Solutions Overview

The three steps from data to action
— perfected by ORCANA

HOW ORCANA WORKS
Plan Generation
Analytics Execution
Insight or Action

Intent Detection: Meticulously designed custom framework

Intent Detection: Meticulously designed custom framework

User query or Data trigger

Scope Detection Agent

(Mini fine-tuned LLM)

Trained on existing client data-sources and Orcana’s analytics capabilities

Entity Extraction Agent

(Medium fine-tuned LLM)

Trained on data-augmented user-queries and custom entities

Clarification Agent

(Small LLM)

Generalist conversational LLM to request customer clarification where required

Plan Generation Agent

(Large fine-tuned LLM)

Highly trained on plan generation structured output tasks; contextually aware of data-sources and analytics capabilities

Plan Validation Agent

(Medium fine-tuned LLM)

Validates plan against data-sources and capabilities, acts as a final fail-safe against any sub-optimal responses

Data Staging 'Plan' & 'Analytics Plan'

Encrypted result generation with zero data-exposure

Encrypted result generation with zero data-exposure

‘Data Staging Plan’ & ‘ Analytics Plan’

Data quality checker module

Orcana sifts through data processing logs to bad data's analysis

Data-staging module

Extract and load preset data or execute preset queries along with custom inclusion and exclusion criteria

Hypothesis generation module

Analyze changes to business metrics across segments and time-periods to execute causal inference models

Analytics/Data-science modules

Identify and stage custom data-science modules such as segmentation; supervised and unsupervised models on structured & unstructured data

Plan executor

A configurable function that extracts data and executes advanced data-science modules

Analytics Output

Result generation: Delivers insights and actions with zero data-exposure

Result generation: Delivers insights and actions with zero data-exposure

Analytics
output

Named entity encryption

(Python)

Orcana anonymizes all named entities by generating encryption keys (Products, Accounts, HCPs, Reps, etc.)

Ingest encrypted context

(Python)

Access org. level context, instructions,, and business rules using encryption keys from step '1'

Results generation

(LLM)

Engage a fine-tuned LLM to generate a reliable response or recommend and action

Query <> Result fit assessment

Assess the fit between query (encrytped) and results generated

Named entity de-encryption

(Python)

Deanonymize the results with all original named entities

Insight OR
Action
Product Architecture

HYBRID AI ARCHITECTURE DELIVERS HIGH ACCURACY AND REPEATABILITY

Application Layer Most competitors' stack Low accuracy & Repeatability; High hallucination End-to-end GenAI architecture * RAG: Retrieval-Augmented Generation Result generation RAG* based context ingestion Pre-trained reasoning Code-generation on-the-fly 99% X 99% X 80% X 70% Orcana's Stack High accuracy & repeatability; No-to-low hallucination Hybrid AI (Gen AI + Targeted AI) architecture * GenAI for unstructured reasoning and agentic Result Generation (GenAI*) Configurable business context 99% X 100% X 100% X 100% OUTPUT Insight or action INPUT Data trigger or user uestion Advanced data-science accuracy Overall ~ 50% ~ 99% VS Modular logic-based reasoning Parameterized python zrepo Intent detection (GenAI*) End-to-end GenAI engine Proprietary Targeted AI engine

Addresses enterprise-wide needs

ACCURACY, ACTIONABILITY, AND ABIDANCE

Accuracy and workflow enablement for Business teams. Configurability and scalability for Tech teams. Compliance and security for Legal teams. All in one AI-native platform.

1
Hybrid AI Architecture

Where accurate analytics is powered by python-based inhouse engine

2
Central AI Platform

with hyper-verticalized agents that enable scale and interoperatbility

3
Zero data exposure to LLMs

Is ensured through reliable non-GenAI analytics and two-way encryption

DELIVERS CONTEXTUAL AND REPEATABLE
INSIGHTS AND ACTIONS

DELIVERS CONTEXTUAL AND REPEATABLE INSIGHTS AND ACTIONS

01

Structured data

  • Sales data
  • Call and Mktg. activity
  • HCP Profile
  • Call-plan
02

Therapeutic Ontology

  • Pubmed docs (NIH)
  • drugs.com for competitor
  • Product label for our brand
  • Internal clinical-research docs
03

Unstructured HCP intel

  • Speaker event (Topic+co ntent)
  • Rep activity (key-topics)
  • Marketing engagement
  • Product website's click data
Metric Map

METRICMAP™ ENABLES REPEATABLE AND
EXPLAINABLE ANALYSIS, AT SCALE

Illustration of how Orcana would systematically traverse and analyze data for a user-question, say,
'Why did Brand TRx for last month decrease?'

Brand TRx
MoM analysis at overall level & by specialty, decile, tiers, sub-national, behavioural segments, etc
TRx / Writer
# Customers
NBx / Writer
Cont. Rx / Writer
New brands writers
Old brands writers
New to therapy
Switch-in
Refills
Switch-out
New market writers
Old market writers
Discontinued writers
Continuing writers
Line progression
In-line switches
Line progression
In-line switches
Product mix
patient mix
Product mix
patient mix
Orcana 3D Analytics Chart
Breadth of analytics offered

ORCANA DELIVERS MID-TO-HIGH IMPACT ANALYSIS, ON DEMAND

IMPACT
LOW
MEDIUM
CUSTOM SUMMARY
TREND-BREAK ANALYSIS
AD HOC ANALYSIS
COHORT ANALYSIS
HYPOTHESIS TESTING
ROOT-CAUSE ANALYSIS
OPPORTUNITY ANALYSIS
LOW
MEDIUM
HIGH