The method
First make Kajaria AI-ready. Then make it AI-native. Do not wait for one to finish before the other starts.
Most AI programmes fail in one of two ways: two years of platform work with nothing to show, or a scatter of demos that never connect. We run both streams together — the foundation and the fast returns.
Two streams
One builds the asset. One pays the bill.
Make the company queryable
Stream A — AI Transformation
The data lake, the identity layer, the governance and the company brain. This is the asset that makes every future application cheap and every number trustworthy.
- One governed lake across tiles, slabs, bathware, adhesives and plants
- One canonical view of dealer, customer, SKU and project
- Certified metrics — one version of volume, mix and cost per msm
- Company brain: ask anything, in any language, with citations
- Every new application delivered in weeks on shared foundations
Show returns in one quarter
Stream B — Quick Wins
Voice-AI-led applications that go live on data Kajaria already has, in 6 to 12 weeks each, each with a measured number attached. They fund the transformation and build internal belief.
- Dealer and showroom voice desks — no enquiry lost, no call unlogged
- Collections voice agent — days off DSO in the first quarter
- Plumber and applicator engines — influencer networks activated at scale
- Attach-rate engine — adhesive on every tile invoice
- Price and reputation radar — market moves seen the day they happen
Applications live over time (cumulative count, on plan)
Quick wins carry the first two quarters. From the third quarter the platform stream overtakes them, because every new application is built on foundations already paid for. This chart counts applications, not rupees — returns are baselined against Kajaria's own numbers before any value is claimed.
The technical foundation, in plain terms
A data lake first, because everything else stands on it
Kajaria's data sits in SAP, in the dealer portal, in plant controllers, in service inboxes and in spreadsheets. Step one is to put it in one governed place with one name for every dealer, product and plant line. Only then does AI become cheap and trustworthy.
Systems of record
SAP ERP · dealer portal · plant SCADA & kilns · service desk · web, call and campaign data · JV feeds
Ingestion fabric
Change-data capture on transactional systems, streaming collectors on plant telemetry, schema checks at the door
Governed data lake
Raw, cleaned and certified zones on object storage, with history, lineage and per-brand access boundaries
Semantic & identity layer
One dealer, one SKU, one project, one applicator — plus certified metric definitions used by every screen and model
Model & agent layer
Forecasting and scoring models, voice agents, retrieval assistants and workflow agents, all monitored and auditable
Surfaces
Board command centre, role copilots, voice channels, dealer and applicator apps, write-back into SAP
How it stacks up
Surfaces
Models & agents
Semantic & identity
Governed lake
Sources
Data flows up from the systems Kajaria already runs. Decisions flow back down into them — that is the difference between a dashboard and a company that runs on AI.
Twenty-four months
Three phases, each with a clear exit test
Days 0-120
Phase 1 — AI-Ready
Make Kajaria queryable
Stand up the data lake, connect the systems of record, resolve dealer, product and plant identities, and put governance in place — while three quick-win applications go live on real data in parallel.
Work
- Connect SAP, dealer portal, plant SCADA, service desk and marketing sources into one lake
- Resolve one canonical ID per dealer, SKU, plant line, project and applicator
- Certify the metrics: volume, mix, realisation, cost per msm — defined once, used everywhere
- Access, PII and audit controls before a single agent touches customer data
Done when
- Any executive question answerable from one place, same number every time
- Data freshness under an hour for commercial data, minutes for plant telemetry
- Three applications live and measured, not piloted
Months 4-12
Phase 2 — AI-Scaled
One platform, many applications
Turn the lake into a delivery platform so each new application takes weeks, not quarters — shared voice agents, shared models, shared governance.
Work
- Voice agent fleet across dealers, plumbers, applicators, showrooms and collections
- Shared model layer for demand, churn, credit and yield with monitoring
- Write-back into SAP and the dealer portal so decisions become actions
- Adoption programme: every function has a named owner and a measured KPI
Done when
- 20+ applications in production on one platform
- Every model monitored, every agent decision replayable
- New application delivery measured in weeks
Months 12-24
Phase 3 — AI-Native
The company runs on its own brain
Cross-brand intelligence no single line could build alone: one customer view across tiles, slabs, bathware and adhesives, and autonomous operating loops with declared human checkpoints.
Work
- One customer and dealer graph across all brands, powering cross-sell by default
- Closed-loop pricing, replenishment and plant recipes
- Autonomous service, collections and quotation loops with human approval gates
- Company brain open to every employee, in their own language
Done when
- Cross-brand attach rate structurally higher, not campaign-dependent
- Cost per msm and DSO improved and holding
- AI is how the company works, not a project it runs
Stream B in practice
The quick-win queue
Each of these ships in 6 to 12 weeks on data Kajaria already has, with a number attached before we start.
DeployOne Connect
Showroom Voice Concierge
Every walk-in, missed call and web enquiry answered and followed up by a voice agent that speaks the customer's language and books the visit.
DeployOne Connect
Plumber Loyalty Voice Engine
Outbound and inbound voice agent that registers plumbers, confirms scheme points, answers queries and keeps the influencer network warm at scale.
DeployOne Connect
Dealer Voice Desk
A vernacular voice agent that takes dealer orders, answers stock and dispatch questions and logs every call — 24x7, in 12 Indian languages.
DeployOne Predict
Attach-Rate Engine
Spots every tile invoice shipped without adhesive and triggers the offer to that dealer through voice and field sales.
DeployOne Connect
Collections Voice Agent
Polite, compliant, multi-language reminder calls tied to ageing buckets, with promise-to-pay capture written back to SAP.
DeployOne Connect
Applicator Voice Helpline
An always-on multilingual voice expert for applicators: which product, how much, what curing time, what to do on a problem site.
DeployOne Connect
Experience Centre Voice Concierge
Captures every walk-in, calls back within the hour in the visitor's language, books the designer meeting and never loses a lead.
DeployOne Sense
Price & Competitor Radar
Continuous crawl of Morbi and branded price lists, e-commerce listings and trade chatter, alerting when a market moves.
DeployOne Dash
Field Force Copilot
Tells each field officer who to visit today, what to say, and what the dealer bought last — on their phone.
DeployOne Connect
Distributor Voice Desk
Voice agent taking distributor orders, quoting grades and confirming dispatch without a call centre.
DeployOne Engage
Carpenter Influence Engine
Registers, rewards and engages carpenters in local languages — the people who actually choose the brand.
DeployOne Sense
Bathware Review Radar
Continuous read of Amazon, Flipkart, Google and YouTube reviews for Kerovit against Jaquar, Cera and Hindware.