Past Sessions

• Event Details: MAU Vegas
• Date: 22 May 2025
• By: Tejas Manohar, Co-Founder & Co-CEO, Hightouch and Jon Carden, CMO & Head of Growth, Fundrise
• Estimated read time: 6-8 minutes

Quick Read Summary

Personalization has been “possible” for a decade, yet most marketing teams still cannot deliver it profitably at scale. The blocker is not a lack of data or tools, it is the cost of turning those inputs into decisions, content, and delivery that consistently outperform a simple one size fits all approach.

Fundrise’s experience is a useful blueprint because it reflects a reality many leaders avoid saying out loud, the more segments you add, the more content you need, and the less confident you become that any of it is moving the business. John Carden described a point where the cost of personalization rose exponentially while the return dropped, so the team stopped trying and standardised messaging.

What changed was not a new channel or a new workflow hack, it was a new operating model, generate more high quality content from fewer source documents, then use agents to decide which message to send, to whom, when, and how often, optimised against a business outcome like win back revenue.

The practical takeaway for CMOs and growth leaders is straightforward, treat personalization as a decisioning problem, not a segmentation problem, then invest in the content inputs and governance that let machines do the routing without sacrificing brand trust.

The Real Bottleneck Is Not Data, It Is Decisions

If your team feels stuck, it is probably because personalization work still behaves like manual production, not like a scalable system. Many organisations have more customer data than ever, and more platforms than ever, yet still struggle to deliver “the right content to the right customer on the right channel” consistently.

The uncomfortable truth is that most modern stacks increase optionality faster than they increase capability. One speaker cited an ecosystem of roughly 12,000 martech platforms, and every one of them claims to enable personalization in its own lane. The hard part is orchestration across lanes, across timing, and across competing goals.

Interpretation: This is why “more tools” often produces less momentum. Every new platform adds new variants to test, new audiences to define, new operations to manage, and more surface area for inconsistency. If your process still depends on humans proposing micro segments and manually mapping content to each segment, the system does not compound, it fragments.

Practical action: Audit personalization effort as a unit economics problem, not a creativity problem. For each lifecycle objective, write down the marginal cost of adding one new segment and one new content variant, then compare it to the marginal lift you can realistically measure. If the curve is going the wrong direction, you need a different mechanism than more A B testing.

When Personalization Breaks, Trust Oriented Brands Feel It First

Brands that win on credibility face a stricter constraint, you cannot “scale content” by lowering the bar. Fundrise operates across real estate, private equity, private credit, and venture style exposure for retail investors, and by 2025 it had grown to more than $3 billion in assets under management and more than 2 million customers in the US.

Their audience spread makes the personalization math brutal. The same platform serves an 18 year old making a first $10 investment and a 55 year old professional investor managing millions. Meanwhile the product catalogue broadened into multiple asset classes and more than 10 funds.

Carden’s conclusion will feel familiar, the cost of personalization rose exponentially as breadth increased, while the return dropped, so the rational move was to stop.

“The actual cost of personalization was rising exponentially… and the return on that investment was dropping precipitously.” John Carden, CMO and Head of Growth, Fundrise

Interpretation: Many teams interpret this moment as a failure of execution, or a reason to hire more people. It is usually a failure of mechanism. If the system requires humans to do the matching of message to person, then scaling audience breadth creates a non linear workload, and scaling workload eventually forces a quality tradeoff. Trust sensitive categories reach that breaking point sooner.

Practical action: Protect quality by narrowing what humans must author, and expanding what machines can derive and distribute. Your content strategy should include a small number of “source of truth” assets that you are willing to put executive scrutiny behind, and a controlled method for turning those assets into many formats without changing the underlying claims.

The Content Waterfall, One Great Asset, Many High Utility Outputs

Scaling content does not mean publishing more, it means extracting more decision ready variants from the best material you already have. Fundrise’s model starts with a flagship asset, its quarterly investor letter, roughly 2,000 to 3,000 words, written to a very high standard because trust is the lifeblood of the investor relationship.

From that single asset, Carden described producing about 15 derivatives in 20 minutes to 90 minutes, including a short version, an FAQ, a podcast outline, versions focused on specific asset classes, and versions written for different personas.

The quantified result is the kind of metric that resets expectations inside a leadership team, roughly three times as much content produced, in about one third of the time, which he framed as about a nine times efficiency gain.

Interpretation: This is the first half of an effective AI decisioning system. Agents cannot pick the right message if you only have one message. Variety is not optional, it is the raw material for relevance. The point is not that every derivative is brilliant, it is that you can afford to generate enough options to let performance data do the filtering.

Practical action: Build a “content waterfall” playbook. Start with one canonical narrative per quarter or per product cycle, then define a repeatable set of derivatives, such as executive summary, beginner FAQ, expert deep dive, asset specific angle, objection handling, win back framing, and short form SMS friendly snippets. Treat each derivative as an input that can be tested and routed by AI decisioning, not as a standalone campaign.

AI Decisioning Replaces Micro Segmentation With Outcome Optimisation

The strategic shift is to stop guessing segments, and start optimising actions against an outcome. Before agents, Fundrise’s team would propose micro segments and run targeted sends, but the segments were too small and too time intensive to be consistently ROI positive.

With AI decisioning, the workflow changes. You set a goal, like win back dollars, feed in many possible actions and content options, add guardrails, and let the agent choose the right person, the right message, the right time, and the right frequency.

Carden described a win back program that used to send one new piece of content per week. After shifting to the agent driven model, over roughly four to six months, Fundrise saw a four times increase in dollars coming back to the platform, with no ongoing intervention beyond producing content.

Interpretation: This is what “personalization at scale” should mean in 2025, not infinite segments, but continuous allocation of attention based on feedback. It is closer to how sophisticated ad platforms operate than how lifecycle teams have traditionally operated, which is why Carden compared it to buying ads where “the content is the targeting.”

Practical action: Pick one lifecycle objective where speed to learning matters and where you have enough content variety to support an agent, for example win back, first conversion, or cross sell into a second product line. Define success as an outcome metric, not open rates. Then operationalise guardrails, such as frequency caps, compliance rules, and brand tone constraints, so the system can explore safely while it learns.

Why Reinforcement Learning Beats Traditional A B Testing for This Job

If you are still running endless A B tests, you are paying an operational tax that gets worse as your option set grows. The core issue is statistical and managerial, every additional variant increases the time to significance, increases the number of moving parts, and increases the risk of test interference.

The alternative described here is reinforcement learning, often implemented in marketing as contextual multi armed bandits. The concept is simple, balance exploration, trying new options, with exploitation, doubling down on what works, rather than splitting traffic evenly across all options for long periods.

An analogy used was a robot trying to find the best food in a city. It can keep going to restaurants it already knows are good, or it can explore new places that might be better. The system learns, then shifts allocation as evidence accumulates.

Interpretation: The promise is not just better performance, it is faster convergence with less waste. Instead of committing large audience slices to weak variants just to “complete the test,” the agent reduces exposure to underperformers and concentrates attention where value is emerging. For marketers, this converts a calendar based testing culture into a continuous optimisation culture.

Practical action: Reframe experimentation governance. Instead of asking, “what are we testing this month,” ask, “what action set are we giving the agent, and what new options are we injecting weekly.” Measure quality of inputs, such as breadth of content variants and clarity of guardrails, alongside outcome metrics. Also document where you still need classic experiments, such as brand positioning tests, and where you can let reinforcement learning do the routing.

Integration Matters Because Your Stack Is Not Going Away

The fastest path to value is an agent layer that orchestrates the tools you already use. Hightouch positioned its AI decisioning product as integrating into existing ESPs and engagement tools rather than replacing them, and cited an integration footprint of roughly 300 connectors across common marketing platforms.

They also made a key technical clarification, this is not “put all customer data into a single large language model prompt.” The core decision engine is reinforcement learning, with additional machine learning models and selective use of generative models layered in to enhance the process, including concepts like uplift modelling for incrementality.

Interpretation: For senior leaders, this should reduce two common fears, operational disruption and data risk. You can adopt AI decisioning as a control layer, keep your delivery systems intact, and still take advantage of adaptive optimisation without treating an LLM as the decision maker of record.

Practical action: Treat implementation like a product launch, not a tool install. Start by mapping which systems will remain the system of record for messaging, which system will define eligible audiences, and where guardrails will live. Then define an escalation plan for unexpected outcomes, such as over messaging a cohort, and build dashboards that reveal what the agent is learning, not just what it is sending.

The CMO Implication, The Team’s Value Moves Up the Stack

As execution gets cheaper, judgment gets more expensive. Carden’s most important point was organisational, the bottleneck shifts away from “getting the work done” and toward understanding the business, the product, and the customer deeply enough to set the right goals and supply the right inputs.

He also made the leadership reality explicit, expectations rise because output and quality are now achievable in days, not weeks. Being reliable at routine execution, like sending emails on schedule, is no longer a differentiator.

Interpretation: This is the second half of the AI decisioning story. Tools do not remove accountability, they concentrate it. If the team can generate more variants and deploy them faster, then strategy, prioritisation, and customer insight become the differentiators. The marketing team of the future becomes smaller, more analytically grounded, and more product fluent.

Practical action: Update your hiring and performance criteria. Look for marketers who can define a goal precisely, translate customer understanding into action sets, partner with product and data teams, and evaluate tradeoffs across acquisition, engagement, and win back. Then protect time for strategic work by automating or agentifying the operational burden that used to consume the week.

Conclusion, Personalization Wins When You Treat It as a System

Personalization is no longer limited by the number of campaigns your team can manually produce, or the number of segments you can defend in a meeting. It is limited by whether you can supply a steady stream of high quality content options, define clear outcome goals, and operate an AI decisioning layer that continuously learns which action to take for which customer.

Fundrise’s results point to the same thesis many mobile and growth leaders are arriving at, agents create leverage in two directions at once, they reduce the cost of producing and routing content, and they increase the performance you can extract from the same audience and the same channels.

If you want “content personalization at scale” to be more than a slogan, start where the leverage is, build a content waterfall from a trusted source asset, then let reinforcement learning marketing systems do what humans struggle to do consistently, explore widely, converge quickly, and keep optimising without burning out the team.

About the Speakers

Tejas Manohar , Co-Founder & Co-CEO, Hightouch, Tejas is the co-founder and co-CEO of Hightouch, a leading data activation platform. He specialises in building tools that empower marketing and growth teams to use data more effectively. His work focuses on bridging the gap between data infrastructure and business outcomes.

Jon Carden , CMO & Head of Growth, Fundrise, Jon leads marketing and growth at Fundrise, the largest private real estate investment platform in the US. With a background in strategy and performance marketing, he has been instrumental in scaling Fundrise’s investor base and deploying AI to drive engagement and ROI.