer different questions.
Finally, the response should interpret and match the result. A request for the nearest whole number differs from one requiring an exact figure. If the calculation gives a result absent from the choices, the solver should re-examine its interpretation rather than silently select something nearby.
Some numerical questions ask whether a statement is true, false or cannot say. In that format, incomplete evidence is a legitimate reason to choose cannot say. It is not an invitation to estimate a missing value.
Suppose a display contains a company's revenue but no costs. A statement about its profit margin cannot be established from revenue alone. More calculation does not fix the absence of the required input.
There is also a difference between evidence missing from the underlying question and evidence missing from the screenshot. A hidden sheet could contain the answer. Before accepting cannot say, establish whether the tool has received the complete relevant material.
Check the data selection first, because an error there invalidates everything after it. Look at the named row, year and units. Then check whether the operation matches the wording of the question.
A quick scale check can expose obvious errors. If two nearby values are being compared, a result suggesting an enormous change should make you look again. Similarly, a share of a total should be compatible with the quantities actually shown.
Do not mistake a long explanation for a more dependable one. Under time pressure, a compact calculation with labelled inputs may be easier to assess. The question is whether the essential reasoning is visible and correct.
TestSolve reports 92% on a 50-question benchmark measured on 18 September 2026. That is the available overall accuracy figure. It does not supply a numerical-only result or an Aon-specific score.
Tables and charts are a clear use case for the product because screen reading addresses their input problem. That is a statement about the workflow, not a new performance statistic. Your first free question is useful for checking whether your particular layout is read correctly.
The tool also does not supply a universal right answer to personality questionnaires or strengths inventories. Game-based assessments score how you play, and solving a still image does not perform those interactions. Use outside assistance only where your assessment instructions permit it.
AI can attempt numerical reasoning questions from complete text, tables or charts. It must select the correct data as well as calculate correctly. Check the inputs, units and interpretation in the worked answer instead of relying only on the final option.
For questions already displayed in tables or charts on a computer, TestSolve offers direct screen reading and phone delivery. ChatGPT or a free maths tool can be sufficient for simpler inputs. The best fit depends on how much information you need to transfer.
Aon lists 12 minutes and 37 tasks for scales numerical. That works out at roughly 19 seconds each if you spread the time across every task, but it is not an individual question limit. Other numerical modules can have different timings.
Do not assume a screen reader can navigate hidden sheets automatically. TestSolve reads the information shown on your screen, so the relevant data needs to be visible. If the wrong sheet is displayed, the solver may lack the evidence needed for that statement.
Free chat tools and calculators can help, and some AI sites advertise trial access. TestSolve's first question is free without an account. The paid options are $14.99 for 30 questions or $19.99 for 50, with no subscription.
TestSolve has a macOS desktop app, a Windows app and a browser version with screen sharing. It reads the visible question and sends the worked answer to your phone. Include the chart's labels, units, legend and answer options in what the tool can see.
If your main obstacle is getting a table or chart into an assistant, start with the original screen. Try TestSolve on your first numerical question and inspect the figures and working on your phone. First question free, no account, no subscription.
<!-- Editorial source notes, not part of the article body: Aon timing: https://www.aon.com/en/capabilities/talent-and-rewards/prepare-for-your-online-assessment Aon sheet navigation, three-option format and completion expectations: https://assets.aon.com/-/media/files/aon/capabilities/talent-and-rewards/practice-tasks---numerical-reasoning.pdf Timing arithmetic: 720 seconds divided by 37 tasks is approximately 19.46 seconds per task. This is an average, not an item limit. Microsoft retirement: https://x.com/MicrosoftMath/status/1931103750753599946 Microsoft historical scope: https://news.microsoft.com/en-in/microsoft-math-solver-ai-app/ Product facts and benchmark supplied by TestSolve. Hidden-sheet navigation, per-format accuracy and response-time guarantees are not claimed. -->
First question free. No account, no subscription.