Artificial Idea | AI careers · practical prompts · no hype Thursday, November 13, 2025 · Issue #30 · Prompt Tutorial
The data barrier, removed
The data analysis prompt: how non-technical people can make sense of spreadsheets with AI
Every professional is drowning in data and starving for insight. The gap between those two conditions is not a statistics problem. It is a prompting problem.
Issue #29 made the case that the AI transition is producing different opportunities and challenges at different career stages, and that the professionals navigating it best are those who engage with it honestly and specifically rather than through generic upskilling advice. Across every career stage and every story in that issue, one capability appeared consistently as a differentiator: the ability to work with data well enough to make decisions grounded in evidence rather than intuition alone.
This is the capability with the steepest perceived barrier among non-technical professionals and the lowest actual barrier once AI tools are applied to it correctly. The perceived barrier is a statistics and programming barrier. The actual barrier is a question-formation barrier. The professionals who work with data effectively are not primarily those who can run a regression or write a Python script. They are those who can identify the right question to ask of a dataset, evaluate whether the answer they receive is reliable, and communicate what the data shows to an audience that did not see the analysis.
All three of those capabilities are developable with the prompts in this issue, without a statistics course, without a data science background, and without anything beyond access to a spreadsheet and a conversational AI tool.
The question formation problem
Before the prompts, the foundational concept that makes data analysis accessible to non-technical professionals is worth establishing clearly, because it changes how the prompts are used and why they work.
Most non-technical professionals approach data with a question they cannot quite formulate. They have a spreadsheet, they have a sense that it contains something relevant to a decision they are trying to make, and they lack the technical vocabulary to ask for what they need in a way that produces it. The result is either that they do not engage with the data at all, outsourcing the analysis to someone technical and receiving conclusions they cannot evaluate, or that they produce superficial analysis, the sum and the average, that answers no meaningful question about the decision they are trying to make.
The gap is not technical. It is interrogative. The non-technical professional does not need to learn how to analyse data. They need to learn how to ask the right questions of it, how to evaluate whether the answers they receive are reliable, and how to communicate those answers to others. AI handles the technical execution of the analysis. The professional handles the intelligence that makes the analysis relevant to the actual decision.
That division of labour is the one these prompts operationalise.
Prompt 1: The dataset orienteer
The problem it solves: understanding what a dataset contains, what questions it can and cannot answer, and what its limitations are before attempting any analysis, which is the step most non-technical professionals skip and the one most responsible for analysis that misleads rather than informs.
You are a senior data analyst helping a
non-technical professional understand
a dataset before conducting any analysis.
Here is a description of my dataset:
[describe the data: what it contains,
where it came from, what each column
or field represents, how many rows
or records it contains, and the
time period it covers]
Or here is a sample of the data itself:
[paste the first 20-30 rows if the
dataset is not confidential]
The decision or question I am trying
to answer with this data:
[describe the business or professional
question as specifically as possible]
Please:
1. Assess whether this dataset is
capable of answering my question
as stated, and if not, what
additional data would be required
2. Identify the three most important
things to understand about this
dataset before conducting analysis:
its limitations, its likely sources
of error, and any structural
characteristics that could
mislead an inexperienced analyst
3. Reformulate my question into
the two or three specific,
measurable sub-questions that
the dataset can actually answer
and that together address my
original question
4. Identify the analysis most likely
to produce a misleading result
with this data if conducted
without appropriate caution,
and what the caution requires
Do not conduct any analysis yet.
Help me understand what I am
working with before I ask
anything of it.
The instruction not to conduct any analysis yet is the constraint that makes this prompt serve the question formation problem rather than bypassing it. Non-technical professionals who jump directly to analysis without orientation consistently make the same errors: they ask questions the data cannot answer, they do not know when results are misleading, and they cannot evaluate the reliability of conclusions because they do not understand the dataset's limitations. Orientation before analysis prevents all three.
Prompt 2: The pattern finder
The problem it solves: identifying the most significant patterns in a dataset without needing to know in advance what those patterns are, which is the analytical capability most useful for exploratory work where the decision-relevant insight is not yet known.
You are a data analyst conducting
exploratory analysis to surface
the most decision-relevant patterns
in a dataset.
Dataset description or sample:
[paste or describe]
Business context: [what organisation
or function this data comes from,
what decisions are being made
in this context, and what would
count as a useful finding]
Please:
1. Identify the three most significant
patterns in this data, described
in plain language that a
non-technical decision-maker
can act on, not in statistical terms
2. For each pattern, assess its
reliability: is it a robust signal
or a pattern that could be an
artefact of the data's limitations
or a statistical coincidence
3. Identify any pattern that appears
counterintuitive relative to
what someone familiar with
this business context would expect,
and flag it for closer examination
4. Identify what the data does not
show that would be important
for decision-making in this context,
the absence of a pattern being
as informative as its presence
For each finding, tell me what
additional analysis would confirm
or disconfirm it before I act on it.
Point four, what the data does not show, is the analytical observation most consistently absent from data analysis produced by non-technical professionals and most consistently present in analysis produced by experienced ones. Absence of evidence is evidence of absence only under specific conditions, and identifying those conditions requires understanding the data well enough to know what it would show if the pattern existed. This prompt builds that orientation into the analysis.
Prompt 3: The comparison builder
The problem it solves: conducting meaningful comparisons between groups, time periods, or categories in a dataset without falling into the comparison errors that produce misleading conclusions, which are the most common analytical errors in non-technical data work.
You are helping me conduct a comparison
analysis that will be used to support
a professional decision.
The comparison I want to make:
[describe exactly what you want
to compare: two time periods,
two customer segments, two products,
two teams, or any other comparison]
The dataset: [describe or paste]
The decision this comparison will inform:
[be specific about what you will
do differently based on the result]
Please:
1. Conduct the comparison I have
described and state the result
in plain language
2. Assess whether the comparison
is a fair one given the data:
are the groups being compared
similar enough in the relevant
dimensions that the comparison
is meaningful, or are there
confounding differences that
could explain the result
independently of what I am
trying to measure
3. Identify the most likely
alternative explanation for
the result I have found,
assuming the result is real
4. Tell me what sample size
or data volume considerations
affect the reliability of
this comparison, in plain
terms rather than statistical ones
5. State whether the result
is strong enough to act on
or whether it is directional
evidence that requires
further confirmation
Be direct about the difference
between a result that is
statistically reliable and one
that is suggestive but uncertain.
Non-technical decision-makers
are best served by honest
uncertainty quantification,
not by false precision.
The instruction to be direct about the difference between reliable and suggestive results is the constraint that makes data analysis honest rather than persuasive. Non-technical professionals frequently receive data analysis that presents uncertain findings with the confidence of established ones, because the analyst is incentivised to produce a clean answer rather than an honest one. This prompt builds the honesty into the output rather than leaving it to the analyst's discretion.
Prompt 4: The insight communicator
The problem it solves: translating the findings from a data analysis into a communication that a non-technical audience can understand, evaluate, and act on, without losing the nuance that makes the analysis reliable.
You are helping me communicate
data analysis findings to an
audience that did not conduct
the analysis and is not
technically sophisticated.
The analysis I conducted and
its key findings: [describe
the analysis and paste or
summarise the results]
My audience: [their role,
what decisions they are making,
what they already believe
about this topic, and how
much time they will give
this communication]
The action I want them to take
or the decision I want them
to make based on this data:
[be specific]
Please produce:
1. A headline finding in one
sentence that is accurate,
specific, and actionable,
not a description of
what the analysis did
but a statement of
what it found
2. Three supporting points
that substantiate the
headline finding, each
in two sentences,
each grounded in
a specific number
from the analysis
3. The most important
caveat or limitation
that the audience
needs to know before
acting on this finding,
stated in plain language
4. The recommended action,
stated specifically
enough that the audience
cannot misinterpret
what they are being
asked to do
Constraint: every number
cited must come from
the analysis I have described.
No illustrative figures,
no rounded approximations
presented as precise,
no numbers that cannot
be traced back to the data.
The constraint that every number must be traceable to the actual analysis is the one that protects the communicator's credibility. Data communications that include illustrative or approximate figures, presented as if they come from the analysis, are one of the most common sources of credibility damage in professional settings. The constraint prevents it.
Prompt 5: The decision validator
The problem it solves: checking whether a decision that is about to be made is actually supported by the data cited in its support, which is the analytical step most frequently skipped and most consequential when skipped.
You are a rigorous analytical reviewer
checking whether a decision is
adequately supported by the data
cited in its favour.
The decision being made:
[describe specifically]
The data cited in support of it:
[describe or paste the analysis
or findings being used to
justify the decision]
Please:
1. Assess whether the data cited
actually supports the specific
decision being made, or whether
it supports a related but
different conclusion that
is being extended beyond
what the data warrants
2. Identify the gap between
what the data shows and
what the decision requires
to be well-founded,
stated as specifically
as possible
3. Identify what additional
data or analysis would
be required to make
this decision well-founded,
and whether obtaining
it is feasible given
the decision timeline
4. Give your assessment of
whether this decision
should proceed as proposed,
be modified based on
the data limitations,
or be deferred pending
better information
Be direct. A decision made
on inadequate data that
is presented as data-driven
is worse than a decision
made on judgment that
is presented as such,
because it has false
confidence attached to it.
The closing instruction reflects one of the most important principles in evidence-based decision-making and one of the least frequently applied. Decisions made on genuine judgment, acknowledged as such, can be evaluated and revised when the judgment proves wrong. Decisions made on inadequate data, presented as data-driven, are defended with the authority of evidence they do not actually have, which makes them harder to revise when they prove wrong and more damaging when they are.
The thirty-minute data practice
These five prompts used in sequence turn any dataset into a decision-relevant analysis in under thirty minutes for a professional who has no technical background and has never written a line of code.
The sequence: orient before analysing, find patterns before testing hypotheses, build comparisons carefully, communicate findings honestly, and validate that the decision is actually supported by the data before making it.
That sequence is what experienced data analysts do. It is what separates analysis that informs decisions from analysis that decorates them. The technical execution is now available to anyone with a prompt and a dataset. The intelligence that makes the execution relevant to the actual decision has always been the scarce resource, and it remains the scarce resource.
These prompts put the technical execution within reach. The intelligence is yours to supply.
Monday we are examining one of the most consequential and least discussed dynamics in the AI and careers space: the specific moment at which the accumulation of small, consistent investments in AI capability begins to produce returns that are visible to people other than the investor. The compounding curve has an inflection point, and understanding where it is changes how you think about the investments that precede it.
The inflection point is closer than most professionals currently believe. Monday explains why.
— The Artificial Idea team
