Rebuilding a CPG Innovation Pipeline — From 12-Week Decision Cycles to 3, Powered by LLMs
How Xmartlabs cut concept-to-decision time by 75% for a multinational CPG brand by replacing sequential validation with AI-assisted parallel workflows.
3 weeks
Discovery & architecture
6 weeks
Market intelligence platform
7 weeks
AI-assisted validation engine
Context
Staying ahead in a market where consumer trends shift in weeks
A multinational consumer goods company with 200+ SKUs across Latin American markets set out to accelerate how they brought new products to life. Their innovation cycle — from opportunity identification to launch — took 8 to 14 months, a timeline that was increasingly difficult to sustain as consumer trends began shifting in weeks.
The innovation team worked with market research reports, cross-departmental spreadsheets, and weekly alignment meetings — tools that had supported their process but were becoming a bottleneck as the pace of the market accelerated. Their goal was clear: connect market data, consumer feedback, and production capabilities into a more responsive workflow. Not “more AI” as a buzzword — but targeted technology applied to specific operational constraints.
Challenge
Three constraints we designed around together
01
Data fragmentation
Trend, sales, and consumer feedback were distributed across 6 separate systems — each maintained and trusted by different teams. The opportunity was to connect these signals in a way that respected existing workflows and avoided costly migrations.
02
Validation bottlenecks
Multi-department validation is inherently complex in a large organization. With marketing, supply chain, and regulatory teams each running multiple review rounds, the goal was to restructure coordination — not eliminate rigor — so teams could validate in parallel without losing accountability.
03
No real-time visibility
Keeping stakeholders aligned required hours of manual report compilation per update cycle. Together we defined what real-time visibility should look like — surfacing the right information to the right people automatically, without adding operational burden to the team.
Approach
Integrate before you migrate
We worked in three phases with a lean, cross-functional team of 6. The biggest risk was not technical — it was adoption. Previous tool rollouts had burned the supply chain team, so we dedicated 20% of the project to hands-on workshops that co-designed workflows with them instead of training them on a finished tool.
Phase 1 · 3 weeks
Discovery & architecture
Mapped existing workflows through interviews with 12 stakeholders across 4 departments. 60% of cycle time was lost in handoffs, not in the actual work. We designed an event-driven architecture that connected existing data sources without requiring migration, reducing change resistance.
12
Stakeholders
4
Departments
0
Migrations
Phase 2 · 6 weeks
Market intelligence platform
Built a web platform integrating consumer trend data (social media, search trends, panel data) into a unified dashboard. The system generates dynamic consumer personas and country profiles for each LATAM market, so opportunity reports reflect regional nuance — not a generic Latin America view.
4+
Data sources
LATAM
Per-country
1
Unified UI
Phase 3 · 7 weeks
AI-assisted validation engine
Developed an automated workflow where each concept runs through parallel (not sequential) validations with every department. Decision-makers receive an AI-generated executive brief. Every LLM decision is fully traced via Langfuse — regulatory teams can audit the complete reasoning chain.
4×
Parallel lanes
100%
Traced
1
Exec brief
Results
75% faster cycles and 4x more concepts evaluated
The rebuilt innovation pipeline delivered measurable impact across every key dimension — not just speed, but throughput, operational load, and cross-department alignment.
12 → 3 wks
Concept to go/no-go decision time
From 12 weeks down to 3
4x
Concepts evaluated per quarter
From 8–10 up to 35–40
20 → 2 hrs
Weekly report compilation
Time previously spent on manual updates
4 depts
With real-time pipeline visibility
Previously limited to a single team
“What stands out about Xmartlabs is their AI knowledge and technical expertise. Their structured and collaborative approach was key to a smooth workflow.”
Project Lead
AI Consulting Company — via Clutch
Technology
How we built it
LLMs applied where they create the most value: synthesizing fragmented data and automating the waiting time between human decisions.
AI / LLM
LangGraph with checkpointing for workflow orchestration; hybrid search (semantic + full-text) over a pgvector knowledge base for contextual trend synthesis.
Architecture
Decoupled ETL pipeline feeding a central data lake; pgvector vector store with embeddings + retrieval (RAG) built on the company’s existing knowledge base.
Validation Engine
Parallel multi-department validation with automated commercial, technical, and regulatory scoring feeding an AI-generated executive brief.
Observability
Langfuse for full LLM traceability — every decision is logged and auditable, supporting regulatory review and compliance workflows.
Delivery Model
Lean cross-functional team of 6, with 20% of project time dedicated to adoption workshops co-designed with the supply chain team.
