Data is a representation of reality, not reality itself.

DATA & INTELLIGENCE · CONSUMER TECHNOLOGY

Cralgo explores how systems capture behaviour, how people interpret those signals and how organisations turn them into decisions.

ONE SYSTEM OF OBSERVATION INSIDE THE CONSUMER ENVIRONMENT

Reality · what was measured · what was structured

ON THIS PAGE

Data & Intelligence continues a question the consumer environment raises: the system shapes what the customer can see, do and decide — and a platform captures behaviour through the structure it has chosen to observe. What, then, can an organisation actually know from the signals its systems produce?

• The premise
• Data, evidence, intelligence
• Three lenses
• Measurement & metrics
• Definitions & quality
• Customer data
• Evidence & experiments
• Architecture & pipelines
• Inference
• Feedback loops
• Decisions
• When more data may not help
• Outcomes

• Consumer Technology
• Commerce Platforms
• Evidence
• Research

THE PREMISE

Systems observe selectively.

Every dataset is the residue of decisions taken before anyone looked at it. What was measured, what was left out, how an event was defined, how entities were structured, what context was preserved, what was aggregated away, what happened before measurement began — and what behaviour the act of measuring encouraged.

A SYSTEM MAY CAPTURE

• clicks
• searches
• purchases
• returns
• customer attributes
• inventory
• fulfilment
• campaign responses
• service interactions
• product usage

IT DOES NOT AUTOMATICALLY CAPTURE

• intent
• hesitation
• confusion
• trust
• context
• alternatives considered
• organisational causes
• what never entered the system

Recorded behaviour is evidence of something happening — not a complete explanation of why.

Intelligence begins where those signals are interpreted in context: alongside what the organisation was doing at the time, what the customer was trying to achieve, and what the system was never in a position to see.

THREE DISTINCTIONS

Data, evidence and intelligence are not the same thing.

These are distinctions for reading this question clearly. They are not Cralgo pillars, stages or a framework — and the movement between them is never automatic.

1. Data

   What was captured

   A structured record of selected events, attributes and states, shaped by the definitions and instrumentation in place when it was collected.

2. Evidence

   What that observation can legitimately support

   A narrower claim than the dataset appears to offer. Evidence depends on what was comparable, what was observable and what remained unknown.

3. Intelligence

   The judgement applied

   The interpretation of that evidence in context, and the decision about what matters next — including the decision to keep looking.

IN BETWEEN

Between data and intelligence sit interpretation, assumption, uncertainty and organisational context. A dashboard is not automatically evidence. Analytics is not automatically understanding. A model output is not automatically intelligence.

THREE LENSES

Signals are produced by people and read by people.

1. Psychology

   The human — twice over

   How do people appear in the data, and how do people read it?

   • attention, choice and hesitation
   • abandonment and workarounds
   • purchase and repeat behaviour
   • judgement and confidence
   • bias and expectation
   • experience and incentive
   • trust in the numbers themselves

   People produce signals through behaviour, and interpret them through judgement. The analyst, the merchandiser and the executive reading the report are also human.

2. Technology

   The system — primary lens

   What can be captured, structured, joined, retained and inferred?

   • what is instrumented
   • how events are structured
   • what is joined, what stays separate
   • what is retained and for how long
   • what becomes visible, and to whom
   • how quickly the signal arrives
   • what is inferred rather than observed
   • what is aggregated, and so lost

   Technology sets the outer boundary of what any interpretation can reach. Nothing downstream can recover a distinction the system never made.

3. Organisations

   The environment

   Which signals are believed, and which questions get asked?

   • which metrics matter
   • who owns the data
   • which definitions are accepted
   • what behaviour gets rewarded
   • what teams can access
   • who is permitted to act
   • what evidence changes a decision
   • how governance works
   • which questions get asked at all

   The most consequential filter is often the earliest one: the question nobody thought to ask leaves no trace in any dataset.

Data does not enter an organisation neutrally. It enters through definitions, incentives and decisions.

Outcome

The outcome is the decision or behaviour that follows — a change in what is offered, promised, prioritised, bought, built or believed. It emerges from the interaction of the three lenses and is not a fourth lens of its own.

MEASUREMENT & BEHAVIOUR

A measure can describe behaviour and also begin to shape it.

Once a metric becomes important inside an organisation, people optimise around it — sometimes toward the outcome it stands for, sometimes toward the number alone.

• What behaviour does this metric encourage?
• What gets ignored because it is not measured?
• What becomes easier to game?
• Does the metric represent the outcome, or only a proxy for it?
• Which trade-off disappears from view once this number is on the wall?

METRICS AND OUTCOMES

The metric is useful only if we know what it is standing in for.

Conversion, revenue and engagement are not outcomes by default. A number may be a signal, a proxy, an intermediate result or an operational indicator. Which of those it is depends entirely on the question being asked — and ecommerce reporting rarely records the question alongside the figure.

• a signal that something changed
• a proxy for something harder to observe
• an intermediate result on a longer path
• an operational indicator of system health

DEFINITIONS, QUALITY & VISIBILITY

Most disagreements about numbers are not about numbers.

DEFINITIONS

Many data disagreements are organisational disagreements expressed through definitions.

Different teams may answer the same question differently, and each answer may be locally correct.

• What is an active customer?
• What counts as a return?
• What counts as a conversion?
• What is available inventory?
• How is retention measured?
• What is one order?
• What is a successful delivery?

SHARED DEFINITION

Consistency of definition can be useful. It does not make interpretation objective.

A single source of truth is often described as an end state. It is more usefully read as a set of narrower questions: which definitions genuinely need to be shared, which context legitimately differs between teams, what is the source of record for a given fact, and what remains uncertain regardless of where it is stored.

DATA QUALITY

Is the data good enough for the decision being made?

Quality is not cleanliness in the abstract. It is fitness for a specific decision, across completeness, consistency, timeliness, accuracy, traceability and meaning. Data adequate for a weekly trading conversation may be inadequate for a pricing change, and vice versa.

• completeness
• consistency
• timeliness
• accuracy
• traceability
• meaning

DASHBOARDS & REPORTING

A dashboard makes selected signals visible. It does not decide which interpretation is correct.

• Who is this for, and what do they do differently because of it?
• Which decision is it meant to support?
• Which context is missing from the view?
• What happens when two indicators disagree?
• What should trigger an investigation rather than a comment?

CUSTOMER DATA

Different systems hold different traces of the customer relationship.

No system sees the entire person. Commerce, service, marketing, fulfilment and product systems each hold a partial trace, recorded for their own purpose — which is the honest starting point for any customer analytics or customer data platform question.

• What view is actually needed, for which decision?
• What consent exists for this use, and what was it given for?
• Which identity resolution is reliable, and where is it assumed?
• What is being inferred rather than observed?
• What is simply unknowable from these traces?

FIRST-PARTY DATA

First-party data is often described as an advantage. It is more precisely a question of provenance, consent, meaning, context and completeness: where the signal came from, what the person agreed to, what it actually records, what surrounded it, and how much of the relationship it misses.

PERSONALISATION

Data feeds personalisation, and the parent question stands unchanged: when does better information increase relevance, and when does inference reduce trust, autonomy or discovery? Personalisation is one consequence of observation, not the identity of this question.

Consumer Technology

CONSENT & EXPECTATION

• What has the person agreed to?
• What would they reasonably expect from this interaction?
• Which inference exceeds the interaction that produced it?
• What does continued trust require here?

EVIDENCE & COMPARISON

Data can become evidence. It is not evidence merely because it exists.

ATTRIBUTION

Attribution assigns credit. It does not necessarily establish causality.

• What was genuinely observable?
• What was assumed by the model?
• What alternative explanations existed?
• How much confidence does this justify?

The useful conversation is about observability and confidence, not about which attribution model wins.

EXPERIMENTATION

Measurement becomes stronger evidence when the question and comparison are designed before the result is known.

A/B testing and structured experiments can isolate an effect that reporting cannot. The strength of the conclusion still depends on design, context, secondary outcomes and duration — and a statistically detectable difference may be too small to matter operationally.

• the question, fixed before the result
• a comparison that is genuinely comparable
• secondary and unintended outcomes
• practical significance, not only statistical
• what the result would have to be to change a belief

QUANTITATIVE AND QUALITATIVE

Numbers can show where to look. Observation can help explain what happened there.

Intelligence does not arrive only through numerical data. Evidence may include metrics, direct observation, interviews, support and service interactions, recorded behaviour, operational patterns and findings from the field. Excluding the qualitative usually narrows what can be explained, not what can be counted.

Evidence

ARCHITECTURE, PIPELINES & LATENCY

A data model is an organisational assumption expressed technically.

• Which entities exist, and which do not?
• Who defines them, and who can change them?
• What history is retained, and at what grain?
• What can be joined, and what stays permanently separate?
• What does this structure make impossible to ask later?

Architecture

PIPELINES

Can the evidence reach the decision reliably enough?

Data engineering and analytics engineering matter here as reliability questions rather than as a catalogue: lineage, latency, transformation, schema change, observability, failure behaviour and ownership. Sophisticated modelling downstream cannot compensate for a signal that arrives late, silently changes shape, or quietly stops.

• lineage
• latency
• transformation
• schema change
• observability
• reliability
• ownership

LATENCY

Which decision genuinely needs the signal immediately?

Real-time capability is sometimes essential — inventory exposure, fraud, service recovery. It also adds complexity, cost and new failure modes. Latency should fit the decision it serves, not the ambition of the platform.

INFERENCE

Inference is not understanding.

Machine learning and applied AI can classify, predict, rank, recommend, generate and detect patterns at a scale no team can match. None of that is the same as knowing why something is happening.

• What exactly is being inferred?
• With what confidence, and how is that confidence expressed?
• From what data, collected under what conditions?
• What context is absent from the model's view?
• What happens when the inference is wrong?
• Who acts on it, and who remains accountable?

PREDICTION

The prediction becomes part of the environment it is predicting.

• A recommendation changes what gets seen.
• A risk score changes how someone is treated.
• A demand forecast changes what is bought and where it is placed.

Predictive analytics rarely observes a system from outside it. Once acted upon, the prediction is one of the forces producing the next set of signals.

RECOMMENDATION

What does the system choose to make visible?

• relevance
• diversity
• customer autonomy
• commercial incentive
• feedback loops
• trust

A recommendation system is a decision about attention. Optimising it against a single metric usually narrows what customers ever encounter.

WHERE PATTERN BECOMES ASSUMPTION

What pattern is the system learning — and why does that pattern exist?

• what was measured, and what was not
• historical behaviour under different conditions
• groups missing or under-represented in the data
• how outcomes were labelled
• what the optimisation target rewards
• how humans interpret the output
• which organisational incentives shaped the target

FEEDBACK LOOPS

Data does not only describe the environment. Decisions made from it can change the environment.

1. 01 The system observes behaviour
2. 02 A model or a person interprets the signal
3. 03 The organisation acts
4. 04 The customer environment changes
5. 05 New behaviour is produced
6. 06 The system observes again

This is not a framework or a maturity model. It is simply how observation participates in the outcome it is trying to describe — and why a metric that improves after an intervention may be recording the intervention rather than the underlying reality.

Signals · interpretation · decision · new signals

DECISIONS, INCENTIVES & OWNERSHIP

Intelligence is not the dashboard. It is the quality of the decision that follows.

• Who sees the evidence?
• Who interprets it, and with what standing?
• Who is able to act on it?
• What threshold triggers action rather than discussion?
• What evidence would change a previously held belief?
• What happens when the data conflicts with experienced judgement?
• Where does judgement remain necessary, and unavoidable?

CONFLICTING SIGNALS

One team optimises conversion, another margin, another inventory, another delivery speed, another customer satisfaction. A capable system can show all of these at once, honestly and simultaneously.

Which outcome has priority when the signals conflict?

That is an organisational question. No amount of instrumentation answers it.

OWNERSHIP

• Who owns definitions?
• Who owns quality?
• Who can change a metric once teams depend on it?
• Who resolves disagreement between readings?
• Who can access sensitive customer data?
• Who owns the decision, not merely the dataset?

GOVERNANCE

Governance matters when it makes the provenance, meaning and permitted use of data clearer.

Where it produces committees instead of clarity, it has become a cost rather than a capability.

RESTRAINT

When more data may not help.

• The question itself is unclear.
• The wrong thing is being measured, accurately.
• No one owns the decision the data would inform.
• Quality is adequate but trust in the numbers is low.
• Organisational incentives point in conflicting directions.
• The context needed to interpret the signal is missing.
• The metric cannot explain causality, however precise it becomes.
• The team lacks the capability or authority to act on it.
• A single qualitative observation would answer it faster.
• Further instrumentation would add noise rather than evidence.

In several of these situations the useful next step is to narrow the question, not to widen the collection. More data is not itself an outcome.

OUTCOMES

What may change when observation and interpretation improve.

• better decision quality
• clearer evidence
• stronger customer understanding
• reduced ambiguity
• more relevant personalisation
• better operational visibility
• more reliable forecasts
• clearer ownership
• faster learning
• stronger organisational capability

Not all of these, and rarely at once. Which of them is available depends on the question, the systems, the definitions and who is able to act.

EVIDENCE

Did the observation actually change what the organisation believed?

• a decision made differently, and traceably so
• a belief revised in the face of the signal
• a definition agreed across teams
• an intervention whose effect was isolated
• an inference checked against what followed
• a question answered well enough to close

Evidence asks what observations legitimately support about what happened in reality. Data & Intelligence asks how signals are captured, structured, interpreted and used. The two are related and not interchangeable.

Evidence

RESEARCH

Open questions Cralgo keeps returning to.

• How does the act of measuring change the behaviour being measured?
• When does more instrumentation reduce, rather than increase, understanding?
• How do organisations decide which reading of the same data to believe?
• What does delegating judgement to inference change in the people who remain accountable?

Research asks the question. Data may provide signals. Evidence tests what those signals support. Judgement decides what to do next — a way of reading this page, not a company-wide framework.

Research

WHERE THIS SITS

One system of observation among several.

• Consumer Technology

  The systems that shape what the customer can see, do and decide.

• Commerce Platforms

  The platform produces and acts on many consumer signals. This page is about how those signals become observations, interpretations and decisions.

• Product & Customer Experience

  What the person was encountering at the moment a signal was produced.

• Data & Intelligence

  How signals are captured, structured, interpreted and used in decisions.

• Architecture

  Structure determines which questions remain askable later.

• Evidence

  What observations legitimately support about what happened in reality.

• Research

  The questions Cralgo is investigating, and why they matter.

WHEN THE DATA QUESTION BECOMES WORK

Sometimes the question needs building, not only discussing.

Depending on what the organisation is trying to know, and what it is able to act on, this work may be conversational, analytical or deeply technical. Data analytics, business intelligence, analytics engineering and applied machine learning are activities inside that work, never the identity of the company doing it.

• clarify the question
• examine the signals that already exist
• define what would be useful to observe
• review data structures and models
• map competing definitions
• improve instrumentation
• design experiments and comparisons
• build data pipelines
• develop analytics and reporting
• create decision-support systems
• apply AI where it is justified
• observe whether decisions actually changed
• revisit the question

Which of these is appropriate is a conclusion, not a starting position. Often the first honest step reduces what needs to be built.

Work with Cralgo

Data records part of what happened.

The harder question is what the organisation should believe because of it.

Explore the evidence Explore Consumer Technology Work with Cralgo