In this AI Time Journal Q&A, we spoke with Rajesh Jain, Founder and Group Managing Director of Netcore Cloud, about where AI changes marketing decisions, why holdout groups matter, and which capabilities enterprises should build themselves by 2027.
Where are enterprises seeing the clearest measurable gains from AI in customer engagement, beyond experimentation or personalization demos?
The clearest gains come where AI changes a decision. Who to message, when, and whether to message at all are decisions with money attached.
In a Netcore client deployment, an online food brand's WhatsApp return had stalled at 7X across 1,117 campaigns. It began choosing recipients by predicted likelihood to open and click. Return reached 13X across four campaigns, with conversion up 94% and revenue per message sent up 3.5X. That comparison is against its own earlier sends, not a holdout group, and I would say so plainly.
The distinction matters. A campaign that converts more has not shown that AI caused the conversions, because the system may simply be targeting people who were going to buy anyway. Holdout groups settle that question, and I would not call a result proven without one.
Sometimes the best decision is to send nothing. A customer close to buying may need a reminder, a recommendation or silence. Telling those apart is where the value sits.
How is the role of first-party customer data changing as marketing teams introduce AI-driven decisions and workflows?
First-party data used to be something you segmented and planned campaigns around. With AI making decisions, it becomes the picture the system acts on, and that picture has to be current.
In another Netcore client deployment, a large Southeast Asian mobile operator was targeting audience lists that were weeks old, for an app used by millions in real time. After moving to live, session-based signals it reported audience growth of 7.3X quarter on quarter and click rates up 37%. Same customers, same app. The difference was how current the signals were.
Three things follow. Identity has to be resolved, so one customer is one record across systems. Consent and purpose travel with the record, so every automated decision can check them. Outcomes flow back, so a conversation or an enquiry improves the next decision.
More data is not the goal. A brand that uses what it knows to make discovery easier earns trust. One that uses the same data to send repetitive messages loses it, and then pays media platforms to find the customer again. I call that the reacquisition tax.
What does it take operationally to connect AI-driven customer journeys across channels rather than deploying isolated features?
Start with what the customer sees. A customer buys on the app and later receives a WhatsApp offer for the same product. Each channel did its job. Nobody owned the customer.
Fixing that takes four things. One identity across channels. One place where the next action is chosen, with channel tools carrying out that choice instead of making their own. Shared rules on consent, frequency and quiet periods, because a customer meets one brand, not five senders. And one owner for the journey, with a retention or revenue target, in place of channel owners each improving their own open rates.
Do not automate everything at once. Pick one journey with money in it, such as abandoned carts, repeat purchase or early churn. Set a baseline, connect the data and channels it needs, test, and expand once the numbers hold. The system also has to know when to wait or stop.
In a Netcore client deployment, a Southeast Asian logistics platform did this for checkout, mid-shipment and post-first-order moments. It reports cancelled-shipment recovery up 17% and payment recovery up 10%.
Which metrics best distinguish a successful enterprise AI marketing deployment from one that looks promising but does not deliver lasting value?
Opens and clicks show the system is busy. Four other measures show whether it is useful.
Incremental result against a control group. Repeat purchase, retention and revenue per customer, compared with a comparable group the system did not touch. This separates real impact from purchases that would have happened anyway.
Margin after everything. A discount can lift sales and cut profit. Count discounts, media, technology and the cost of running the AI itself.
Staying power. Cohort retention and repeat-purchase rates over two or three quarters. Many deployments look strong in month one and fade as customer behaviour changes.
Human effort. If people keep overriding or fixing what the system queued, it will not scale, and the override rate usually moves before campaign results do.
Agree the baseline and the measures before launch, so that conversation happens on day one rather than after the results arrive.
What are the hardest implementation challenges around data quality, integration, governance, and customer trust, and how should leaders address them?
The hard problems sit in records, pipelines and rules.
Data quality. Duplicate and late records do the most damage. In a Netcore client deployment, a diagnostics chain had duplicate patient profiles and could not tell individuals from families. A family-level identity model let its campaigns address a household correctly. Decide which few signals change a decision, such as a purchase, a complaint or a consent change, and keep those current first.
Integration. Legacy systems export in batches. Stream the events that matter instead of trying to connect everything.
Governance. Every automated decision needs an owner. Define what the system may decide alone, what needs approval and what stays with a person, such as unusual discounts, sensitive data and regulated claims. Keep logs showing why a customer received a message.
Trust. Customers accept automation in routine moments and resist it when money, health or a complaint is involved. Slow down there, and say plainly how data is used.
My advice is bounded autonomy. Start with low-risk decisions such as send time or channel, watch the results, and widen the system's authority only when they justify it.
Looking toward 2027, which capabilities should enterprise marketing leaders build internally, and where should they rely on technology partners?
Build what carries your judgement. Buy what is becoming standard.
Five capabilities belong inside the company. The definition of a valuable customer and the outcomes that matter, because no vendor can set those for you. Measurement: designing experiments, using control groups, reading incremental results. Supervision: setting objectives and limits for AI systems, watching what they decide and stepping in when they go wrong. Collaboration between marketing, data, engineering and compliance around shared outcomes. And a habit of continuous testing, since models and customer behaviour will keep changing.
Partners should supply what is costly and slow to build alone: customer data infrastructure, identity resolution, models, real-time decisioning, channel orchestration, security and governance controls.
Ownership stays with the enterprise: the customer relationship, data policies and the definition of success. Ask partners to share accountability for outcomes as well as deliver software. A vendor paid on seats or messages has no reason to care whether retention improves.
By 2027 I expect brands to differ in three things: how current their customer picture is, how well they supervise decisions, and whether they can prove results.
