05 / Products
Stack/Agentic Data Analysis
Ask the question. Let Stack investigate.
Stack connects to your analytical data, forms hypotheses, tests them against the evidence and turns its findings into a structured analysis — including charts, narrative and a finished presentation.
Investigation
A question branches into hypotheses, then converges on evidence.
The problem
Conventional analytics often requires you to know in advance which metric matters, which dimensions to inspect, which dashboard to open and which query to run. Open-ended questions — why something moved, what appears to be driving a change, which segment is behaving differently — still need an experienced analyst to perform many iterative steps by hand.
Stack is designed to automate and accelerate that investigative workflow. It does not replace the judgement of a professional analyst. It takes the first long walk through the data so that your judgement can start from evidence rather than from a blank query.
It does not simply translate that question into one SQL query. It behaves more like an analytical team: interpret the question, inspect the available data, form hypotheses, test those that can be tested, compare evidence, reject weak explanations, refine the analysis, and decide how the findings should be communicated.
Traditional BI asks what the dashboard shows. Stack asks what might be happening, and what evidence supports that reading. The result is an investigative workflow, not a guaranteed account of the truth.
How Stack works
From an open-ended question to a board-ready analysis.
The defining feature is the workflow. Stack interprets the question, inspects available data, forms hypotheses, tests those that can be tested, compares evidence, rejects weak explanations, refines the analysis, and then decides how the findings should be shown.
Figure · How Stack investigates
- 01
Question
- 02
Understand data
- 03
Form hypotheses
- 04
Test hypotheses
- 05
Refine analysis
- 06
Select evidence
- 07
Visualise
- 08
Build narrative
- 09
Generate presentation
Agentic hypothesis testing
More than a query. More than a chart.
Stack does not only retrieve rows or generate a visualisation. It considers multiple candidate explanations, investigates them, and uses evidence to strengthen or weaken those explanations. That produces an iterative analytical process: generation, testing, follow-up, comparison, segmentation, trend analysis and — where the data justifies it — statistical support.
This is not autonomous scientific proof. Hypotheses can be wrong. The value is a structured investigation that you can read, challenge and continue.
Figure · Question to conclusion
Question
Why did the metric move?
H1
Retention
H2
Price
H3
Mix
H4
Cost
H5
Season
Evidence
Surviving explanations, discarded ones, and the cuts of the data that justified each reading.
Deliverable
A narrative, charts and a sequence: a presentation a person can review.
Questions Stack is built for
- 01
Why has customer retention fallen this quarter?
- 02
What is driving margin erosion?
- 03
Which factors appear to predict churn?
- 04
Why is one region outperforming another?
- 05
Where are the biggest operational inefficiencies?
- 06
What explains the change in conversion rate?
- 07
Which customer segments are behaving differently?
- 08
What changed before the decline in support satisfaction?
- 09
What are the most important patterns in this data that management should know about?
Data access
Pointed at the analytical environment you already have.
Stack is designed to work with enterprise data sources such as data lakes, warehouses, analytical databases, structured datasets and business intelligence environments. Integrations follow the estate you already have. We do not publish a vendor list the implementation does not yet warrant.
Presentation intelligence
The analysis does not stop at the finding.
Stack determines which findings matter, which charts communicate them, how to sequence the argument, and what supporting evidence to include. It can then generate a PowerPoint. That is an analytical deliverable — a structured reading of the evidence — rather than automatic slide decoration.
Example journey
Why did profitability fall in Q2?
Illustrative example — not a customer result
Investigated
- Revenue
- Customer mix
- Region
- Product mix
- Acquisition costs
- Discounts
- Operational expenditure
- Retention
- Pricing
- Seasonality
Illustrative narrative
“Profitability declined primarily due to margin compression in two product segments, compounded by increased acquisition costs in one region.”
What the investigation can include
- 01
Hypothesis generation
Turn an open-ended question into a set of candidate explanations that can be investigated.
- 02
Hypothesis testing
Run analyses that strengthen or weaken those explanations against the available data.
- 03
Follow-up analysis
Pursue the lines of enquiry that survive the first pass, and drop those that do not.
- 04
Comparison
Place segments, periods, regions and products next to each other so differences are visible.
- 05
Statistical evidence
Use quantitative support where the data and the question justify it — without dressing noise as proof.
- 06
Anomalies and trends
Surface breaks, outliers and movements that a fixed dashboard may never be asked to show.
- 07
Segmentation
Separate the groups that actually moved from the averages that hide them.
- 08
Presentation intelligence
Choose findings, charts and sequence so the output is an argument a person can follow, not a pile of slides.
Use cases
Classes of question the investigation is built for.
01
Commercial performance
The problem
Revenue, margin and mix questions are rarely answered by a single chart. The useful question is what changed, where, and what else moved with it.
How it is addressed
Stack can investigate performance across products, regions, channels and time, then assemble the evidence into a narrative your commercial team can review.
02
Customer behaviour
The problem
Retention, churn and conversion shifts often have several plausible causes. Dashboards show the metric; they do not run the investigation.
How it is addressed
Point Stack at the relevant customer and event data and ask what appears to have changed. You can then review the hypotheses it tested and the ones it discarded.
03
Operational analysis
The problem
Inefficiency is easy to feel and hard to locate. The work is finding where time, cost or failure actually concentrates.
How it is addressed
Stack can look across operational measures, compare units and time periods, and surface the patterns that deserve a closer look from your team.
04
Marketing effectiveness
The problem
Campaign, channel and creative questions are usually asked after the fact, against incomplete attribution and too many cuts of the data.
How it is addressed
Stack investigates the cuts that the question implies, reports what the data can support, and is clear where the evidence is too thin.
05
Financial analysis
The problem
A movement in profitability has a list of usual suspects — price, mix, cost, volume — and the work is deciding which of them the data actually supports.
How it is addressed
Stack can walk that list as an investigation, then produce a reviewable pack rather than a single unexplained variance line.
06
Product analytics
The problem
Product teams need to know which behaviours changed, not only which dashboard tile went red.
How it is addressed
Ask what changed before a drop in activation, engagement or conversion, and review the evidence Stack assembled.
07
Executive reporting
The problem
Leadership packs take time because someone has to decide what matters, what to show, and in what order.
How it is addressed
Stack can draft that argument from the data — findings, charts, sequence — for someone on your team to accept, edit or reject.
08
Investor reporting
The problem
Explaining a period to investors or a board requires a coherent account, not a warehouse of charts.
How it is addressed
Stack can assemble a structured reading of the period. The reading remains a draft for professional review, not a statement of record on its own.
Human review
Support for professional analysis, rather than a substitute for it.
Generated analyses should be reviewable, and traceable to supporting data where the technical path allows. Stack is not an unquestionable autonomous analyst. It is a system that can do the investigative labour and leave you able to see how the argument was built.
Limits worth being clear about
Investigate and test. Do not treat the output as a statement of record.
- 01
Hypotheses can be wrong. The system is designed to test them, not to declare them true.
- 02
Data quality matters. Missing fields, biased samples and broken joins will shape the result.
- 03
Findings require evidence. Where the data cannot support a claim, the analysis should say so.
- 04
Human review remains valuable. Generated analyses are drafts for professional judgement.
Enterprise deployment
These products are delivered as engagements rather than self-serve licences. We start with your problem, your data, and the systems you already run.