If you are a CPG operator in India and you have sat through five vendor pitches in the last quarter, you know the pattern. Each tool promises end-to-end transformation. Each comes with a deck. None of them quite fit your stack, your data, or your team. By the time the procurement cycle ends, the people who briefed it have moved on.
The honest answer is that the AI tool landscape for CPG is layered. Each layer has two or three credible options. Most engagements that succeed buy the right tool at the right layer and integrate it carefully, rather than picking one platform and stretching it across everything.
The stack, layer by layer
The shape is clear once you draw it. Source systems at the bottom. Data and BI on top of that. Planning above that. Order and operations above that. Front of house at the top, where customers and the market touch the business. Each layer needs a different kind of AI, and confusing the layers is the most common reason engagements fail.
Layer by layer, what to consider
Source systems
SAP, Oracle, Tally, ERPs, distributor management systems, POS feeds. Nothing AI here. The tools at this layer are what they are. What matters is how clean the data is on the way out, because every layer above depends on it. The most common reason an AI build stalls is bad data at this layer.
Data and BI
This layer is solved. Power BI for Microsoft-heavy shops, Tableau where the visualisation budget is large and the visual quality matters, Looker where the data lives in BigQuery and the team is modelling-disciplined, Metabase for a fast cheap start. The full BI tool comparison is in a separate article.
The AI layer here is mostly auto-generated narratives on top of dashboards and the natural-language query interfaces. Useful, increasingly standard, not transformative on its own.
Planning
Demand forecasting is where most CPG companies should focus the AI build first. The mature platforms (o9, ToolsGroup, Anaplan) are real but expensive, and require multi-month implementations. The in-house Python build with a senior team typically ships in six weeks at one-fifth the licence cost, and the integration is yours forever. We have covered the build approach in detail here.
The decision shorthand: if you have over five thousand crore revenue and a planning team of fifteen plus, buy the platform. Below that, build with a senior boutique.
Order and operations
The AI use cases here are the highest-ROI for most Indian CPG businesses. Order intake from distributor WhatsApp messages and emails. Collections and reconciliation. Customer service across channels. The agents that handle these jobs run on Claude or OpenAI for the language layer, with Tally or SAP APIs for the writeback. Few vendors sell this end-to-end at the price point Indian mid-market can stomach. This is the build layer.
The good news is that the underlying technology is now cheap, fast, and reliable enough that the build is in weeks, not quarters. The hard part is the integration with whatever ERP and accounting system you have, which a generic vendor cannot do well.
Front of house
Customer chat AI, content generation, lead generation. The buy layer. Yellow.ai for Indian customer chat is mature. Apollo and similar platforms for lead generation. Hypotenuse.ai or others for marketing content. Drift and Intercom for inbound chat on the website.
The trap to avoid is buying these tools without an opinion on the workflow they slot into. The tool is usually fine. The flow that surrounds it determines whether it earns its keep.
The build, buy, partner decision per use case
The shorthand: where the data shape is unique to your business (forecasting, planning, order intake), build with a senior partner. Where the use case is industry-generic (customer chat, lead generation), buy. Where the question is strategic and one-off (trade promotion ROI, pricing strategy), partner with a consulting firm.
The mistakes show up at the corners. Buying a generic forecasting tool that ignores your channel mix specifics. Trying to build a customer chat system from scratch when Yellow.ai already exists. Hiring a consulting firm to run an ongoing campaign that an agency should run instead.
What we are NOT recommending
Three categories deserve scepticism.
End-to-end "AI platforms" that promise to cover the whole stack. They overshoot at one layer and underdeliver at three. The reference customers are usually pilot projects that have not generalised.
RPA tools for AI use cases. RPA is fine for screen-scraping legacy systems. It is the wrong tool for anything language-heavy. A modern LLM-based agent handles language work that RPA cannot.
Generic Indian SaaS pitched as "AI for CPG". Many such offerings exist, most are reskinned forecasting or BI tools with a marketing layer added. Look at what the underlying technology is, not what the website claims.
How to actually pick
For each use case in your roadmap, ask three questions:
- How custom is the data shape? If it is generic, buy. If it is unique to your business, build.
- How critical is the integration to your existing stack? If it is critical and complex (your ERP is custom, your DMS is bespoke), build. If it is straightforward and well-documented, buy.
- How long do you want to own the capability? If for five years, build. If for one campaign, buy or partner.
What to do this week
Pick the three highest-priority AI use cases on your roadmap. Run them through the build, buy, partner matrix above. Honest answers usually shift the procurement direction. If you want a no-pitch conversation about your specific roadmap with someone who has built each of these layers many times, talk to us.