01
ValidateConfirm the signal
Definition parityCalendar alignmentCoverage checkRestatement check
Like-for-like
reconciliation
reconciliation
Investigation design system / FMCG
Investigations, engineered.
Your analytics stack has everything. Except the investigation logic that turns a business question into a rigorous route through evidence.
Designs the algorithm
Uses your data reality
States evidence limits
Business questionMetricsCutsHypothesesDataMethodsToolsEvidence
Product demonstration / 01
Why did value sales fall vs last year?
A web application that designs the complete investigation around the question, the company and what can actually be observed.
Investigation plan / NC–017
02
LocateFind concentration
Retailer contributionRegional patternSKU concentrationWeekly onset
Contribution
ranking
ranking
03
DecomposeSeparate drivers
Volume effectPrice/mix effectDistribution proxyPromotion split
Bridge +
variance tree
variance tree
04
ExplainTest hypotheses
Lost distributionPromotion gapPack shiftCategory pressure
Competing
explanations
explanations
05
EstablishState evidence level
TriangulateSeek disconfirmationRecord limitsDefine next evidence
Evidence
assessment
assessment
Hypothesis routes
ObservationLoss concentrates in 12 SKUs at two retailers
Compare weighted distribution proxy with value contribution by week.
Evidence levelAssociative
The proxy narrows the route; it does not establish the mechanism.
Required nextNeeds store-level shelf availability
Do not advance the claim beyond the available evidence.
FindingSupportStateLimit
82% of decline sits in two retailersRetailer × SKU contribution viewDescriptiveDoes not explain why
Distribution proxy fell before salesFour-week sequence alignmentAssociativeStore evidence missing
Availability drove the declineNo direct shelf observationMechanism not establishedCollect store-level evidence
Operating sequence / 02
Ask. Describe your reality. Get the investigation.
Define the decision
Scope the metric, comparison, entity, period and decision the investigation must support.
Declare constraints
Describe data availability, dimensions, tools, team capability and evidence gaps.
Receive the plan
Follow it. Build from it. Orchestrate with it. The algorithm remains inspectable.
One investigation model / 03
One logic. Three ways to use it.
The product is the investigation design, not a replacement dashboard.
01
Analyst guide
An ordered route through definitions, breakdowns, methods, hypotheses and evidence checks.
Follow now02
BI blueprint
A precise specification for the views, measures and interactions a team should build.
Build next03
Future automation
A governed investigation sequence that can later coordinate tools and agents.
Orchestrate laterFrom fragments to system / 04
The missing layer is not another chart.
Knowledge scattered across people and files
Analyst memory
Old presentation decks
Dashboard habits
Method documents
Unstated assumptions
A company-specific investigation algorithm
Ordered investigation phases
Defined methods and views
Competing hypotheses
Data and capability constraints
Evidence limits
Driver separation / 05
Decomposition tells you what moved. Not why.
A bridge locates the mathematical contribution. POLYLENS then designs the routes needed to examine plausible mechanisms.
Value change=Volume effect+Price / mix effect
Contribution viewIllustrative structure
Volume effect
−6.8
Price effect
+1.9
Mix effect
−3.5
Portfolio entry
+0.6
Portfolio exit
−0.6
Interpretation boundary: the bridge describes contribution. It does not prove a root mechanism.
Competing routes / 06
The answer is rarely one cause.
A strong plan keeps several explanations alive, specifies what would support each one and records where evidence ends.
Value sales
−8.4% vs LY
−8.4% vs LY
Distribution loss
Locate retailer and SKU concentration; compare the proxy sequence with sales.
Promotion gap
Separate base and promoted sales; compare event timing and depth.
Pack / mix shift
Test whether the portfolio moved towards lower-value packs or channels.
Category pressure
Compare brand change with category and retailer context where available.
Competitor activity
Requires competitor media or promotion evidence that is not currently available.
Shelf availability
Needs store-level shelf availability before the mechanism can be assessed.
Localisation / 07
Not outliers. Patterns that only exist as a group.
Top-down navigation follows the business hierarchy. Bottom-up localisation regroups fine cells by similar behaviour. They answer different questions.
Procedure comparisonWhere is the decline concentrated?
Declared hierarchy
National grocery / −8.4%
Retailer A / −12.1%
North−15.4
Central−10.8
South−4.1
Retailer B / −9.7%
North−3.6
Central−14.2
South−8.3
Fine cells
48 visible / illustrativeSKU × retailer × region × week
Persistent
declinePromo
gapsDistribution
signalLate
onsetMix
shiftStable?
declinePromo
gapsDistribution
signalLate
onsetMix
shiftStable?
Top-downTraverse known dimensions and quantify where contribution accumulates.
Bottom-upRegroup fine cells by shared trajectories, then interpret each cluster against business context.
Reasoning architecture / 08
The analyst, made inspectable.
The analyst is not one magic model. It is an ordered set of reasoning operations with explicit hand-offs and limits.
Frame
Metric, comparison, scope, decision.
Validate
Definitions, periods, coverage, quality.
Locate
Hierarchies, contribution, onset.
Decompose
Drivers, bridges, relationships.
Challenge
Hypotheses, alternatives, disconfirmation.
Bound
Evidence level, gaps, next collection.
Knowledge provenance / 09
Not generated on the fly. Built from established practice and explicit research.
Every investigation pattern should show where it comes from, how it is adapted and what claim it can support.
Method provenance ledgerTraceable / reviewable
Operating knowledge
Senior analyst workflows
Company playbooks
Decision rituals
Established methods
Variance decomposition
Diagnostic reasoning
Evidence assessment
Reusable modules
Validate → locate
Driver bridges
Hypothesis routes
Company adaptation
Available dimensions
Data gaps
Tool capability
Bounded plan
Ordered algorithm
Evidence limits
Next-data requests
SourceMethodAdaptationBounded claim
Designed tension / 10
Standardised where rigour matters. Personalised where context differs.
What should not depend on the analyst
01
Validate definitions before interpreting change.
02
Separate mathematical contribution from causal explanation.
03
Test competing explanations and seek disconfirmation.
04
State evidence limits before recommending action.
What must reflect the company
A
Metric definitions, hierarchies and reporting calendar.
B
Available datasets, detail level and known quality constraints.
C
Analytical tools, team capability and feasible methods.
D
Decision rights, commercial context and next-evidence options.
System first / 11
Before you automate analysis, map the investigation.
Automation becomes more useful when the reasoning sequence, inputs, checks and stopping conditions are explicit first.
Question contract
Define the decision, metric, scope and comparison.
Investigation model
Order methods, views, routes and evidence checks.
Human execution
Guide analysts through a shared, inspectable sequence.
Tool specification
Translate the model into BI and data requirements.
Governed orchestration
Coordinate agents and tools around the approved algorithm.
Early access
Stop reinventing the investigation.
Join the waitlist for POLYLENS. We are working with FMCG teams that want rigorous investigation logic before more dashboards or automation.
POLYLENS is an early prototype, and there’s no signup yet. If the approach is useful to you, write to me and I’ll keep you posted.
No automated data connection is implied. Early access focuses on investigation design.