Ask open questions
Ask “Why did margin fall in the north?” or “Where is growth stalling?”, not a query. Stack interprets the question against the data.
The question that matters most is rarely one a dashboard was built to answer.
An analyst that interrogates your data
Ask your data anything. Get evidence back.
Ask an open question like “What are the current barriers to growth?” Stack runs the statistics, forms hypotheses, tests every one, and hands you an evidenced PowerPoint.
Every slide has a receipt. And you see what it proved wrong, not just what it proved right.
Question
“What are the current barriers to growth?”
- supported
H1Retention fell because more of the base is new customers
Mix-shift decomposition: most of the decline is mix, not rate.
- rejected
H2Established customers are churning faster
Within-cohort retention barely moved quarter on quarter.
- weakly supported
H3A regional problem is driving churn
Region matters only after splitting by tenure.
- supported
H3aNew customers in the North churn faster
Follow-up: the regional effect is concentrated in new North customers.
- rejected
H4Worse support is causing churn
A weak correlation that disappears once tenure is controlled for.
- rejected
H5The change is seasonal
No matching pattern in the same quarter of earlier years.
Sound familiar?
You have the data. You still don’t have the answer.
Leadership asks why. The answer joins the analyst queue and arrives in three weeks.
The dashboard shows what moved. It never says why.
Everyone has a theory, and the most senior one wins.
The board pack gets built the night before, from whatever charts exist.
How it works
From an open question to an evidenced deck.
01
Connect your data
Point Stack at your warehouse and analytical sources. Read-only, with a data classification on every source.
02
Ask an open question
“What are the current barriers to growth?” In your own words. Stack works out what it means against the data it can see.
03
Run the maths
A statistics engine profiles the data and looks for trends, correlations, segments, anomalies and shifts in mix.
04
Test every idea
Stack forms hypotheses, plans how to prove or disprove each one, runs the tests, and follows up on what survives.
05
Get the deck
What survives becomes an evidenced PowerPoint: findings, charts and narrative, every slide linked to its test.
The average customer doesn’t exist
The headline number moved. Nobody behaved differently.
Retention fell by more than three points. The obvious story is that customers got less loyal. Split by tenure and that story falls apart: established customers and new customers each behaved almost exactly as before.
What changed is who the customers are. More of the base is new, and new customers always churn more. Stack separates mix from rate, so you fix the real problem instead of the one the average suggests.
| Retention | Q1Q2 |
|---|---|
| All customers−3.3 pts | 87.2% 83.9% |
| Established customers−0.4 pts | 91.0% 90.6% |
| New customers−1.0 pts | 72.0% 71.0% |
−3.3 pts
Overall retention, Q1 to Q2. On the dashboard, customers look less loyal.
81%
Of the fall is mix: more of the base is new customers, who always churn more.
Illustrative · fictional Harborlight Analytics demo data · bars start at 60%
Noise isn’t proof
Half the value is what it proves wrong.
Test twenty ideas and one will look significant by luck. Most tools would put it on a slide. Stack corrects for the number of tests it runs, reports effect sizes, and marks a hypothesis rejected or inconclusive when the evidence isn’t there.
That means the most senior person’s pet theory gets tested, not argued about.
Tested one at a time
Product colour “predicts” churn, p = 0.031.
Corrected for 20 tests
After Holm correction, nothing is significant. Colour was chance.
What it does
It works like your best analyst. It just never runs out of time.
Connect your sources
Connect warehouses and analytical environments over MCP. Credentials are encrypted and every source carries a data classification.
It starts where your data already lives. No new warehouse, no spreadsheet exports.
A real statistics engine
Trends, seasonality, correlation, regression, t-tests, ANOVA, chi-square, effect sizes, cohorts, anomalies, mix-shift and contribution analysis.
The maths is computed, not guessed by a language model. You can check every number.
Hypotheses, generated
Stack turns your question into candidate explanations: a segment, a channel, a price change, a seasonal effect, a shift in who your customers are.
It raises the explanations nobody in the room would have, including the uncomfortable ones.
Tests designed to disprove
For each hypothesis, Stack plans the analysis that would support or undermine it, then runs it.
Like a good analyst, it decides what would prove an idea wrong before it looks.
Honest verdicts
Every hypothesis ends supported, weakly supported, rejected or inconclusive, with its confidence and the evidence behind it.
You see what isn’t true as well as what is. Pet theories get tested, not argued.
It keeps digging
Explanations that survive the first pass spawn sharper follow-up hypotheses. Weak ones are dropped.
The second question is usually where the answer is.
An evidenced PowerPoint
Stack chooses the findings, charts and order so the output is an argument, not a pile of charts. Every slide links back to the test behind it.
The deck is what people take into the meeting. Every claim in it has a receipt.
Every step on the record
Watch the investigation as it runs. Afterwards, open the methods view to see every query, tool and result.
No black box. Your analysts can audit it, so your executives can trust it.
The deliverable
Every slide has a receipt.
Stack picks the findings that survived, the charts that show them best, and the order that makes an argument. Data-quality caveats come first, not in a footnote.
Every slide links to the hypothesis, the test and the data behind it. When the board asks “how do we know?”, the answer is one click away.
Finding 1 of 3
Retention fell because the mix changed, not because customers got worse
Share of customersQ1Q2
Established customersNew customers
Overall retention −3.3 pts. 81% of that is a shift toward new customers.
Evidence: H1 · supportedMix-shift decompositionmonthly_retention
Questions people ask
The questions your dashboard was never built to answer.
Leadership
“What are the current barriers to growth?”
Finance
“Why did margin fall, and how much of it is mix?”
Commercial
“Why is win rate falling in mid-market deals?”
Customers
“Why did retention fall in Q2?”
Marketing
“Which channels bring customers who stay?”
Operations
“Where are the bottlenecks in fulfilment?”
Your data stays where it is
The maths is real. Your records stay put.
Profiles, not raw rows
The language model sees dataset profiles and statistical results. Your raw records never go into a prompt.
Read-only by design
Queries are limited to read-only statements, one at a time, with row limits. Stack can’t change or delete your data.
Classified and encrypted
Every source carries a classification: public, internal, confidential or restricted. Credentials are encrypted at rest.
Your team, your access
Admins connect sources, analysts run investigations, viewers read the results.
What it won’t do
It won’t tell you what you want to hear.
An analysis is only as good as the data and the people reading it. So Stack shows its working.
- Hypotheses can be wrong. Stack is built to test them, not to declare them true.
- Data quality shapes the answer. Stack reports missing fields and broken joins before the charts.
- Where the data can’t support a claim, the deck says so.
- It drafts, people decide. Every analysis is a draft for professional judgement.
Request a demo
Forty-five minutes. One open question. One evidenced deck.
Stack is yours to run. Your team connects the data and asks the questions. The demo shows you how.
01
Your question
Tell us what your leadership keeps asking that takes too long to answer.
02
Ask it
Watch Stack take an open question on demo data: profiling, hypothesising, testing.
03
The verdicts
Supported, rejected, inconclusive, and the statistics behind each one.
04
The deck
The evidenced presentation, and a click from any slide to its evidence.
05
Your data, safely
How Stack connects read-only, and why raw rows never reach the model.
06
Your questions
Ask anything, then decide whether a pilot on your own data makes sense.
What would you ask your data if the answer came back with evidence?
See Stack take an open question, test every explanation against your data, and turn what survives into an evidenced deck.