Integrating Synthetic Into CX Operations
Treat synthetic tools as an upstream accelerator to a standard personalization and experimentation model.
The Intersection of Synthetic Approaches and Customer Experience
TL;DR: How synthetic data and personas accelerate the design, testing, and delivery of better experiences—without waiting weeks for field research.
Personalization only improves customer experience when it is continuous, contextual, and tied to measurable outcomes. McKinsey’s research shows that when organizations execute personalization at scale, they routinely see 10–30% uplifts in revenue and retention alongside 10–20% gains in marketing efficiency—a tangible signal that relevance, delivered consistently, compounds value for both customers and the business. The same body of work underscores that personalization has shifted from nice-to-have to table stakes, with companies that activate it at scale putting lifetime value on a steeper trajectory through superior retention and richer relationships. Hitting that upside depends on an operating model that keeps pace: orchestrating the “4Ds”—data, decisioning, design, and distribution—while building a measurement playbook that uses incrementality testing and standard metrics to validate impact in-market. Taken together, the evidence argues for an approach where CX teams can test, learn, and deploy at speed—because the business case for timely relevance is already settled.
What “synthetic” adds to the CX toolkit
Synthetic methods give CX leaders practical ways to move faster without skipping rigor. First, personas that update themselves: rather than treating personas as static posters, teams can generate data-driven stand-ins that refresh with new behaviors and context, aligning with Forrester’s guidance to make personas purpose-built, insights-based, and inclusive so they actually guide decisions. Second, journey sandboxes: gen-AI–powered customer digital twins let teams simulate cross-channel journeys—browse to cart to service—and observe likely drop-offs, churn risk, or sentiment effects before real customers ever see a change. Related work shows how pairing digital twins with generative AI improves scenario exploration and decision quality, which is exactly what CX prototyping requires. Third, rapid hypothesis screening: by running hundreds of low-cost “pre-experiments” synthetically, teams can narrow to the most promising offers, creatives, or policies and then confirm with lightweight human tests—an experimentation pattern long advocated in the literature and now easier to execute at scale . The net effect is faster learning cycles, fewer in-market missteps, and experiences that feel timely and tailored rather than generic.
Where synthetic directly improves the customer’s experience
Synthetic methods let teams shape offer relevance, service quality, and post-purchase engagement before real customers feel the changes. McKinsey’s work on personalization shows that when relevance is executed at scale, companies see double-digit gains in revenue and retention; the catch is maintaining a fast loop from idea to in-market learning. Gartner’s guidance on the Digital Twin of a Customer (DToC) further shows how simulated customers and journeys can anticipate behavior, pressure-test policies, and predict the experience a change will create. Forrester adds that personalization must extend across onboarding, adoption, and renewal—areas where many firms still under-serve customers. Together, these points argue for using synthetic personas and journey sandboxes to screen copy, timing, fees, service rules, and handoffs across the full lifecycle—so teams fix rough edges before they reach real people. The next step is operational: build a simple, repeatable flow that turns signals into simulations, simulations into shortlists, and shortlists into live tests.
References
https://hbr.org/2018/11/using-experiments-to-launch-new-products
https://hbr.org/2017/09/the-surprising-power-of-online-experiments
The Emergence of Synthetic Personas
Personas are still one of the best ways to spread customer empathy across large organizations, but most teams admit their versions go stale long before the next planning cycle. Forrester defines customer personas as archetypes that represent real subsets of your base—people who share goals, needs, and behaviors—which is a useful anchor for what we’re trying to simulate at scale. Forrester has also warned that many enterprise personas are built on outdated methods that amplify bias and collect dust rather than guide decisions; they urge a shift to purpose-built, insights-based, and inclusive personas that teams actually use. Synthetic personas meet that brief by generating data-driven, updatable stand-ins that behave like real segments without relying on last year’s research binder.
The practical appeal is speed and freshness. With large language models and related techniques, teams can spin up personas conditioned on current signals—first-party behavior, market context, even channel constraints—and then simulate how those personas respond to offers, creative, or service policies. Early evidence suggests this isn’t just a parlor trick: researchers have shown that LLM-based “AI participants” can reproduce a wide range of published marketing effects at scale, enabling rapid hypothesis tests before fielding with humans. In parallel, McKinsey notes that gen AI now lets marketers produce highly relevant messages for micro-communities at a volume and cost previously out of reach, which is precisely where synthetic personas shine as a planning and testing layer.
For enterprise teams, the operating model is straightforward: use synthetic personas to pressure-test ideas quickly, then reserve human research for the few decisions that warrant it. Treat them as decision accelerators, not replacements. A sensible workflow pairs synthetic simulations with small, targeted human validations, updates the personas on a set cadence, and bakes in bias checks and documentation—aligning with Forrester’s guidance on modern persona quality. When you do this well, you also give the CFO something tangible: faster time to market and measurable uplift from better personalization, outcomes that McKinsey has linked to double-digit gains in revenue and retention when personalization is executed effectively.
We’ll carry this forward in the next post by connecting synthetic personas to customer experience design—how these “plausible customers” help teams prototype journeys, stress-test service policies, and raise the floor on relevance before the real customers ever see a thing.
Defining “Synthetic” in the Marketing Context
This article series explores the role of synthetic research and data in this speed to action.
Synthetic methods in marketing begin with data that never existed in a CRM yet behaves as if it did. Gartner frames synthetic data as information “generated by applying a sampling technique to real-world data or by creating simulation scenarios where models and processes interact to create completely new data”. Put simply: machines spin up plausible records—behaviors, preferences, even quotes—without lifting a single row from a customer database. For our purposes, “synthetic” also covers personas and research protocols that originate in algorithmic models rather than focus-group recruitment lists.
While anonymized or de-identified data strips away direct identifiers, it can still be reverse-engineered. Synthetic artifacts avoid that headache because they mirror statistical patterns without sharing a molecule of the original substance—a distinction MIT Sloan highlights in its discussion of privacy-safe data generation and Gartner’s projection that 60% of AI-ready data will be synthetic by 2024. In other words, marketers get the flavor without the fingerprints, satisfying regulators and risk officers in one swoop.
Synthetic outputs come in three broad flavors: structured tables for tabular modeling, unstructured text or images for qualitative insight, and multimodal bundles that fuse clicks, sentiment, and context into a single sandbox. A 2024 SAS report (drawing on Gartner research) estimates that by 2026 three out of four enterprises will lean on generative AI to create synthetic customer data, underscoring just how mainstream these techniques are becoming.
Generative AI adds rocket fuel to the mix. Forrester notes that GANs and VAEs now shape synthetic datasets that “capture real-world patterns and scale efficiently, improving analytics accuracy and speed” while keeping analysts on the right side of privacy law. The result is less time begging Legal for clearance and more time stress-testing new offers before your competitors have finished their coffee.
With definitions, distinctions, and use cases in hand, we can turn to the next logical step: building personas that move as fast as your data. In future articles, we’ll explore how synthetic personas give teams a living, breathing test market—without the airfare, incentives, or awkward small talk.
References
https://www.gartner.com/en/information-technology/glossary/synthetic-data
https://www.forrester.com/report/synthetic-data-for-customer-insights/RES182031
The Need for Instant Insights at Scale
This article series explores the role of synthetic research and data in this speed to action.
Competitive advantage hinges on speed. Decision cycles that once spanned quarters are now expected to resolve in days—sometimes hours. By the time a team syncs up for their weekly status meeting, the moment of opportunity may already have passed. This isn’t hyperbole; it’s the new operating reality. Gartner’s 2024 CMO Spend and Strategy Survey found that “demonstrating ROI” (69%) and “revenue generation” (59%) are marketers’ top priorities, yet average marketing budgets have dropped to 7.7% of total company revenue—a 15% year-over-year decline (Gartner, 2024). This combination of shrinking resources and rising expectations leaves little room for drawn-out research cycles. Insight delayed is advantage denied.
More fundamentally, time—not money—has become the most constrained resource in enterprise decision making. McKinsey reports that senior executives spend nearly 40% of their working hours making decisions, yet over 60% of that time is considered ineffective. For a typical Fortune 500 company, this inefficiency can cost up to $250 million annually in lost productivity (De Smet, Jost, & Weiss, 2023). The implication for marketing teams is clear: every hour spent wrangling unstructured data from a legacy dashboard is an hour not spent crafting meaningful, timely experiences. Synthetic data and personas—by generating ready-to-analyze, “plausible people” on demand—help eliminate that bottleneck. Instead of waiting weeks for customer feedback or panel insights, marketers can simulate interactions and test hypotheses within hours.
At the same time, customers have little patience for brand experiences that feel generic or out of step with their needs. As Harvard Business Review has cautioned, over-reliance on slow, outdated inputs leads to “engineered insincerity”—where automation mimics empathy but fails to deliver relevance (Scheibenreif, 2023). Personalization at scale only works when underlying signals are updated continuously. Otherwise, yesterday’s segmentation becomes today’s churn trigger. This makes the case not just for synthetic speed, but for synthetic adaptability. As this book will argue, synthetic personas and data—when governed responsibly—enable marketing teams to move at the pace of customer expectation, not just campaign calendars.
References
De Smet, A., Jost, P., & Weiss, L. M. (2023, August 30). How to make better decisions in the age of urgency. McKinsey & Company. https://www.mckinsey.com/featured-insights/mckinsey-guide-to-excelling-as-a-ceo/how-to-make-better-decisions-in-the-age-of-urgency
Gartner. (2024, May 29). CMO Spend Survey 2024: Proving ROI and Driving Growth With Less. Gartner, Inc. https://www.gartner.com/en/marketing/research/cmo-spend-survey
Scheibenreif, D. (2023, October 2). Personalization at Scale Requires Real-Time Customer Data. Harvard Business Review. https://hbr.org/2023/10/personalization-at-scale-requires-real-time-customer-data