Writing
Blog
Notes from the people behind ThirdVoice: plain, evidence-led, and India-first.
Measurement & attribution
-
A/B Tests Are Decided Before They Run
The discipline that separates a real experiment from an anecdote is pre-test sample-size mathematics — significance, power, minimum detectable lift, baseline rate. If those were not fixed before launch, the dashboard is not evidence.
-
Dashboards Nobody Acts On Are Worse Than None
The right outcome metric for an analytics command centre is time-to-action, not dashboard count. Unused dashboards manufacture the illusion of capability, and the integration work beneath them is always the underestimated cost.
-
Last-Touch Attribution Is a Measurement Artifact
Most conversions follow multiple visits and cross-channel touches over days. Crediting the final click is not a finding about customers — it is a limitation of the tool, and it systematically misallocates budget toward channels that close rather than channels that open.
-
Pre-Post Analysis Is Not Mix Modelling
Cookie deprecation revived marketing-mix modelling, but most of what Indian brands call attribution is pre-post analysis — which never controls for concurrent activity, and therefore cannot tell you what your marketing actually caused.
-
Social listening earns its budget the week it changes a decision
The value of social analytics is not the annual brand-health score; it is the signal that something is changing, fast enough to act on. The right question moved from 'who owns your social data?' to 'who owns your social intelligence — and did it change a decision this week?'
Customer data & CDP
-
Data you cannot activate is a liability
First-party data is only as valuable as the activation infrastructure around it. Collection without identity resolution, consent verification, and channel activation produces regulatory exposure under India's DPDP Act, not commercial return.
-
First-party data strategy does not survive contact with India's indirect economy
Playbooks written for Western direct-to-consumer brands assume the brand meets its consumer. Most Indian brands never do. The prerequisite is a data-capture channel — loyalty, pack, partnership — not a platform.
-
Most CDP deployments in India are data warehouses with better marketing
A customer data platform earns its keep only when data flows out to activation. Most Indian deployments only flow in. The unified profile is a build-vs-buy-vs-connect decision, not a platform purchase.
MarTech stack decisions
-
Maturity assessments fail as scorecards
Digital-maturity assessments that stop at scoring produce roadmaps clients agree with but never fund. Every gap must be priced, sequenced, and tied to a business-value case — or the assessment is an opinion, not an investment decision.
-
Western martech goes blind in B2B2C India
Marketing stacks built for direct-to-consumer retail cannot see India's intermediated reality. The durable playbook: intermediary intelligence first, proxy signals as interim measurement, and owned channels valued as data assets, not revenue channels.
-
Your consultant's platform is not your strategy
Every services firm that also owns or resells platforms carries a structural conflict of interest, and clients deserve disclosure. The honest differentiator in marketing technology advice is demonstrable platform neutrality — by design or by governance.
Operating model & strategy
-
The capability you rent stays rented
Managed marketing services close the capability gap fast, but after two or three years the models, the data mart and the know-how live in the vendor's delivery centre. Buy the bridge — but only with a knowledge-transfer workstream priced in from day one.
-
The right strategy two years early is a wrong strategy
India's e-commerce formation years proved that a correct strategy deployed ahead of the market is economically equivalent to a wrong one. Timing is a strategy variable, not an execution detail — and it must be interrogated as hard as the strategy itself.
Personalisation & identity
-
If you cannot name the segment, it is not a segment
Behavioural segmentation earns its keep at the naming step, not the algorithm. A cluster you cannot describe in plain words is over-fitted or under-differentiated — and the response-rate staircase remains the funding case for behavioural data.
-
Next best action fails on the offer library, not the model
Next best action, not personalisation, is the right frame for applied AI in marketing — and when a programme fails, it fails on the offer library, not the scoring or the delivery.
Retail media
-
Retail media closed the loop a decade of analytics never could
Retail media is the structural fix for the consumer-goods attribution problem. It gives brands closed-loop ad-to-purchase measurement inside the channel they already sell through — something no analytics stack built on proxy data ever delivered.
-
Retail media is an operating discipline now, not a media line
Retail media has professionalized: its own currency (incremental return on ad spend, share of voice, bid discipline), its own workflows, its own staffing. Brands still treating it as marketplace ads are a full maturity cycle behind the practitioner standard.
Applied AI in marketing
-
AI is this decade's SMAC
The AI urgency argument of 2024–26 is structurally identical to the SMAC argument of 2013. The CMO who lived through that cycle already has the mental model for this one — including where it oversold.
Customer experience
-
Journey maps do not travel
A customer journey map adapted from a global template misrepresents the Indian customer, whose channels run through WhatsApp, UPI, short video, QR-on-pack and kirana activation. Map the as-is before the to-be, and rebuild every map from India-specific research.
Loyalty & CRM
-
Your loyalty programme is a data asset or it is nothing
Most loyalty programmes are undifferentiated point schemes that accrue liability on purchases that would have happened anyway. The real value is first-party data capture and mindstate-matched treatment — and 'everyone gets the same offer' is the tell.