Your Next Customer Is Simulated
Why the Companies That Win Will Test Their Products on AI Personas Before Real Humans Ever Touch Them
Last month, I wrote about The Death of the UI and argued that software interfaces are failing because they force humans to adapt to product architecture instead of the other way around. That piece struck a nerve. The feedback was immediate and split cleanly into two camps: people who agreed the current model is broken, and people who asked, “Okay, but what do we actually do about it?”
This is the answer.
Simulate the customer before they ever arrive.
The Blind Spot in Every Product Team
Here is something that should bother every product leader: you spend months building features, designing flows, and writing documentation, and then you launch and pray.
The feedback loop is brutal. Real users hit your product, get confused, open support tickets, churn, or worse, they leave silently and you never know why. You measure NPS scores after the damage is done. You run A/B tests that take weeks to reach significance. You rely on user research sessions with 8 to 12 people and extrapolate from there.
All of this is reactive. You are measuring failure after it happens.
What if you could measure failure before your product ever ships?
Enter the Simulated Customer
The concept is straightforward but the implications are enormous.
Instead of waiting for real users to struggle with your product, you create AI-powered personas that interact with your software the way real humans would. Not scripted bots clicking through predefined flows. Not regression test suites checking if buttons render. Actual simulated humans with goals, frustrations, varying levels of technical literacy, and real intent.
Imagine this:
Persona A: A non-technical marketing manager trying to set up a new campaign dashboard for the first time. No prior experience with your product. No onboarding call booked. Just a login and a goal.
Persona B: A senior operations lead migrating from a competitor platform, importing 50,000 contacts, and expecting feature parity.
Persona C: A CEO who has 4 minutes between meetings and needs to pull a revenue report for a board deck.
Now let each of these simulated personas loose on your product. Don’t give them a tutorial. Don’t give them documentation. Just give them their intent and watch what happens.
Where do they get stuck? How many steps does it take? Do they reach for the help widget? Do they abandon the task entirely?
This is not QA. This is pre-launch customer experience testing.
Why Traditional Testing Misses the Point
The software testing industry is undergoing a massive transformation. In 2026, enterprise QA budgets have ballooned as companies race to adopt AI-driven testing, autonomous agents, and continuous validation. The market is on track to more than double in size over the next decade. That growth is real.
But here is the gap: almost all of it is focused on system correctness.
Does the API return the right response? Does the form validate correctly? Does the page load under 3 seconds? Does the integration sync without errors?
These are important questions. They are also the wrong questions if you are trying to understand whether a real person can actually use your product to accomplish their goal.
System correctness tells you the machine works. It does not tell you the human succeeds.
The shift I am describing is from testing systems to simulating experiences.
A Concrete Example
Let us take something every SaaS company deals with: user onboarding.
In the traditional model, the product team designs an onboarding flow, the engineering team builds it, QA verifies it works, and then real users go through it. If the drop-off rate at step 3 is 40%, you find out three weeks later from your analytics dashboard. Then begins the cycle of hypotheses, redesigns, and another round of A/B testing.
In the simulation model, you run 500 simulated personas through the onboarding flow before launch. Each persona has a different role, technical background, and goal. You measure:
Completion rate across persona types
Average step count to reach activation
Points of confusion where simulated users pause, backtrack, or change direction
Support dependency and whether the help widget or docs reduce friction
Drop-off probability at each stage
You can even run variants. Version A has a guided tooltip walkthrough. Version B has a video overview. Version C has no onboarding at all, just the raw product.
Compare the results. You now have data-driven conviction about which onboarding approach works best, for which persona type, before a single real user has logged in.
That is the power of the simulated customer.
The Three Layers of Simulation
The way I think about this, there are three distinct layers:
Layer 1: System Simulation. This is what most companies already do, or are moving toward. Automated testing that verifies the software behaves correctly across states, permissions, and configurations. It answers: does the system work?
Layer 2: User Simulation. This is where the shift happens. AI personas attempt real tasks within the product without scripted paths. They explore, make mistakes, get confused, retry. It answers: can a human succeed?
Layer 3: Experience Simulation. This is the frontier. You introduce variables like support layers (chatbot, knowledge base, live agent routing), different UI treatments, and varying levels of documentation. You measure how each variable affects the outcome for each persona type. It answers: what combination of product and support architecture maximizes success?
Most companies operate at Layer 1. The ones that will dominate the next decade will operate at Layer 3.
Why This Matters for Support, Not Just Product
Here is something that surprised me as we explored this at Marketrix: the biggest unlock is not in product design. It is in support architecture.
Think about it. Every SaaS company has a support team, a knowledge base, maybe a chatbot, maybe live agents. But how do you know if your support layer actually reduces friction? How do you know if adding a Zendesk widget helps or hurts? How do you know if your knowledge base articles are being surfaced at the right moment?
Today, you measure this through ticket volume, CSAT surveys, and resolution time. All lagging indicators. All measured after the user has already struggled.
With simulated customers, you can A/B test your entire support architecture proactively.
Run the same persona simulation twice. In version A, the product has an embedded AI assistant. In version B, it has traditional documentation. In version C, it has no support layer at all.
Compare task completion rates, step counts, and drop-off probability across all three.
Now you know, with data, whether your AI assistant is actually helping or whether it is just another popup users dismiss.
This transforms support from a reactive cost center into a measurable design variable.
The Product Manager’s New Superpower
If product managers could, through simulation, gauge the correctness of software flows (QA), and gauge the perception of users interacting with it (CX), they can make bold decisions with conviction, as they have simulated the outcome a priori.
Think about what that unlocks:
Ship with confidence. You have already stress-tested the experience across dozens of persona types before launch day.
Prioritize with data. Instead of debating which feature to build next, simulate the impact of each on user success rates.
Reduce support costs proactively. Identify and fix friction points before they become support tickets.
Personalize at scale. Different persona types struggle at different points. Tailor the experience before they ever ask for help.
The product manager of 2026 does not just read dashboards. They run simulations.
What This Looks Like in Practice
At Marketrix, we are building exactly this. We combine UI-operating simulations with persona-based user simulations to help companies understand whether their software actually enables users to succeed, and how different support layers influence that success.
The workflow looks like this:
Define your personas. Who are your users? What roles do they hold? What is their technical sophistication? What are they trying to accomplish?
Set up the simulation. Point the system at your product. No SDK required, no code changes. The simulation operates on the UI layer, just like a real user would.
Run the scenarios. Each persona attempts real tasks. The system measures steps, time, confusion points, support interactions, and task completion.
Compare variants. Test different onboarding flows, support configurations, or feature treatments. See which combination maximizes success for each persona.
Iterate before launch. Fix the friction. Adjust the support layer. Re-run. Repeat until the numbers look right.
This is not about replacing user research or real customer feedback. Those remain essential. This is about adding a layer of pre-launch intelligence that catches problems early, validates hypotheses quickly, and gives product teams the conviction to move fast.
The Bigger Picture
We are entering an era where the best companies do not just build products. They simulate the entire customer journey before it happens.
Simulation is becoming the connective tissue between product, QA, support, and customer experience. The silos between these teams dissolve when you can run a single simulation that tests all of them simultaneously.
I believe this is one of the most important shifts in software development since the move to cloud. And just like cloud adoption, it will start with early movers, face skepticism from incumbents, and eventually become the default.
The UI era taught us how to design screens. The simulation era will teach us how to design outcomes.
The companies that simulate first will ship with more confidence, grow faster, and spend less on reactive support. The rest will keep guessing.
Time to stop guessing.
If you’re building a SaaS product and want to explore how simulation can transform your approach to QA, onboarding, or support architecture, I’d love to connect. Reach out at irosha at marketrix dot ai or find me on LinkedIn.
And if you enjoyed this, subscribe below for more insights on the future of AI, simulation, and what it means to build software that actually works for people.