Advertisement
SAP Concur

MarTech360 Interview with Gabriele Franco, Co-founder at Cassandra

Peter•Nov 19, 2026
MarTech360 Interview with Gabriele Franco, Co-founder at Cassandra
Advertisement
SAP Concur Post Top

“The objective is not to make every dashboard agree. Different methods answer different questions. The objective is to know which evidence is strong enough for the decision you are about to make, and which uncertainty still needs a test.”

You began your journey in growth marketing and performance advertising before moving into marketing technology and building Cassandra. What experiences or challenges made you realize that marketers needed a fundamentally different approach to measuring and allocating advertising investments?

The question that matters is simple: if I spend more, what changes in the business? A platform dashboard does not necessarily answer that. It tells you which conversions it can claim. Those are different things.

That is the problem Cassandra is built around. A marketer can have an impressive channel report and still have no defensible answer to the CFO's next question: where should the next euro go? Looking backwards and distributing credit is not the same as deciding what to do next.

Our starting point was measurement and budget allocation: estimate what advertising contributes, understand diminishing returns, and simulate different investment choices. But the ambition has moved further. Even a good recommendation has no business value until somebody acts on it and checks the result.

That is the thread connecting the original Cassandra to what we are building now: measurement, decisions, execution and learning in one loop. The product is the medium. The outcome is the point.

Marketing teams have traditionally relied on platform attribution, last-click models and channel-specific analytics to evaluate campaign performance. Where do these approaches fall short, and what should businesses consider when building a more reliable measurement framework?

The biggest mistake is asking attribution to answer a causal question. Attribution distributes credit among observed interactions. Incrementality asks whether the sale would have happened without the advertising. Improving the first does not automatically solve the second.

Last click can reward the channel that captures existing demand while missing the activity that created it. Platform reports are useful for operating inside a channel, but each platform sees its own slice of the customer journey. Adding their reported conversions together does not give you a reliable business total.

I would build the framework from the business backwards: agree on revenue and customer definitions with finance, reconcile those outcomes with spend, use MMM for the cross-channel view, and use experiments to challenge the model's estimates. Keep platform metrics for diagnostics and execution.

The objective is not to make every dashboard agree. Different methods answer different questions. The objective is to know which evidence is strong enough for the decision you are about to make, and which uncertainty still needs a test.

Marketing Mix Modeling (MMM) has gained renewed attention as privacy changes and fragmented customer journeys make traditional attribution more difficult. How has MMM evolved, and what makes it practical and accessible for marketing teams that may not have dedicated data science resources?

MMM is becoming more useful when it moves from an occasional presentation to a maintained decision process. The important change is not a prettier channel-contribution chart. It is connecting data preparation, model validation, scenario planning and the next budget decision.

Modern tools can automate more of the data pipeline and model workflow, represent uncertainty, estimate saturation, and incorporate experimental evidence. That makes the process more accessible. It does not make the assumptions disappear.

A team without data scientists should still be able to ask: what data went in, which non-marketing factors were considered, how well does the model predict unseen periods, and does its channel estimate agree with a credible experiment? If those questions cannot be answered, accessibility has become a black box.

Cassandra brings MMM and incrementality testing into the same measurement workflow. The practical value is helping a marketer compare investment scenarios without rebuilding the analysis from scratch each time.

Also Read: MarTech360 Interview with Dock Kim, Vice President of Business Development at Bidease

Many organizations collect large volumes of marketing data but still struggle to translate insights into better investment decisions. How can marketing leaders bridge the gap between measurement, strategic planning and actual budget allocation?

Stop ending the process at the insight. An insight without an owner, an action and a check is unfinished work.

Start with a commercial objective and real constraints: the revenue or contribution target, available budget, margin, inventory and the time horizon. Then ask what the next unit of spend is expected to return. Historical average ROAS is not enough. A channel can have an excellent average return and already be close to saturation.

Turn the recommendation into a decision someone can execute: what changes, by how much, on which evidence, who approves it, and what would make us stop or reverse it. After execution, compare the outcome with the hypothesis and update the next decision.

For example, the question should not end at 'Search looks efficient.' It should become 'Is there evidence that additional Search spend will create profitable demand, rather than capture demand we already have?' If that is uncertain, the next action may be a test rather than a budget increase.

That connection between measurement and implementation is central to Cassandra's direction. We want the system to help complete the decision, not produce another handoff.

Incrementality testing is increasingly important for understanding whether marketing activities are genuinely driving additional business outcomes. How should brands combine MMM, incrementality testing and other measurement methods to develop a more complete picture of marketing effectiveness?

Give each method a job. MMM provides a whole-business view across channels and time. A well-designed incrementality test gives you a more focused causal answer about a specific intervention. Attribution helps diagnose the observable customer journey and operate campaigns. None should pretend to replace all the others.

The useful loop is: use the model to identify a consequential uncertainty, design a test around it, compare the result with the model, and use credible experimental evidence to improve the next allocation.

Calibration needs judgement. A test from one market, season or spending level is not automatically the truth for every future decision. Check its design, statistical power, contamination risk and relevance to the budget you are planning. Feeding a weak experiment into a model can make the answer more confident without making it more correct.

At Cassandra, combining MMM with incrementality is the foundation for connecting measurement to action. If we want an AI system to move money, we need a way to challenge what it believes causes growth. A fluent explanation is not a substitute for that.

Cassandra is evolving from a marketing measurement and budget allocation platform into a broader AI-powered marketing system. What inspired this shift, and how do you see AI changing the way marketing teams plan, make decisions and execute campaigns?

We realised that making analysis easier was only part of the job. You can recommend the right investment and still lose the opportunity while a report moves between an analyst, a manager, a media buyer and a creative team.

That is why we are expanding Cassandra from measurement and allocation toward a system that connects planning with the work itself: campaign operations, creative development, reporting and a shared memory of what the team has learned. Measurement remains the foundation. It is what helps us judge whether the work created value.

We are also changing how we build. We are working directly on marketing operations with clients, taking responsibility for a defined objective instead of relying only on interviews about what features they want. If we want to build a product that owns an outcome, we have to experience that responsibility ourselves.

The longer-term ambition is a business model better aligned with the client's results, rather than access to software alone. Automating tasks is a useful entry point. It is not the final differentiation. The harder problem is deciding which work is worth doing, doing it, and learning whether it helped.

With Cassandra AI moving toward supporting activities such as budget adjustments, campaign execution, creative development and reporting, how do you see the balance between AI autonomy and human oversight evolving? Which marketing decisions should remain firmly in human hands?

Autonomy should be earned at the level of a workflow. It should not be a switch you turn on for the entire marketing budget.

I would start with reading data, preparing reports and drafting recommendations. Then allow bounded execution where the objective, permissions, budget limits and escalation conditions are explicit. Spending money or changing a live campaign should require approval unless that precise action sits inside a limit the person has already delegated.

The human keeps the commercial objective, the risk appetite, the brand boundaries and the decision to expand that delegation. There are also two reviews I care about: the proposed action before it has consequences, and the learning before the system treats it as established knowledge.

That second review is easy to miss. A promotion, a stock change or seasonality can look like a creative breakthrough. If the system stores the wrong explanation, it can repeat the mistake across future campaigns.

The marketer's job moves from touching every setting to designing the rules and judging exceptions. Less manual execution does not mean less responsibility. It makes ownership more explicit.

As AI becomes more involved in marketing operations, how can organizations ensure that automated recommendations and actions remain transparent, explainable and aligned with business objectives? What role should data quality and measurement accuracy play in building trust in these systems?

An explanation written after the decision is not enough. I want to see the path from evidence to action.

For a budget recommendation, that means the source data and date range, the business objective, the assumptions, the uncertainty, the proposed change and the permission that allows it. After execution, there should be a record of what actually changed. If execution only partly succeeded, the system should say so.

Data quality comes before eloquence. Reconcile revenue with the business records. Check currencies, time zones, conversion definitions and reporting delays. Otherwise, an agent can compare numbers that look compatible but are not.

Measurement accuracy is also more than fitting historical revenue. A model can fit the total and still allocate channel credit incorrectly. Check predictions on unseen periods, compare with experiments, and make uncertainty visible at the point of decision.

This is the standard we want Cassandra to meet: when the evidence is weak, say what is missing instead of manufacturing certainty. 'Do not act yet' has to be an acceptable output. Trust grows when the system knows the boundary of its evidence, not when it always has an answer.

Marketing teams are under increasing pressure to demonstrate measurable business impact rather than simply report impressions, clicks or conversions. Which metrics should CMOs prioritize when evaluating marketing effectiveness, and how can they connect marketing investments more directly to revenue and profitability?

The headline should be incremental business value, with the cost of producing it attached. A high platform ROAS is not proof of profitable growth.

I would connect the scorecard in layers. First, the business result: incremental revenue and contribution after variable costs and marketing spend, where those can be credibly estimated. Second, the economics of the next decision: marginal return, incremental acquisition cost and payback, using assumptions finance agrees with. Third, the operating indicators that explain movement: conversion rate, creative fatigue, pacing and customer mix.

Keep attributed, modelled and experimentally measured results labelled. Do not present them as interchangeable facts. And treat lifetime value as an assumption to validate against customer cohorts, not a number that makes every acquisition look profitable.

For AI, I would add the cost of review and correction. Producing more assets or saving hours is useful only if it improves the business or creates capacity the team can actually use.

The CMO and CFO should be able to discuss the same investment decision in the same language: what we spent, what we believe it caused, how certain we are, and whether the next euro clears our profitability threshold.

Looking ahead over the next three to five years, how do you expect AI, predictive analytics and automated decision-making to reshape the structure and skill sets of marketing teams? Will marketers increasingly become strategic operators who manage and direct AI-powered marketing systems?

I think the unit of work changes from the task to the system. Today, a marketer is often judged on the campaigns they launch or the reports they deliver. Increasingly, they will be responsible for the system that produces those outputs, and for whether it drives growth.

In that system, measurement is not a report at the end. It is part of how decisions get made. Attribution, MMM and incrementality tests tell the system what is probably causing growth and how sure we are. Alongside them sit signals that can be read in real time, from inside and outside the business: creative signals, such as what is fatiguing or taking off in the account and in the market; offer signals, such as a promotion running on our side or a competitor's; and seasonality. Put those signals together with attribution and incremental measurement, and an automated decision stops being a guess. It can be precise, it can be made at the moment it matters, and it can be implemented.

Here is what that looks like from the client's side. One of our clients had a coupon code that stopped working at checkout. Over a couple of days, checkout conversion fell from around 50% to around 10%. Nobody had to spot it. The AI brain we built noticed the drop, worked out the root cause, and sent the client a request for approval to fix the coupon error. The client approved, the error was corrected, and we avoided losing close to 4,000 euros a day. It happened over a weekend, when the marketer was not working and would not have seen anything until Monday.

For the client, the value is easy to describe. The problem was caught while it was happening, not in next week's report. The cause was identified, not just the symptom. The fix still went through the client's approval. And the money that would have leaked was not lost. That is what it means to put measurement inside the decision loop: attribution shows where the problem is, incrementality tells you what is actually worth protecting or investing in, and the system turns both into an action someone can approve.

That is what I mean by a marketing engineer. Someone who understands marketing well enough to write down how a good decision is made, connect the right evidence and signals, give agents bounded tools, and judge whether the result and the learning are valid.

Predictive analytics becomes an input to that system: a way to compare possible moves and expose their uncertainty, not a forecast to obey. The skill mix combines commercial judgement, measurement, experiment design, data literacy and the ability to build repeatable workflows. Coding can help. Understanding the decision matters more than being able to automate it.

Cassandra's ambition is to provide the foundation for that role, so every team does not have to assemble the measurement, context, execution and learning loop from zero.

I would not predict that every company needs the same job title or team structure. I would predict a clearer dividing line: teams that use AI to produce more activity, and teams that build a system accountable for a business result. We want to build for the second group.

Thanks Gabriele! 

Gabriele Franco is a marketing technology entrepreneur and business leader specializing in marketing measurement, growth strategy, and data-driven advertising. As Co-founder and CEO of Cassandra, he focuses on developing technology that helps businesses understand marketing performance, measure incremental impact, and make more informed advertising budget decisions. His professional background includes founding Hybrida Marketing, where he worked on growth marketing strategies, performance campaigns, and data-driven approaches for businesses. Franco's experience managing marketing investments and working with attribution challenges influenced his move toward Marketing Mix Modeling (MMM) and AI-powered measurement solutions. His work centers on the evolving relationship between marketing analytics, artificial intelligence, and business performance, particularly how organizations can improve budget allocation and evaluate the contribution of different marketing channels.

Advertisement
SAP Concur Post Bottom