Apple •Spotify• Pocket Casts •Youtube •Overcast •RSS

What’s up everyone, today we have the pleasure of sitting down with Mitch Ribar, advisor on AI and customer lifecycle. He is the former Head of Customer Engagement & Data Science at Optus and Executive Director of AI & Marketing Technology at JPMorgan Chase.
Summary: Campaigns, calendars, and a CDP label that no longer means anything: Mitch walks through what’s left standing when AI picks the next best action for every customer. He makes the case for simulation as the new center of gravity, explains why 20 ideas should never all land on real customers, and shows how your A/B tests could be contaminating each other. You’ll also hear why doing nothing can be the smartest move a decisioning system makes. Stay for the end, where he breaks the campaign into the 5 building blocks he’d use to replace it.
In this Episode…
- Why Marketers Now Need to Be Communicators, Technologists, and P&L Owners
- Why the CDP Label No Longer Works and Simulation Is the New Center of Gravity
- How Customer Simulation Lets You Test 20 Ideas and Run Only the Best Few
- The 4 Types of Marketing Simulation and Why First-Party Data Matters Most
- How Agents, Simulation, and Live Tests Fit Into One Continuous Loop
- How a Universal Control Group Fixes Experiment Spaghetti
- Why Doing Nothing Is a Valid Next Best Action
- How to Balance Quarterly Targets With Customer Lifetime Value
- What Replaces the Campaign Calendar in Always-On Marketing
Recommended Martech Tools and Agencies 🛠️
We only partner with products and agencies that are chosen and vetted by us. If you’re interested in partnering, reach out here.
🔄 GrowthLoop: The agentic, composable CDP that drives compound growth by uniting your cloud data + AI into one marketing engine.
🐾 CaliberMind: GTM intelligence and multi-touch attribution platform giving marketers clear revenue analytics on what works – and the power to activate high-intent audiences across 170+ marketing platforms.
📧 MoEngage: Customer engagement platform that executes cross-channel campaigns and automates personalized experiences based on behavior.
🎨 Knak: Go from idea to on-brand email and landing pages in minutes, using AI where it actually matters.
About Mitch

Mitch Ribar advises CEOs, boards, and executive teams at Capita Technology on AI-powered customer systems built to improve retention, lifetime value, and commercial performance. He previously ran Customer Engagement & Data Science at Optus as EGM, a department that covered martech, data science, customer analytics, decisioning, and customer identity, with P&L responsibility tied to lifetime value, churn, upsell, and cross-sell.
Earlier, he held several executive roles at JPMorgan Chase across customer strategy, AI personalization, and marketing technology, and led product for financial services at Quantium. He’s now building customer simulation software with his team.
Why Marketers Now Need to Be Communicators, Technologists, and P&L Owners

Open the org chart at most large consumer brands and you’ll find the same split. Brand and lifecycle sit in one corner, martech in another, data science down the hall, and finance shows up once a year at budget time. Mitch has run data science, martech, and customer lifecycle teams, and he expects AI decisioning to collapse that map.
The marketer who survives the collapse has a long skill list. You need to tell a story, land a message, read an API document, and know your way around a P&L, all in the same week. Mitch calls it going back to the future.
“You’ve got to be equal parts communicator, technologist, and commercial, financially savvy.”
The past he’s describing is about 100 years old. Product was born out of the marketing department at Procter & Gamble, and the job covered 3 questions at once: what do we create for customers, how do we talk to them about it, and how do we price it. Mitch thinks marketing is heading back toward that scope. The unit of work becomes the totality of customer management, which bundles the product, the message, the channel, and the price. Fewer people hold more of it, and in the limit it starts to look like a single role.
Why Enterprises Overconstrain AI Decisioning
Ask marketers about AI decisioning and they sort into 2 camps. One camp feels anxious that a machine will pick the content and the send time. The other camp shrugs, because machine learning has been choosing channels inside their systems for almost a decade. Phil asked where most enterprises sit on that anxiety scale.
Mitch says most of them reach for the least risky option. People who have to hit KPIs and targets protect those numbers, so even a big decisioning program ends up wrapped in guardrails. He’s built several of these programs at different companies and now helps others do it, and his advice is to go all in. Let the system find the next best action. The moment an executive puts a finger on the scale because they’re sure they’re right, the program starts to trip over itself.
Phil adds the complexity that makes people hesitate. One big B2C brand he talked to has 7 tiers of product offerings, each owned by different teams, with data science groups holding different pieces. Decisioning for a single product line looks feasible. Across all 7 it looks like a mountain, so everyone keeps their head down and works on their own line.
Mitch sees opportunity in exactly that mess. Teams that optimize inside one area find a local maximum, the best result available from their own corner. Teams that work together can find the global maximum. Put your commercial hat on, sit down with finance and analytics, and ask what a great customer looks like. Mitch can’t think of a company without synergies between its product lines. Great customers hold more than one product, stay longer, and use multiple channels.
Once you know what great looks like, a flywheel starts. You move more customers onto that path, they spend more and stay longer, and your acquisition gets more targeted. Because you’re retaining more people, you can outspend competitors to win new ones. Mitch expects the next decade to belong to medium and large companies that get better at retention and growth from their existing base, and that advantage carries over into acquisition. The org chart most enterprises run today was drawn around channels and products, and the first companies to redraw it around the customer will collect that compounding effect before everyone else.
Key takeaway: Put one person from data science, one from martech, and one from lifecycle in a room with finance, and have them define what a great customer looks like: how many products they hold, how long they stay, and how many channels they use. Use that profile as the single goal your decisioning system optimizes across every product line. Give the system room to run before anyone adds a guardrail.
Back to the top ⬆️
Why the CDP Label No Longer Works and Simulation Is the New Center of Gravity

What a customer data platform does depends on which vendor you ask, and Mitch has watched that happen from the start. When Segment.io arrived, a CDP was a data aggregator, a way to get every feed into one spot. Today, depending on the vendor, the same label covers identity, segmentation, and activation, and Mitch calls it a really murky spot.
He doesn’t think the CDP is gone. The category as marketers knew it is dead, he says, but every company still needs one core place to organize its customer data, so “long live CDPs.” What changed is the shape. Platforms keep absorbing features from the layers above and below them, and the vocabulary has split into composable CDPs, hybrid CDPs, and other flavors until the label stopped helping buyers pick anything.
“I think the new center of gravity for the stack is simulations.”
The consolidation is easy to see in his examples. Databricks is moving up the stack with a customer lake product it recently announced. GrowthLoop and Braze are both announcing decisioning products. For a buyer sorting through that, Mitch says, it’s a difficult place to be.
Over the last 10 years the community has chased one idea, a single customer brain that does the decisioning. Mitch has swapped the brain for a nervous system. Collecting signals and alerts, and holding memory across every channel, interaction, and moment, is a bigger challenge than installing the brain. The data moves fast, it’s detailed, and you have to combine it across channels before it means anything.
That brings him to his bet. When you have so many variables to account for, especially across a customer base in the millions, simulation becomes the new center of the stack. Mitch is building software with his team around that bet. Buyers who judge a platform by the cleverness of its decision engine will miss that sensing and memory decide whether the engine has anything worth deciding on.
Key takeaway: Audit your stack by function instead of by category label. List which tools sense customer signals, which hold memory, which decide, and which act. Then find the weakest of the 4, because the sensing and remembering layers are usually the ones holding up the decision.
Back to the top ⬆️
How Customer Simulation Lets You Test 20 Ideas and Run Only the Best Few

Say you run a subscription business and retention is slipping. Your options are a discount, a product change, a new message, or nothing at all. Whichever one you pick, the standard play is the same. You expose real customers to it, you set up the experiment, and then the waiting starts.
“You wait, and you wait, and you wait, and you get your statistical significance, and then you check the box and it’s either good or bad.”
Mitch agrees that’s still the gold standard. A simulation lets you explore more ideas offline first, then take only the most promising ones live. Phil framed it in terms any growth team will recognize: instead of testing 20 things next quarter on real customers, you run all 20 on a simulated version of your customer base, and only the best 2 meet actual people.
The reason that matters comes from how decisioning programs are set up. Most run a champion and challenger structure, and Mitch says north of 80% of customers receive the champion experience, because the business has to make money and keep customers happy. The challenger is the experiment trying to beat it. Every test pulls people out of the champion group, and that puts revenue and churn at risk. More simulation means more confidence before you experiment on customers, and it keeps more of your base on the experience you already know works.
Mitch then pushes the idea further. The ultimate version of simulation forecasts how a customer plays out over time. What happens if you send the same offer or the same message again and again? The world doesn’t stand still. Customers change, and competitors change with them. That time dimension is where Mitch’s own software is aimed, and he describes it as the next big step.
Every experiment on a live customer base has a price, paid in the revenue and churn risk of whoever got pulled out of the champion experience. Simulation is the first tool that lets a team shrink that bill without shrinking how much it learns.
Key takeaway: Write down the share of your customer base that currently sits outside your champion experience in live tests. That number is the cost of your testing program. Set it as the figure a simulation layer should bring down.
Back to the top ⬆️

Speaking of disrupting martech… are you still operating like it’s 2013 and you haven’t updated your chatbot? We recently started collaborating with the Docket team and what they are building with their Inbound Demand Agent is super cool. Check it out.
The 4 Types of Marketing Simulation and Why First-Party Data Matters Most

Vendors now use the word simulation for at least 4 different products, and a buyer who treats them as one thing will compare tools that don’t compete. Phil laid out the categories he’d found in his research: journey logic simulation, decision and policy simulation, synthetic customers and digital twins, and multi-agent ecosystem simulation. Mitch says they’re broadly right, and he adds that it’s such a new space that the boundaries are still moving.
| Type of simulation | What it models | Examples Mitch gave |
|---|---|---|
| Journey simulation | The if-then-else rules a customer follows down a path | Onboarding flows and renewal flows |
| Decision and policy simulation | What happens when you change guardrails, priority, or offer mix | Moving a base from an 80/20 marketing-to-newsletter split to 70/30, or cutting a send cap from 3 emails a week to 2 |
| Synthetic customers and digital twins | Agents that mimic your customer base, a research cohort, or a focus group | Panels of agents built from customer data |
| Multi-agent ecosystem simulation | How customers, brands, agents, and competitors interact over time | Causal contexts, with so many variables that the whole world has to be modeled |
Decision and policy simulation has been around the longest, Mitch says, and he gives a shout-out to Pega for being thought leaders there more than a decade ago. The journey simulator is the most basic of the 4. It tests whether the rules you wrote behave the way you expect when a customer moves through them.
“The data that you use to create these synthetic customer panels or these digital twins is the most important thing.”
Synthetic customers get most of his attention. The technology to build agents that mimic your customers or your focus groups already exists, so the quality of the input decides the quality of the answer. Mitch loves first-party data for that reason. If you can capture what customers actually do, you get a much richer output than you’d get from surveys, questionnaires, or reviews, because it’s very difficult to control the bias in those datasets.
He keeps 2 questions in front of every project. How do I get the least biased data possible? He admits it’s never perfect, but you do your best. And how do I control for as many variables as I can? That second question takes him straight to the ecosystem category. Customers, brands, agents, and competitors all move at once, so you need different ways to model what the world looks like and what your customers do in it. Mitch calls that cutting-edge work.
Synthetic customers are only as good as the behavioral data underneath them, which makes the company with the cleanest first-party data a stronger bet than the one with the cleverest agents.
Key takeaway: Before you buy or build a synthetic customer panel, ask what behavioral data trained it and how its bias was controlled. If the answer leans on surveys or reviews, treat its output as a hypothesis worth testing live. Match each vendor to one of the 4 simulation types so you only compare tools that answer the same question.
Back to the top ⬆️
How Agents, Simulation, and Live Tests Fit Into One Continuous Loop

Every simulation category so far has a person at the keyboard deciding what to test. Phil asked about a fifth category, where an agent does that job. On Monday morning the agent reports that it already ran the simulation, here are the 3 interventions that look best, and here’s the result, so go live-test them with real customers.
Mitch wants it closer than “the future” and says his team is researching it heavily. He wouldn’t call it a fifth category, since it interacts with the other 4. One set of agents executes campaigns and actions. Another set does the research and analysis, works down the list of simulation types, and decides what the next best action should be. What the real tests teach goes back into the simulator, which then picks the next thing worth simulating.
Phil summed up the working model, and Mitch confirmed it. The loop runs in 6 steps:
- Collect first-party data and get to know the customer.
- Model the customer base and keep that model dynamic.
- Simulate interventions, from new campaigns to frequency caps to policy changes.
- Execute one of them with real customers.
- Observe the result causally.
- Recalibrate, and go back to knowing the customer.
“You can’t, you can’t simulate your way to victory.”
That loop is what Mitch is chasing, and he adds the warning that goes with it. You still need great first-party data, a deep understanding of your customer base, a model that represents it, and great products and features. Anyone who thinks they can switch on agentic automation, send another million emails, and fix their problems is going to be disappointed. The work humans keep is strategy, creative, ideas, and being the champion for the customer.
Why Simulation Needs a Context Layer
Phil tried to place simulation inside the stack everyone already runs. A sensing layer collects identity, events, transactions, and product usage. A decisioning layer applies policy, constraints, and eligibility to pick an action. An activation layer, the old reverse ETL territory, chooses the channel and the offer. A measurement layer tracks drift and causal lift. He asked where the new simulation piece sits among them.
Mitch’s first reaction is that laying out all those categories shows how complicated the field has become. He likes to joke with his teams that the whole game is moving metadata around: you get the right data to the right spot at the right time. That’s the nervous system idea again, applied to every tool on Phil’s list.
Then he makes a firm claim about the simulation layer. He doesn’t think it exists without a much better context layer. Mitch groups context with the causality work happening in pockets and with the data warehouses, and the hard part is making that layer easy to distribute as well as easy to access. Plenty of companies consolidate and join all their data, then forget to get it out to the right spot.
Once simulation engines are running, they feed the decisioning layer and take some responsibility off it. Decisioning keeps the last mile, which is real-time context at the channel and live signals from customers. Mitch adds a condition. You have to be able to trust the output, so you test your simulations too, because you can’t simulate and hope for the best.
Agents will make interventions cheap to produce, which moves the scarce resource to the customer model that decides whether any of them are worth trying.
Key takeaway: Draw your own loop on one page, marking where customer data is collected, modeled, simulated, executed, and measured. Circle every arrow that depends on a person moving a file by hand, since those are the places a simulator will have stale data. Fix the most important circle before you evaluate any simulation vendor.
Back to the top ⬆️
How a Universal Control Group Fixes Experiment Spaghetti

At WordPress.com, Phil watched a huge volume of people start free websites every day, with a goal of converting them to paid plans. The product team tested the onboarding flow. The data science team tested an ML propensity model. The lifecycle team changed the HTML templates on certain emails. Some teams wanted to change one small thing at a time, others wanted to rebuild the whole experience, and others wanted audiences held out across every test. Everyone was experimenting on the same people at the same moment.
Mitch knows the situation, and he says the bigger the company, the harder it gets, because there are more people. Everyone wants to do a good job and show that their work drove the lift. A/B testing has been drilled into marketers for decades, so every meeting comes down to the same demand to show the results of the test.
“I call them the contaminated tests.”
When everyone tests at once, Mitch doesn’t see how any team can claim a clean result. The tests collide. He’s heard it called spaghetti, and he prefers contaminated. Try this check: add up the lift every team claims, walk over to finance, and ask how much of it hit the P&L. Mitch expects them to laugh first. Then, on a good day, they’d say 10% or 20%.
An air traffic control function for tests would help, and he calls it hard to coordinate when digital runs a red button versus blue button test while marketing runs copy tests on the same customers. So he lands on 2 fixes.
The first is more simulation, which means fewer live tests to manage. Mitch has watched the same pattern in other industries: make something easier and cheaper to do and people do more of it. Testing got so cheap that it now contaminates its own results. Simulation lets everyone keep running quick, cheap tests without all of them landing on real customers.
The second is a universal control group.
“You should have a universal control group. That means you have a set of customers that are completely held out of everything for a long period of time, like 12 months.”
That group catches organic changes in your customer base and shifts from the market and from competitors, because the lifts and dips show up in people nobody contacted. It also improves your test reads. Instead of measuring from an internal baseline, you measure the change between the universal control group and whatever test you’re running. Mitch hopes everyone already does this, and if they don’t, he says to start quickly. Phil adds why it matters for credit. If you can’t say which of 20 experiments caused an uplift, the holdout still tells you how much the combined effort moved the baseline compared with doing nothing at all.
Cheap experiments turned every team into a source of lift claims, and the only reconciliation finance will accept is a group of customers that no team was allowed to contact.
Key takeaway: Carve out a universal control group, a set of customers held out of every campaign and test for 12 months. Report each team’s lift against that group as well as against its own baseline. Then total the lift every team claims and compare it with what finance sees in the P&L.
Back to the top ⬆️
Why Doing Nothing Is a Valid Next Best Action

Everyone building a decisioning program obsesses over the next intervention, the next message, the next thing to put in front of a customer. Mitch thinks sometimes the most valuable action on the list is no action. He sees teams get this wrong all the time, especially new teams, or teams at companies that haven’t done customer lifecycle or customer value management before.
Every customer base has a group that wants to be left alone, or that’s overly sensitive to being stimulated to shop. Contact them and you’ll make them angry. Mitch says the opposite outcome is arguably worse.
“If you wake up a part of your customer base that is, let’s say, quote-unquote, sleeping. You’re stimulating a behavior that’s gonna cause them to go and look around, maybe at your competitors, and you’re gonna lose them.”
He’s watched the expensive version of this mistake. A company spends millions on a big martech program, buys the newest and shiniest tools, gets excited, and starts stimulating churn. Mitch calls that a quick way to destroy a business case for technology that is really great.
Mitch’s answer is to understand the segments and cohorts in your customer base, and then he asks to get on his soapbox about them.
Segments, Cohorts, and Audiences Are 3 Different Things
Mitch has sat in too many meetings where people use these 3 words interchangeably. His definitions:
- Segments are durable. Customers shouldn’t bounce between segments more than once or twice a year. Demographics, psychographics, and core product behaviors set them.
- Cohorts always have a time component. They’re temporal, such as the new customers who joined in week 42 or the ones who joined on a particular coupon code. They cut across segments, and they’re how you understand time and start to forecast.
- Audiences are how you use pieces of segments and cohorts to build the list of who you’re going to talk to about a message.
He compares it to medicine. A doctor from Sydney, Shanghai, Lisbon, or Los Angeles shares a syntax, so anterior and posterior mean the same thing in every operating room. Marketing needs more of that definitional work to enter its next generation with less confusion, and to stop having conversations that start with “what’s a CDP, what exactly is it again?”
A decisioning system with no way to choose silence treats every sleeping customer as an opportunity, and that’s how a martech investment starts producing churn.
Key takeaway: Add “do nothing” to your action library as a real option, and list the groups that should get it first: customers who want to be left alone and customers who would go shopping elsewhere if you woke them up. Agree on one written definition each for segment, cohort, and audience before your next planning cycle.
Back to the top ⬆️
How to Balance Quarterly Targets With Customer Lifetime Value

Sales is behind on quota, so someone asks marketing to email everyone who signed up a few months ago and hasn’t bought yet. Phil calls it the email blast, and in B2B it arrives with a reason attached: we need more MQLs. His question was how a team adopts always-on decisioning while real quarterly metrics still have to be hit.
Mitch calls it the million-dollar question, and the tension he describes is a clash of 2 clocks. Companies run on 12-month budget cycles, those get chopped into quarterly targets, and the targets have to be hit. Customers don’t live on that clock. In the industries Mitch knows best, banking, telecommunications, and software, customer tenure is often 5 or 10 years. Think about how rarely you change bank accounts, cell phones, or insurance providers.
“You can hit a quarterly goal, but what about lifetime value?”
Lifetime is the word he stresses. A customer who stays longer gives you more chances to talk about your products and services and to deliver a great experience. Mitch believes it’s a win-win. Customers are happier because they get what they need, and the business makes more money. The balanced conversation about quarterly numbers only works when a counterweight sits beside them, whether that’s LTV or whatever long-term metric fits the business.
He takes Phil’s email blast as the test case. Suppose the team acquires a whole bunch of customers from a channel it rarely uses, and then asks what that group will look like in 12 months. It could be a big churn risk rising on the horizon, and the customer lifecycle team would have to deal with it, which pulls them away from the work they’d planned.
Mitch doesn’t offer a shortcut. It depends on the company and the industry. His closing condition is specific: establish those long-term metrics, and set up a customer team with a concept of a customer P&L that fits the tenure of your customers. Then you’ve got a fighting chance.
Phil adds from his own marketing ops and lifecycle experience that he wishes more of his past leaders had thought in lifetime value. With email as the core channel, deliverability turns every send into a decision about whether the message is worth the damage it might do to future sends. Companies with customers who stay 5 or 10 years are managed on 12-month budget cycles, and the gap between those 2 clocks is where churn risk hides.
Key takeaway: Pair every quarterly acquisition target with a 12-month counterpart metric such as LTV, and review both in the same meeting. Before approving the next email blast or discount push, ask what that cohort will look like in 12 months and who on the lifecycle team will inherit the churn.
Back to the top ⬆️
What Replaces the Campaign Calendar in Always-On Marketing

Plenty of B2C teams still run on a calendar. October means a spooky-season campaign, December means winter themes, and every theme gets built by hand. Phil asked what Mitch would tell those teams, given that Mitch’s own library of building blocks has no slot for the campaign.
Mitch starts by defending the calendar. It has a time and a place, and it works as a backdrop for coordinating seasonal events and capturing business constraints, including the quarterly KPIs that have to be hit. A big product launch is rare, and everyone should rally around it. But a calendar that runs the whole business is too slow, because it’s not responsive, flexible, or dynamic enough, and the one thing you can guarantee is that your customers and your competitors are dynamic.
“Are your team surprised that Valentine’s Day turns up every year? Are they surprised that Thanksgiving rolls around and you gotta do Black Friday sales?”
On Black Friday, Cyber Monday, or the yearly iPhone launch in telecommunications, Mitch disagrees with the idea that they need a campaign calendar. He admits the creative treatments differ every year. These are also revenue-sensitive moments, and many brands avoid taking risks in them. His answer is to break the campaign down.
Every campaign is a bundle of 5 primitive groups:
- Product: what it is, its price point, and any discounts.
- Channel: where the message goes.
- Content: what the message says and shows.
- Customer: the segment, the audience criteria, or whatever data represents the person.
- Action: the objective, whether that’s a direct response, engagement, or a newsletter.
Glue those 5 together and you can build whatever you want. An always-on decisioning system needs a library full of each primitive, and that library gets you ready for the step after, generative content that’s one-to-one. Mitch’s example: his email would carry cats and Phil’s would carry dogs, with images and text generated for each of them. Phil might love exclamation points. Mitch wants the Oxford comma.
The same structure improves experimentation. When you isolate a variable per primitive group, a pricing test changes the product category and holds everything else fixed: the channel, the content, the action, and the customer targeting. A campaign is a bundle of 5 decisions made once, and a decisioning system remakes each of those decisions for every customer.
Key takeaway: Take your last 3 campaigns and break each one into the 5 primitives: product, channel, content, customer, and action. Store each piece in a library, tagged so a system could pull it independently. Run your next test by changing one primitive and holding the other 4 fixed.
Back to the top ⬆️
How to Decide What Deserves Your Energy Using a Calendar

Mitch spends his working days thinking about systems that decide for customers, so Phil asked how he decides for himself. He’s a father of 2, a motorsport fan, and a founder getting ready to launch. How does he choose what deserves his energy at any moment?
“I can say with 95% accuracy, no matter what happens, I’m doing bedtime and I’m reading the books.”
Family comes first, and bedtime is the fixed point that everything else is scheduled around. Past that, Mitch says he obsesses about his calendar. Wherever he needs to be and whatever he needs to do goes in. Watching Formula 1 gets the same treatment, so he books it and knows the time is set aside.
With so many things going on, he says, you need some backbone to your day, your week, and your month. For him, that backbone is the calendar, and his system sits inside it.
Phil closed with a bonus question: is Bluey a national treasure in Australia? Mitch says it’s on Australian coins now, in $1 and $2 versions that are hard to get. Everyone wanted to collect them.
People who work around always-on systems still need a few hours that don’t answer to anyone, and a calendar entry is the cheapest way to create them.
Key takeaway: Block your fixed commitments first, the way Mitch protects bedtime. Then book the things you love, such as a race or a game, as real calendar events. Treat the calendar as the backbone of the week and let everything else fit around it.
Back to the top ⬆️
Episode Recap

Mitch Ribar argues that every campaign breaks down into 5 primitive groups: product, channel, content, customer, and action. Marketing organizations built around campaigns, channels, and calendars will need to reorganize around the customer as AI systems start choosing the next best action. He expects the marketer’s job to shrink back toward its origins, where one person can tell a story, read an API document, and read a P&L.
The same loop runs through every chapter. You collect first-party data, model the customer, simulate interventions offline, run the best few on real customers, measure the result against a universal control group, and recalibrate. Splitting the campaign into primitives makes that loop testable, because you can change one variable and hold the rest fixed. 2 habits keep the loop honest: “do nothing” sits in the action library as a legitimate choice, and every quarterly target gets a lifetime value counterweight.
For the stack, Mitch sees the CDP label blurring as Databricks, GrowthLoop, and Braze move toward each other, and he swaps the customer brain for a nervous system. Sensing and memory are the harder problem, so a simulation layer needs a context layer underneath it. For the org, he wants a customer team that owns a customer P&L matched to how long customers actually stay, with the retention flywheel funding acquisition.
Mitch admits what’s still open. Simulation is a new space and the 4 categories are only broadly right. Agents that choose what to simulate next are something his team is researching, and he wants them closer than the future. He says there are no shortcuts for reconciling quarterly numbers with lifetime value, because the answer depends on the company and the industry, and he insists that simulated results still need testing before anyone trusts them.
Follow Mitch on LinkedIn at linkedin.com/in/mitchribar.
Full episode ⬇️ or Back to the top ⬆️

✌️
—
Intro music by Wowa via Unminus
Cover art created with Midjourney (check out how)
Apple •Spotify• Pocket Casts •Youtube •Overcast •RSS
Related tags
<< Previous episode
Next episode >>
All categories
- AI (118)
- career (67)
- customer data (70)
- email (66)
- guest episode (195)
- operations (134)
- people skills (35)
- productivity (10)
- seo (16)
See all episodes
Future-proofing the humans behind the tech
Apple •Pocket Casts•Google •Overcast •Spotify •Breaker •Castro •RSS