Sections
Before You Type Anything
Start here
The quality of the answer depends entirely on the quality of the question. Before you open Qlik Answers or Qlik Predict, work through this checklist. Everything here comes from your Day 1 and Day 2 training.
Pull out your Decision Brief. What is the actual decision? Not the topic. The decision with options, an owner, and a deadline. Your Decision Brief question is your prompt.
Classify your inquiry mode. Is this Exploratory (what is going on?), Problem-Specific (why did this happen?), Comparative (which option is better?), or Confirmatory (is our assumption true?)? This determines which tool you use.
Check your Data Requirements. What is the must-have data connection? Do you believe it exists? If you are not sure, that becomes your first question: does this linkage exist in our data?
Name your blind spot. What will be missing from any answer the tool gives you? Write it down now, before you see the results. You will need it for your limitation statement later.
Decide what "good" looks like. Before you ask, define what a useful answer would contain. If you cannot describe what a good answer looks like, your question is not sharp enough yet.
The principle
A vague question gets a beautiful, useless answer. Your Decision Brief is your prompt engineering.
Qlik Answers: Writing Effective Prompts
Descriptive & Diagnostic
Qlik Answers searches across your connected data sources and returns a direct answer in plain language. It works best for Exploratory and Problem-Specific questions. It finds specific information and provides summaries. It does not generate its own rankings or aggregations unless they already exist in your data.
The four elements of a sharp prompt
Every Qlik Answers prompt should name: the metric you care about, the comparison (group A vs group B, or before vs after), the time period, and the segment. If any of these four are missing, the prompt is too vague.
Exploratory: "What is going on?"
"What is the [metric] for [group A] vs [group B], broken down by [dimension]?"
Too vague
"How are programmes doing?"
Sharp
"What is the borrowing rate within 30 days for programme attendees vs non-attendees, broken down by programme format?"
Exploratory: Trend detection
"How has [metric] changed since [event or date], segmented by [dimension]?"
Too vague
"Show me borrowing trends"
Sharp
"How has borrowing per visitor changed since January 2025, segmented by branch and patron age group?"
Problem-Specific: "Why did this happen?"
"Of [specific patron segment], what percentage [took a specific action] within [time window]?"
Too vague
"Why is conversion low?"
Sharp
"Of patrons identified as physical-only borrowers, what percentage also made a digital loan within 60 days of omnichannel interaction?"
Problem-Specific: Segment comparison
"Which [category] has the highest/lowest [outcome] over the past [time period]?"
Too vague
"Which programmes work best?"
Sharp
"Which programme format has the highest borrowing conversion rate over the past 12 months, and how does attendance compare across formats?"
Qlik Answers uses meaning, not keywords
Qlik Answers uses semantic search, so conversational phrasing works better than keyword-style queries. Ask it the way you would ask a colleague, not the way you would type into a search engine.
Qlik Predict: Framing Effective Scenarios
Predictive & Prescriptive
Qlik Predict builds predictive models from your data. Before you model anything, you need to answer one question clearly: what exactly are you trying to predict? This is your target variable, and it must be specific enough for the model to learn from historical data.
Define your target variable precisely
"Predict borrowing" is vague. "Predict whether a programme attendee will borrow at least one book within 30 days of attending" is specific. The target variable determines whether the model does classification (yes/no outcome), regression (a number), or time series forecasting (a trend over time).
Comparative: "Which option is better?"
"If we [take action from Option A], what is the projected change in [outcome] over [time period]?"
Too vague
"What will happen if we change programmes?"
Sharp
"If we shift 25% of current programme offerings from author talks to book clubs, what is the projected change in borrowing conversion over 6 months?"
Comparative: Key drivers
"What are the key factors driving [outcome]? Which ones can we influence?"
Too vague
"Why do people stop borrowing?"
Sharp
"What are the key factors driving member lapse (60+ days without a transaction)? Which factors are within NLB's control to change?"
Confirmatory: "Is our assumption valid?"
"If [leading indicator] drops below [threshold], what is the predicted impact on [KPI]?"
Too vague
"Is our retention strategy working?"
Sharp
"If the 60-day lapse count at a branch exceeds 200 per month, what is the predicted impact on that branch's annual retention rate?"
Scenario comparison
"Compare the projected [outcome] for Option A vs Option B vs Option C over [time period]."
NLB example
"Compare the projected borrowing conversion for three scenarios over 6 months: (A) revise 25% of offerings to reading-focused, (B) reallocate 50% of manpower to high-conversion formats, (C) pilot new formats at 3 libraries."
Read the explanation, not just the prediction
Qlik Predict shows which factors drive each prediction. These are your potential leading indicators. When the model says "programme format is the strongest driver of borrowing conversion," that confirms what the upstream chain suggested: format matters more than volume. The explanation is often more valuable than the prediction itself.
Ask in Stages: Build Understanding Iteratively
Qlik Answers
You do not have to get everything in one prompt. Qlik Answers supports follow-up questions that carry context forward. Real analysis works in stages: start broad to orient yourself, then narrow based on what you find. Each answer shapes the next question.
First question — Orient
"What is the borrowing rate by programme format over the past 12 months?"
Start with the broad landscape. See what the data shows before you narrow.
Follow-up — Narrow
"For the top 3 formats by conversion, show me the monthly trend."
You noticed book clubs are highest. Now check whether that is consistent or spiking.
Follow-up — Segment
"Which patron age groups have the highest conversion in book clubs?"
The format works, but for whom? This starts to answer the audience composition question.
Follow-up — Test the assumption
"Of book club attendees who borrowed within 30 days, were they already active borrowers before attending?"
This is the 78% question. Are book clubs creating readers or entertaining existing ones?
Why this works
Each question is shaped by the previous answer. You could not have asked question 4 without the context from questions 1-3. This is how real analysis works: you follow the evidence, not a script.
After You Get an Answer: The Skepticism Checklist
Critical thinking
An answer from Qlik Answers or Qlik Predict is not a conclusion. It is evidence that needs interpretation. Apply these checks before you act on anything the tool shows you.
Check the sources. Qlik Answers shows which data sources it used to build the answer. Review them. Are they the right sources for your question? Are they current?
What can this answer tell me? Name what is within the scope of the data. Example: "This tells me NLB borrowing patterns by format and branch."
What can it NOT tell me? Name what is outside the scope. This comes from your blind spot documentation. Example: "This does not tell me about reading outside NLB (Kindle, Storytel, personal purchases)."
Write the limitation statement. Draft the exact sentence you would include in any recommendation: "This analysis does not account for..."
For Qlik Predict: read the drivers, not just the prediction. Which factors does the model say matter most? Do they align with your upstream chain? Any surprises could be new leading indicators.
Common Mistakes
Avoid these
These are the patterns that produce poor results. For each mistake, the fix is usually about reframing how you think about the question, not about learning new syntax.
If You Get a Bad Answer
Troubleshooting
A bad answer is usually a signal about your question, not a failure of the tool. Work through these steps before giving up.
Rephrase the question. Qlik Answers uses semantic matching. Different phrasing may surface different results. Try asking the same question from a different angle.
Check whether the data connection exists. If Qlik Answers cannot find relevant data, the linkage you assumed may not exist. That is itself a valuable finding. Document it.
Break a complex question into two simpler ones. "How does programme format affect borrowing behaviour across different patron segments?" is doing too much. Ask about format first, then ask about segments.
Check if you are using the right tool. Qlik Answers describes and diagnoses (what is happening, why). Qlik Predict forecasts and models (what will happen if). Asking the wrong tool the right question still gives a bad answer.
Start over with the Decision Brief. Go back to your structured question. Reread your options and success criteria. Often the problem is not the prompt but that the underlying question was never sharp enough.
Remember
Discovering that a question cannot be answered with available data is not a failed analysis. It is a finding. Document it, name the gap, and bring it back to your team. That is the Data Requirements discipline from Day 1 in action.
Your Prompt Workspace
Draft here
Use this space to draft and refine your prompts before typing them into the tools. Copy them when ready.
Your Decision Brief question (from Day 1)
Your Qlik Answers prompt
Follow-up question 1
Follow-up question 2
Your Qlik Predict scenario
Your limitation statement