In Data We Trust - The Trust-Driven Architecture

Last week, I shared my POV on Apple's Unfair Advantage in AI (from editor

Martin Ma

- change of plans, this is scheduled to be published next on 06/26) - and the core of it really boiled down to one thing: Trust.

But if you think this is just an Apple thing, you couldn't be more wrong. That was just the tip of the iceberg (and yes, for my data friends, the pun is intended). Today, I want to dig deeper.

Trust-Driven Architecture

They say 80% of AI work is really data work. And let's be honest, we're all in a mad dash to build AI-native companies and inject intelligence into every product. But in this race, I see so many of us building these incredible, futuristic skyscrapers on foundations we still don't fully trust. From my perspective, this lack of trust creates a toxic, cascading failure mode that paralyzes organizations from the outside in.

1. It Starts with Your Customer

This is the first and most fundamental wall. If your users don't trust you, they won't give you their consent or their data. It's that simple. In an era where over 80% of consumers are actively concerned about their data privacy, trust isn't a "nice-to-have" - it's the price of admission. Without earning this foundational trust, you have nothing to work with. Your AI ambitions, your personalization engines, your entire data strategy - they're all dead on arrival because you don't even have the fuel to get started.

2. Then, It Infects the Data Itself

But let's say you clear that first hurdle and get the data. The crisis of confidence moves inward. When your internal teams don't trust the data they have, you get one of two outcomes. The first is analysis paralysis, where every project gets bogged down in a dysfunctional tax of endless validation and debate. But the second is far more dangerous: teams push forward anyway, building powerful products and AI models on that shaky foundation. This is where it gets truly scary, because

The biggest risk to your business isn't your AI strategy failing; it's your AI succeeding on bad data.

It leads to flawed financial projections, misguided marketing campaigns, and AI models that amplify hidden biases with terrifying efficiency.

3. Finally, It Breaks Your Teams

This crisis of confidence inevitably erodes trust in the people and teams responsible for the data. When product and engineering teams can't rely on the data platform team, they create workarounds. They build their own shadow data pipelines and hire their own analysts, duplicating effort and creating organizational scar tissue. Instead of innovating, your most expensive talent is forced to spend a huge portion of their time on basic, foundational data work - cleaning, validating, and wrangling data just to get to a starting point. It's a massive, dysfunctional tax that kills velocity because no one trusts the factory to produce clean parts.

How I Think About Earning Trust

So, where do we start fixing this? From my perspective, you have to begin at the source.

From where I stand, trust isn't a feature; it's the entire operating system for your data platform. It’s the only currency that truly matters, and I’ve been thinking a lot about what it takes to earn it from every possible angle, starting with the most important one.

Dimension 1: Trust with Your Users

This is where it all begins, and it's non-negotiable. And let's be real, a privacy policy nobody reads isn't trust; it's a legal defense. True trust is earned by treating people like partners, not data sources.

• It means being radically transparent about why you need their data and what awesome thing you're going to do for them with it. No more vague statements. Be specific.

• It means giving them granular, easy-to-use controls and making it just as easy to say "no" as it is to say "yes." When you do that, you send a powerful message: we respect you more than we want your data.

• It means communicating the entire data lifecycle, from cradle to grave, so there are no fears about what’s happening behind the scenes. Be open about how data is collected, stored, used, shared, and eventually destroyed.

It all forms a trust flywheel - the more trust you earn, the better data people give you, which lets you build better products, which earns you more trust. Let's just learn from the best with Apple's playbook.

Dimension 2: Trust in the Data Itself

Once you've earned your users' trust and have data to work with, the next critical battle is internal: ensuring that everyone in your organization trusts the data itself. This is where the rubber really meets the road. Your analysts, data scientists, and business decision-makers are the primary customers here, and their trust is multifaceted and fragile. It all starts with quality.

1. The Promise of Quality: Getting the Basics Right

I can't say this enough: bad data is worse than no data at all. We've all been in that soul-crushing meeting where the whole hour is spent debating if a number is right. It shatters confidence and grinds progress to a halt.

Now, the six dimensions of data quality - accuracy, completeness, timeliness, uniqueness, validity, and integrity - are an over-discussed point in our industry. We all know they're important. They are the absolute, non-negotiable table stakes. If you don't have a plan for these, you're not even in the game. But getting them right isn't the end goal; it's the prerequisite for everything else. It's the foundation upon which real trust is built.

2. The Promise of Enablement: Making Data Accessible and Understandable

Can people actually find what they need? This is where I see the platforms trying to build trust by lowering the barrier to entry. Look at Snowflake’s new Snowflake Intelligence or Databricks One from Databricks Data + AI Summit announcement. It’s a direct attempt to build trust with the entire business by creating a conversational AI that lets a non-technical user simply "talk" to their data. It’s a powerful statement: "You don't need to be a SQL wizard to get answers." This is about empowering the "data-curious" and making everyone feel like they have a stake in the data culture. Though, out-of-box solutions will only get you to the door, the key is to leverage your own data and unblock your own business needs with the last mile engineering. What's your excuse not to do it now?

3. The Promise of Empowerment: Trust Through Self-Service

As you scale, you can’t have a data platform team as a bottleneck. You have to push ownership out to the data domains where the expertise lives, embracing a Data Mesh philosophy where it applies.

"You build it, You own it." is not just about accountability, but also more importantly empowerment with their skin in the game!

But that only works if you empower them with great, self-serve tools. This is where the platform philosophies get fascinating, and I'm totally in the camp of believing these platform tools needs to be both "buy" + "build" - buy commodities, partner to learn, and build to your strength. Talking about which, just a couple of weeks ago, we learned - Snowflake is building trust by integrating dbt natively, simplifying a beloved engineering workflow. Databricks is making a different trust play by open-sourcing Declarative Pipelines, betting that the freedom of open standards will win more hearts and minds. Both are trying to solve the same trust problem: making it easier for engineers to do their best work without being bottlenecked or vendor-locked.

4. The Promise of Effectiveness: The Ultimate Measure of Trust and ROI

This brings me to a concept I’ve been thinking about a lot lately. We talk a lot about "Data ROI", "TCO" and "value," but those terms can feel too narrow and purely financial. I believe we need a more holistic measure, one that captures the true health and efficiency of a data culture. I call it Data Effectiveness.

To me, Data Effectiveness is the ultimate measure of trust made tangible. It goes beyond a simple ROI calculation. It's a comprehensive yet very simple, cost-attribution model that forces us to answer the tough questions and prove that the trust placed in our data - or the cost of not having it - is justified. It’s about moving beyond vanity metrics and getting brutally honest about the real-world impact of our data investments.

A holistic formula for this concept can be expressed as:

\(\text{Data Effectiveness} = \frac{\text{Value Generated} - \text{Direct Investment} - \text{Trust Tax}}{\text{Direct Investment}}\)

Here’s how I think it's different:

• It explicitly accounts for the "Trust Tax". A standard ROI might calculate the cost of a data scientist's salary. Data Effectiveness asks: How much of that salary are we wasting because she has to spend 80% of her time on data janitorial work, rebuilding basic features from scratch because she doesn't trust the official source? That wasted time is the "Trust Tax," and it's a massive, hidden drain on productivity that Data Effectiveness puts on the balance sheet.

• It measures the cost of organizational friction. It’s easy to say a dashboard is "valuable." But is it still valuable when you account for the cost of the data ingestion pipelines, the storage, the compute for transformations, the visualization license, and the three extra meetings per week where teams argue about whether the numbers are right? That friction is a direct cost of low trust.

• It frames the conversation around impact, not just output. Data Effectiveness isn't just about whether we built a thing; it's about whether that thing effectively solved a problem. (think Amazon's work-backwards Product Management) Did that new model actually reduce customer churn, or did it just get deployed? Did that certified dataset actually speed up time-to-insight for the marketing team, or does it just sit there?

Calculating this isn't an academic exercise. It's how you build trust with your CFO. It's about creating a direct, defensible line from every dollar invested in the data platform to tangible business outcomes - revenue growth, cost savings, productivity gains, and risk mitigation. When you can prove your data's effectiveness, you're no longer asking for budget; you're presenting a profitable investment backed by a culture of trust.

Dimension 3: Trust in Your Data Platform, Teams, and Partners

Once you have confidence in the data itself, the final, crucial dimension of trust comes into focus: trust in the entire engine room responsible for managing that data. This includes the technology platform, the internal teams who run it, and the external partners who support it. This is where execution lives or dies.

1. Trust in Your Own Team

Leadership as the Source of Trust. All external trust is rooted in the trust that exists within the data team itself. A leader cannot expect the organization to trust their data if they do not first cultivate a high-trust environment for their own people.

This internal trust is built on several key leadership imperatives:

• Provide a Clear Vision and an Executable Plan: A team needs a clear direction and a believable strategy for getting there. Remember - lead with a clear vision, and lead with a clear impact.

• Empower and Enable: Leaders must listen to their teams and empower them to take ownership and make decisions, creating an environment where they feel safe to innovate and even fail.

• Demonstrate Technical Credibility: A leader must be able to engage in detailed technical discussions and understand the nuances of a design, while never losing sight of the broader business objectives.

• Be Available and Supportive: A leader's role is to serve their team, removing obstacles and providing the resources and support they need to succeed.

2. Trust in Your Platform & Internal Teams

A platform is nothing without the people behind it. Building trust here means fundamentally shifting the role of a data platform team from that of a gatekeeper to an enabler.

- For Your Data Producers

Think about the engineers building your core applications. For them, the data platform can often feel like a black hole. To build trust, the platform must be positioned as a trusted steward, guaranteeing that the data they produce will be handled with integrity. This is why the recent platform wars are getting so personal. The strategic acquisitions of PostgreSQL providers - Snowflake buying Crunchy Data and Databricks acquiring Neon and their release of Lakebase - aren't just technical moves. They are massive, billion-dollar investments in trust, signaling to developers that the platforms are serious about supporting the transactional and AI-native apps they want to build. (btw, who buys Supabase next?)

- For Your Domain Owners and Consumers

As we scale, the old model of a central data team acting as a bottleneck just doesn't work. It kills speed and erodes trust. Adopting a Data Mesh philosophy - where you push ownership out to the domain experts - is the answer, but it's not a free-for-all. True empowerment, and the trust that comes with it, is built on a few promises from the platform team:

• We'll give you a great self-service platform, not a list of rules. Our job is to be enablers.

• We'll provide radical transparency. To own their data, teams need full visibility. This means top-notch observability into data health and crystal-clear lineage. No black boxes.

• The platform will be rock-solid. Empowerment is meaningless if the foundation is shaky. We have to commit to operational and engineering excellence - reliability, scalability, and resilience. Your data products should never fail because our platform did.

• We promise not to break your stuff. This one is simple but critical. The platform team can't make platform changes that pull the rug out from under our domain teams. That's a catastrophic breach of trust that requires robust, predictable change management.

3. Trust with Your Product, Privacy and Legal Partners

This is a big one for me. We have to treat our product, legal and privacy teams as design partners, not as a final, feared checkpoint. Trust is built when you engage them early, showing that you're trying to find the right balance between innovating and being compliant, not just trying to get away with something.

• Always remember they're your partners in crime and people you should count on to push the boundary and unlock the most value of your data without losing your customer's trust.

• The ultimate test of this trust, however, is what happens when things go wrong. A well-rehearsed incident response plan is the single greatest trust-builder. It’s not just about having a plan on paper; it’s about having the "muscle memory" from practicing it, so that when a crisis hits, your response is defined by immediate ownership and transparent communication. That’s how you show you can be trusted to do the right thing, -even- especially when it’s hard.

4. Trust with Business Leaders and Stakeholders

Earning trust with business leaders and stakeholders isn't about showing them your cool new tech; it's about proving you're a valuable partner in driving their success. They don't think in terabytes; they think in P&L. Here's what I mean:

• Deliver on a Foundation of Understanding. The first step is simple but so often missed: truly understand their needs. Be genuinely helpful. Your job is to make their job easier and their outcomes better. Trust begins when they see you as someone who delivers value, not just someone who manages a cost center.

• Pitch Investments, Not Expenses. I've said this before, and I'll say it again: stop asking for money for technology. Instead, pitch an investment in solving a business problem. Start with a clear, quantifiable pain point they care about, frame your data initiative as the direct solution, and use a framework like "Data Effectiveness" to prove the ROI. It's about accountability and partnership.

• Build a Platform for Their Future, Not Just Their Present. It's not enough to solve today's problem. You earn deep, lasting trust by showing you're building a platform that anticipates their future needs. This means architecting for scalability, flexibility, and the next wave of innovation. When you show them you're not just a service provider but a strategic partner invested in their long-term success, you build a foundation of trust that is incredibly difficult to break.

5. Trust with Platform Partners & Community

Finally, your data platform doesn't exist in a vacuum. You have to earn trust with the entire ecosystem. Data is a commodity, and the platforms are getting commoditized even faster. In this world, you can't build behind closed doors.

• Be Open with Your Partners. Treat your hyperscalers and tech partners like extensions of your own team. Share your designs and let them poke holes in your architecture. Their feedback is invaluable, and that level of vulnerability builds a deep, collaborative trust that goes beyond a simple vendor relationship.

• Embrace the Community. Don't just adopt open-source technology; contribute back. Whether it's through code, documentation, or just sharing your learnings, giving back to the community that you benefit from is a powerful way to build credibility and establish your team as a trusted leader in the space.

My Final Take: It All Comes Down to This

We've spent a lot of time talking about the different dimensions of trust, and it's easy to get lost in the details of governance, platforms, and ROI calculations. But I want to zoom out for a second and leave you with one final thought.

For years, we've treated data as an asset to be collected and hoarded. We've built bigger and bigger warehouses, more and more complex pipelines, all in the pursuit of accumulating more data. But that era is over. In a world where AI can generate infinite information, the value is no longer in the data itself.

The value is in the decisions (with human in the loop or not) that data enables.

The last real, sustainable competitive advantage isn't your algorithm, your data volume, or your tech stack. It's your organization's ability to execute with speed, agility, and unwavering confidence. It's the speed at which your teams can move from a question to an answer to an action. It's the confidence your leaders have to make bold bets, knowing the information they're standing on is solid ground, not quicksand.

That capacity for execution is a direct function of trust.

Trust is what turns data from a liability into a lubricant. It eliminates the friction of doubt, the tax of verification, and the paralysis of uncertainty. It's the invisible infrastructure that allows talent, technology, and strategy to finally click into place.

So my challenge to you, and to myself, is this: stop architecting just for data. Stop architecting just for AI.

Start architecting for trust.

Because in the end, that's the only thing that will truly set you apart.

I’d love to hear your thoughts. Where do you see the biggest trust gaps, and what are you doing to close them?