INDUSTRY
Consumer Packaged Goods and Distribution
Enabling Product Innovation and Omnichannel Experience in CPG
Highlights
- Consumer packaged goods (CPG) organisations are increasingly leveraging artificial intelligence (AI) to improve demand forecasting, personalised promotions, pricing, and inventory optimisation.
- However, many AI initiatives struggle to move beyond pilots due to delays, rework caused by unclear roles, scattered documentation, ad hoc escalations lack of transparency, and limited trust in model outcomes.
- Traditional agile practices, while effective for software delivery, have to evolve to address the exploratory, data-dependent, and continuous learning nature of AI.
- A pragmatic agile-for-AI framework tailored for CPG, that introduces discovery sprints, hypothesis-driven backlogs, business-focused prioritisation, and AI-specific governance, will help organisations deliver faster business value with trust, compliance, and scalability.
On this page
On this pageinpage
Why AI delivery is hard
The consumer packaged goods (CPG) industry operates in an environment of constant change where consumer behaviour shifts with seasons, promotions, and regional preferences
Data is fragmented across point-of-sale systems, loyalty programmes, syndicated sources, and digital channels. Artificial intelligence (AI) models built for one geography or retailer often fail elsewhere due to differences in data distribution, regulations, and buying patterns.
For example, a demand forecasting model that performs well in North America may underperform in Europe if local holidays, regulations, or shopper behaviour are not considered. In this context, speed alone is insufficient; business stakeholders also need transparency, explainability, and confidence in AI-driven decisions.
Challenges in AI delivery
Most CPG organisations follow a linear operating model. Data science teams explore data and build models, which are then handed off to technology teams for validation and deployment. This approach introduces several challenges:
- Delays in validation and sign-off when technology teams deliver late in planning cycles
- Rework necessitated by gaps between synthetic or lab data and real-world retail data
- Limited visibility into progress, dependencies, and sprint commitments
- Skill and tooling gaps between data science and engineering teams
- Stakeholder mistrust due to limited explainability of model outputs
These issues slow time to value and reduce confidence in AI initiatives.
Traditional agile needs to evolve
Traditional agile assumes that requirements can be progressively refined and that outcomes are largely predictable.
AI work violates these assumptions. Model performance depends on data quality, availability, and experimentation. Stakeholders often refine expectations only after seeing early model outputs.
Without explicit space for discovery and learning, teams rush into execution with incomplete understanding, leading to frequent reprioritisation and costly rework.
Reimagining agile for AI
To address these realities, agile must evolve across four dimensions.
These four dimensions are show in the table below:
| Dimension | Traditional agile | Agile for AI in CPG |
| Requirements | Mostly known upfront | Discovered through data exploration (discovery and design sprint) |
| Backlog | Feature and task-based | Hypothesis-driven experiments |
| Success measure | Delivery of features | Business outcomes and impact |
| Governance | Post-delivery checks | Built into DoR (definition of ready) and DoD (definition of done) |
A short, time-boxed discovery sprint before PI (planning increment) or release planning allows teams to explore real retail data, validate assumptions, and assess feasibility. For example, before committing to a festive-season forecasting initiative, teams validate access to points of sale (POS), loyalty, and holiday data, establish a baseline forecast, and identify regional data gaps.
This sprint creates shared understanding across business, data science, and technology teams and significantly reduces downstream surprises (see Table 2).
| Focus area | Typical outcomes |
| Data readiness | Confirmed access to POS, loyalty, and external data |
| Feasibility | Baseline model performance and constraints identified |
| Dependencies | Retailer feeds, labeling, infrastructure needs |
| Planning inputs | Refined hypotheses and realistic PI commitments |
Instead of treating AI work as feature delivery, backlog items are framed as business hypotheses. Each item clearly articulates the expected outcome.
Examples in a CPG context:
- If weather and holiday features are added, then seasonal forecast accuracy will improve by X%
- If customer demographics are included, then basket size prediction accuracy will increase by Y%
- If an explainability dashboard is introduced, then category managers will approve recommendations faster
This approach shifts focus from outputs to measurable business impact.
A weighted shortest job first (WSJF) approach enables CPG leaders to prioritise AI initiatives based on economic impact. Smaller initiatives such as explainability or regional adaptation often deliver faster business value than large model builds.
AI-specific definitions of ready (DoR) and done (DoD) embed governance, quality, and trust into delivery. These include data-readiness checks, performance thresholds, bias and fairness expectations, explainability requirements, and monitoring plans.
How the model works in practice
Cross-functional CPG squads combine business analysts, data scientists, engineers, and product owners with shared accountability for outcomes. Agile ceremonies remain intact but are enhanced with AI-specific checkpoints such as data readiness reviews, explainability demos, and model health discussions.
Unified work boards provide visibility into dependencies such as retailer data feeds and labelling requirements, reducing surprises and escalations.
Business benefits
Adopting this agile-for-AI model delivers tangible benefits.
They include (see Table 3):
- Faster time to value through early validation and reduced rework
- Improved forecast accuracy, promotion return on investment ROI, and inventory optimisation
- Increased stakeholder trust through explainable and transparent AI
- Reduced operational and compliance risk
- Scalable AI delivery across retailers and geographies
| Agile practice | Business value for CPG |
| Hypothesis-driven backlog | Clear ROI and faster decision-making |
| WSJF prioritisation | Maximised economic value from AI investments |
| AI-specific DoR, DoD | Trust, compliance, and reduced rework |
Unlock business value from AI
Agile remains the backbone of successful AI delivery in the CPG industry.
It, however, , must evolve to address AI’s uncertainty, data dependency, and governance needs.
By introducing discovery sprints, hypothesis-driven backlogs, business-focused prioritisation, and AI-aware definitions of readiness and completion, CPG organisations can unlock sustained business value from AI.
This evolved agile approach ensures that AI initiatives are not only fast, but also trusted, responsible, and aligned with business outcomes.