Salut. Je m’appelle Michaël Lozano. Pendant six ans, j’ai travaillé sur Palantir Foundry chez Airbus, via Capgemini. J’ai livré plus de 80 produits data. 98% de disponibilité. 4.8/5 de satisfaction client. Des chiffres qui claquent, mais derrière, il y a des nuits blanches, des pipelines qui cassent à 3h du mat’, et des réunions interminables pour expliquer pourquoi le modèle ne marche pas si les données d’entrée sont pourries.

Aujourd’hui, je suis freelance. Et j’ai construit un outil open-source : DataFlow Canvas. C’est un visual pipeline builder pour Foundry. Drag-and-drop, génération automatique de code PySpark avec les fameux `@transform_df`, gestion de la lineage, collaboration en temps réel via Firebase. Bref, l’outil que j’aurais rêvé avoir quand je débutais sur Skywise.

Mais avant de vous parler de DataFlow Canvas, il faut qu’on parle d’un problème plus large. Un problème structurel. Pourquoi la France manque cruellement d’experts Palantir Foundry. Et pourquoi, si vous cherchez quelqu’un qui sait vraiment ce qu’il fait, vous tombez souvent sur des profils… approximatifs.

Foundry, c’est pas du Python qu’on apprend à la fac

Palantir Foundry, c’est une plateforme. Une grosse plateforme. Pas juste un Jupyter Notebook avec un joli logo. C’est un écosystème complet : Ontology, Code Workbooks, Foundry ML, Pipeline Builder, Contour, Workshop, Slate… La liste est longue. Et chaque module a ses propres subtilités, ses pièges, ses bonnes pratiques qui ne sont écrites nulle part.

On n’apprend pas Foundry à l’école. Il n’y a pas de cours « Introduction à Foundry » à Télécom Paris ou à l’INSA. Il n’y a pas de certification officielle reconnue par l’industrie (les certifications Palantir existent, mais elles sont internes, réservées aux employés Palantir et à leurs partenaires directs). Résultat ? On apprend Foundry sur le tas. En cassant des choses. En lisant la documentation en diagonale à 2h du matin parce que le pipeline de prod est tombé. En demandant à un collègue qui, lui aussi, apprend sur le tas.

C’est un cycle vicieux. Les entreprises veulent des experts Foundry. Mais personne ne forme les experts Foundry. Donc les seuls « experts » disponibles sont ceux qui ont eu la chance (ou la malchance) de tomber sur un projet Foundry dans leur carrière. Et même parmi eux, beaucoup ont une connaissance superficielle. Ils savent cliquer dans l’interface. Ils savent écrire un `@transform_df` basique. Mais dès qu’il faut optimiser un pipeline qui traite 10 To de données, ou déboguer un problème de lineage cassé, ou architecturer une Ontology qui tient la route à l’échelle… là, il n’y a plus personne.

Mon expérience chez Airbus : Skywise et la maturité data de l’A350

Chez Airbus, j’ai travaillé sur Skywise, la plateforme data collaborative d’Airbus. Mon rôle ? Construire des produits data pour la maintenance des avions. Concrètement : prendre des téraoctets de données de capteurs, de logs de maintenance, de plans de vol, et les transformer en insights actionnables pour les compagnies aériennes.

Un exemple concret. Sur le programme A350, on a travaillé sur la maturité data des avions. Chaque avion génère des milliers de paramètres en vol. La question n’était pas « est-ce qu’on a les données ? » (on les avait). La question était : « est-ce que ces données sont fiables, complètes, et exploitables pour faire de la maintenance prédictive ? »

On a construit des pipelines Foundry pour évaluer la qualité des données de chaque avion. Un pipeline qui, pour chaque vol, vérifiait : est-ce que tous les capteurs essentiels ont envoyé des données ? Y a-t-il des valeurs aberrantes ? Des trous temporels ? Et on alimentait un dashboard qui donnait un score de maturité data par avion.

Ce genre de projet, c’est typiquement là où Foundry brille. Mais c’est aussi là où Foundry punit les architectes négligents. Un pipeline mal conçu, avec des joins mal optimisés ou des transformations inutiles, et vous vous retrouvez avec un job qui tourne pendant 12 heures au lieu de 30 minutes. Sur un environnement partagé, ça bloque tout le monde. Et croyez-moi, quand vous bloquez 50 data engineers chez Airbus, vous recevez des appels pas très sympas.

J’ai vu passer des dizaines de consultants sur ces projets. Beaucoup partaient au bout de 6 mois en disant « je connais Foundry ». Mais est-ce qu’ils comprenaient vraiment l’Ontology ? Est-ce qu’ils savaient pourquoi `@transform_df` était préférable à un code workbook brut pour la gouvernance ? Est-ce qu’ils maîtrisaient les mécanismes de caching de Foundry pour éviter de recalculer les mêmes datasets 10 fois ? Pour la plupart, non. Ils avaient appris à survivre. Pas à maîtriser.

Pourquoi c’est si dur de devenir un vrai expert Foundry

Foundry, c’est comme un iceberg. La partie visible, c’est l’interface web. C’est joli, c’est intuitif, on clique, ça marche (souvent). La partie immergée, c’est tout ce qui se passe en dessous :

– L’Ontology, qui est le cœur métier de Foundry. Si vous modélisez mal votre Ontology, tout ce que vous construisez dessus sera bancal. C’est comme construire une maison sur du sable. – Le système de builds, avec ses caches, ses invalidations, ses dépendances. Comprendre pourquoi un rebuild ne se déclenche pas, ou pourquoi il rebuild tout alors qu’il ne devrait rebuild qu’un seul dataset, c’est un art. – La sécurité et la gouvernance. Foundry a un modèle de permissions complexe. Comprendre les différences entre les permissions au niveau du dataset, de l’Ontology object, et du code workbook, c’est essentiel pour ne pas accidentally donner accès à des données sensibles à toute l’organisation. – L’optimisation des performances. Foundry tourne sur Spark en backend. Si vous ne comprenez pas comment Spark fonctionne (shuffles, partitions, skew), vos pipelines seront lents. Point final.

Tout ça, ce n’est pas documenté de manière centralisée. Il faut accumuler de l’expérience. Et l’expérience, ça prend du temps. Des années.

DataFlow Canvas : l’outil que j’aurais voulu avoir

C’est pour ça que j’ai construit DataFlow Canvas.

DataFlow Canvas, c’est un visual pipeline builder open-source pour Foundry. L’idée est simple : au lieu de devoir écrire tout le code PySpark à la main, vous dessinez votre pipeline sur un canvas. Vous drag-and-drop des datasets, des transformations, des sorties. Et l’outil génère automatiquement le code Foundry correspondant, avec les bons decorators `@transform_df`, la bonne gestion des inputs et outputs, et même la documentation de base.

Mais DataFlow Canvas, c’est pas juste un générateur de code. C’est aussi :

– Un moteur d’architecture IA. Vous décrivez un besoin métier en langage naturel (« je veux joindre les données de vol avec les données de maintenance et calculer le temps moyen entre deux pannes »), et l’IA vous propose un squelette de pipeline. Vous validez, vous ajustez, et hop, le code est généré. – De la collaboration en temps réel. Via Firebase, plusieurs personnes peuvent travailler sur le même canvas en même temps. Plus besoin de s’envoyer des fichiers JSON par email ou de se battre sur un Git repo. – De la gestion de lineage et de versioning. Chaque modification est trackée. Vous pouvez voir l’historique de votre pipeline, revenir en arrière, comparer les versions. – Un catalog de transformations pré-built. Joins, filtres, agrégations, window functions… Les transformations les plus courantes sont disponibles en un clic. Plus besoin de retaper les mêmes bouts de code encore et encore.

DataFlow Canvas, c’est mon cadeau à la communauté Foundry française (et mondiale). C’est open-source, c’est gratuit, et j’espère que ça aidera les équipes à accélérer leur adoption de Foundry tout en évitant les pièges classiques.

La France a besoin d’experts Foundry. Vrais.

Si vous lisez cet article, c’est peut-être que vous cherchez un expert Foundry. Ou que vous voulez devenir un expert Foundry. Dans les deux cas, voici mon conseil :

Ne vous contentez pas de l’interface web. Creusez. Comprenez ce qui se passe en dessous. Lisez la documentation (oui, elle est longue). Expérimentez. Cassez des choses en environnement de dev. Demandez à des gens qui ont de l’expérience. Et si vous ne trouvez personne… contactez-moi.

Je suis freelance. Je suis spécialisé sur Palantir Foundry. J’ai 6 ans d’expérience, 80+ produits livrés, et une obsession pour la qualité des architectures data. Je parle français, anglais, espagnol, et italien. Je suis certifié SAFe 5 Agilist et PSPO I. Et surtout, je suis passionné par Foundry au point d’avoir construit un outil open-source pour faciliter son utilisation.

Si vous avez un projet Foundry, si vous voulez auditer votre architecture, si vous avez besoin de former vos équipes, ou si vous voulez juste discuter de la meilleure façon de modéliser votre Ontology… parlons-en.

📩 Contactez-moi via studioinitiative.com

🔗 Retrouvez-moi sur LinkedIn

🛠️ Et allez voir DataFlow Canvas sur GitHub — c’est gratuit, c’est open-source, et ça pourrait bien vous faire gagner des semaines de travail.

—

Michaël Lozano est Senior Product & Delivery Lead Data, expert Palantir Foundry avec 6+ années d’expérience chez Airbus (Capgemini). Il accompagne maintenant les entreprises dans leur transformation data via Studio Initiative.

L’article J’ai construit un Pipeline Builder pour Palantir Foundry — Voici pourquoi la France manque d’experts est apparu en premier sur Studio Initiative.

You’ve probably wondered why some approaches work better than others?
In the world of technological implementation, it’s not always what you think.

The Problem with the Traditional Approach

Many companies treat their implementations as simple IT projects:
fixed specifications, waterfall planning, a single delivery after several months of development.
This approach completely overlooks the very nature of what certain platforms enable:
a continuous evolution where value is created through iteration, not in the final delivery.

The Product Mindset: Delivering Value Early and Often

A product approach completely transforms this dynamic:

1. Start with a precise business use case

Not « building a platform », but « solving a concrete business problem ».

1. Deliver every 2-4 weeks

Each iteration brings a measurable improvement.

1. Measure adoption, not just technical completion

The real KPI is not « the project is complete » but « actual usage ».

1. Integrate feedback into the immediate backlog

User workshops become the prioritization for the next sprint.

What the Data Shows

1. According to (PDF) AI Powered Data Governance -Ensuring Data Quality and …:
   This article explores how AI and ML can automate compliance checks, detect anomalies, track data lineage, and streamline validation processes.

2. According to How Data Governance Improves AI Success? – Medium:
   Data governance ensures that AI models are developed on high-quality, trustworthy, and governed data, cutting down risks, improving decision

Why This Works Particularly Well

Some platforms are designed for iteration:
their components can evolve without breaking what already exists,
enabling rapid and secure deployment.

Concrete Result: From Resistance to Enthusiastic Adoption

In a recent mission, we went from
less than 20% adoption with the traditional approach
to over 85% in under 4 months with the product approach.
The difference? Listen, adapt, deliver frequently.

Your Next Step

If you’re considering an implementation or have already started
but adoption is struggling, seriously consider this product approach.
It may require more discipline in terms of measurement and frequent feedback,
but the return on investment in terms of real business value is incomparable.

Want to discuss how to apply this product mindset to your specific context?
I offer a free 30-minute strategic call to explore your current challenges
and see how a product delivery approach could transform your initiative.

Schedule your discovery call

L’article Data Governance: Ensuring Traceability and Quality in AI-Driven Projects est apparu en premier sur Studio Initiative.

You’ve probably wondered why some approaches work better than others?
In the world of technological implementation, it’s not always what you think.

The Problem with the Traditional Approach

Many companies treat their implementations as simple IT projects:
fixed specifications, waterfall planning, a single delivery after several months of development.
This approach completely overlooks the very nature of what certain platforms enable:
a continuous evolution where value is created through iteration, not in the final delivery.

The Product Mindset: Delivering Value Early and Often

A product approach completely transforms this dynamic:

1. Start with a precise business use case

Not « building a platform », but « solving a concrete business problem ».

1. Deliver every 2-4 weeks

Each iteration brings a measurable improvement.

1. Measure adoption, not just technical completion

The real KPI is not « the project is complete » but « actual usage ».

1. Integrate feedback into the immediate backlog

User workshops become the prioritization for the next sprint.

What the Data Shows

1. According to Building an Autonomous AI Assistant with a Heartbeat – LinkedIn:
   I gave my AI assistant a heartbeat. Not a metaphor — a literal cron job that wakes it up 5 times a day and says: « What needs attention?

2. According to How to Build a Self-Maintaining AI System with Heartbeat and Wrap …:
   Instead of relying on a human to re-prime an agent before every run, you build two complementary skills into the agent itself: a heartbeat scan that periodically syncs the agent’s knowledge of the…

Why This Works Particularly Well

Some platforms are designed for iteration:
their components can evolve without breaking what already exists,
enabling rapid and secure deployment.

Concrete Result: From Resistance to Enthusiastic Adoption

In a recent mission, we went from
less than 20% adoption with the traditional approach
to over 85% in under 4 months with the product approach.
The difference? Listen, adapt, deliver frequently.

Your Next Step

If you’re considering an implementation or have already started
but adoption is struggling, seriously consider this product approach.
It may require more discipline in terms of measurement and frequent feedback,
but the return on investment in terms of real business value is incomparable.

Want to discuss how to apply this product mindset to your specific context?
I offer a free 30-minute strategic call to explore your current challenges
and see how a product delivery approach could transform your initiative.

Schedule your discovery call

L’article Leveraging Heartbeat Routines for Continuous Self-Improvement in AI Assistants est apparu en premier sur Studio Initiative.

You’ve probably wondered why some approaches work better than others?
In the world of technological implementation, it’s not always what you think.

The Problem with the Traditional Approach

Many companies treat their implementations as simple IT projects:
fixed specifications, waterfall planning, a single delivery after several months of development.
This approach completely overlooks the very nature of what certain platforms enable:
a continuous evolution where value is created through iteration, not in the final delivery.

The Product Mindset: Delivering Value Early and Often

A product approach completely transforms this dynamic:

1. Start with a precise business use case

Not « building a platform », but « solving a concrete business problem ».

1. Deliver every 2-4 weeks

Each iteration brings a measurable improvement.

1. Measure adoption, not just technical completion

The real KPI is not « the project is complete » but « actual usage ».

1. Integrate feedback into the immediate backlog

User workshops become the prioritization for the next sprint.

What the Data Shows

1. According to Building Proactive Multimodal AI Agents by Shai Alon – YouTube:
   … proactive, autonomous partners capable of understanding and acting across text, image, video, and audio. Shai breaks down the ‘Agent

2. According to Building proactive AI agents – by Bryan Houlton – Orin Labs:
   We needed a combination of workflows and triggers that allowed Orin to be proactive and dynamic, not stagnant and static. Orin is now somewhat proactive (doesn’t require the user to press a button…

Why This Works Particularly Well

Some platforms are designed for iteration:
their components can evolve without breaking what already exists,
enabling rapid and secure deployment.

Concrete Result: From Resistance to Enthusiastic Adoption

In a recent mission, we went from
less than 20% adoption with the traditional approach
to over 85% in under 4 months with the product approach.
The difference? Listen, adapt, deliver frequently.

Your Next Step

If you’re considering an implementation or have already started
but adoption is struggling, seriously consider this product approach.
It may require more discipline in terms of measurement and frequent feedback,
but the return on investment in terms of real business value is incomparable.

Want to discuss how to apply this product mindset to your specific context?
I offer a free 30-minute strategic call to explore your current challenges
and see how a product delivery approach could transform your initiative.

Schedule your discovery call

L’article Building Proactive AI Agents: Lessons from the Agent Autonomy Kit est apparu en premier sur Studio Initiative.

You’ve probably wondered why some approaches work better than others?
In the world of technological implementation, it’s not always what you think.

The Problem with the Traditional Approach

Many companies treat their implementations as simple IT projects:
fixed specifications, waterfall planning, a single delivery after several months of development.
This approach completely overlooks the very nature of what certain platforms enable:
a continuous evolution where value is created through iteration, not in the final delivery.

The Product Mindset: Delivering Value Early and Often

A product approach completely transforms this dynamic:

1. Start with a precise business use case

Not « building a platform », but « solving a concrete business problem ».

1. Deliver every 2-4 weeks

Each iteration brings a measurable improvement.

1. Measure adoption, not just technical completion

The real KPI is not « the project is complete » but « actual usage ».

1. Integrate feedback into the immediate backlog

User workshops become the prioritization for the next sprint.

What the Data Shows

1. According to AI Ethics: What Is It and Why It Matters for Your Business:

AI Ethics: what it is and why it matters for your business. In this guide, we’ll explore what AI ethics entails, why it’s important for your business, and how you can implement ethical practices to …

1. According to AI and Ethical Leadership: Why Business Ethics Must Evolve:
   AI and ethical leadership require technological literacy alongside moral awareness. Leaders must recognise that ethical risk is often embedded

Why This Works Particularly Well

Some platforms are designed for iteration:
their components can evolve without breaking what already exists,
enabling rapid and secure deployment.

Concrete Result: From Resistance to Enthusiastic Adoption

In a recent mission, we went from
less than 20% adoption with the traditional approach
to over 85% in under 4 months with the product approach.
The difference? Listen, adapt, deliver frequently.

Your Next Step

If you’re considering an implementation or have already started
but adoption is struggling, seriously consider this product approach.
It may require more discipline in terms of measurement and frequent feedback,
but the return on investment in terms of real business value is incomparable.

Want to discuss how to apply this product mindset to your specific context?
I offer a free 30-minute strategic call to explore your current challenges
and see how a product delivery approach could transform your initiative.

Schedule your discovery call

L’article Ethics in AI: Why It Matters for Business Leaders est apparu en premier sur Studio Initiative.

You’ve probably wondered why some approaches work better than others?
In the world of technological implementation, it’s not always what you think.

The Problem with the Traditional Approach

Many companies treat their implementations as simple IT projects:
fixed specifications, waterfall planning, a single delivery after several months of development.
This approach completely overlooks the very nature of what certain platforms enable:
a continuous evolution where value is created through iteration, not in the final delivery.

The Product Mindset: Delivering Value Early and Often

A product approach completely transforms this dynamic:

1. Start with a precise business use case

Not « building a platform », but « solving a concrete business problem ».

1. Deliver every 2-4 weeks

Each iteration brings a measurable improvement.

1. Measure adoption, not just technical completion

The real KPI is not « the project is complete » but « actual usage ».

1. Integrate feedback into the immediate backlog

User workshops become the prioritization for the next sprint.

What the Data Shows

1. According to AI & Business Analyst Role: Augmentation, Skills, & Future Strategy:
   The role of the Business Analyst (BA) has always been pivotal, acting as the critical bridge between business needs and technical execution.

2. According to The augmented business analyst with generative AI – Thilo Hermann:
   Gen AI is augmenting various business analysis tasks, focusing on writing epics, user stories and acceptance criteria, creating use cases and personas,

Why This Works Particularly Well

Some platforms are designed for iteration:
their components can evolve without breaking what already exists,
enabling rapid and secure deployment.

Concrete Result: From Resistance to Enthusiastic Adoption

In a recent mission, we went from
less than 20% adoption with the traditional approach
to over 85% in under 4 months with the product approach.
The difference? Listen, adapt, deliver frequently.

Your Next Step

If you’re considering an implementation or have already started
but adoption is struggling, seriously consider this product approach.
It may require more discipline in terms of measurement and frequent feedback,
but the return on investment in terms of real business value is incomparable.

Want to discuss how to apply this product mindset to your specific context?
I offer a free 30-minute strategic call to explore your current challenges
and see how a product delivery approach could transform your initiative.

Schedule your discovery call

L’article The Role of a Business Analyst in AI-Agent Augmented Teams est apparu en premier sur Studio Initiative.

The problem with the « project » approach in Foundry implementations

Many companies treat their Palantir Foundry implementations as simple IT projects: fixed specifications, waterfall planning, a single delivery after several months of development. This approach completely overlooks the very nature of what Foundry enables: a continuous evolution platform where value is created through iteration, not in the final delivery.

I’ve seen too many teams spend 6 months building a perfect ontology… only to discover that business needs had fundamentally changed during that time. The result? A technically flawless platform that remains desperately empty because no one actually uses it.

The product mindset: delivering value early and often

A product approach completely transforms this dynamic:

1. Start with a precise business use case
   Not « building a data platform », but « reducing supply chain incident resolution time by 30% in 8 weeks »

2. Deliver every 2-4 weeks
   Each iteration brings a measurable improvement: a new operational view, an automated ingestion pipeline, an AIP application for a specific process

3. Measure adoption, not just technical completion
   The real KPI is not « the ontology is complete » but « how many active users use this feature each week? »

4. Integrate feedback into the immediate backlog
   Weekly user workshops become the prioritization for the next sprint, not a end-of-project review

Why this works particularly well with Foundry

Foundry is designed for iteration:
– The ontology can evolve without breaking existing applications (thanks to the separation of business/technical)
– AIP allows rapid deployment of operational interfaces
– Ingestion pipelines are naturally modular and scalable
– Security and governance are applied as the platform expands

Concrete result: from resistance to enthusiastic adoption

In a recent mission for an industrial European client, we went from:
– Months 1-3: Long discovery workshops, detailed specifications, low user engagement
– Months 4-6: Product approach with biweekly deliveries, weekly user workshops
– Result: 85% adoption among operational teams in less than 4 months, versus less than 20% with the traditional project approach

Your next step

If you’re considering a Foundry implementation or you’ve already started but adoption is struggling, seriously consider this product approach. It may require more discipline in terms of measurement and frequent feedback, but the return on investment in terms of real business value is incomparable.

Want to discuss how to apply this product mindset to your specific context? I offer a free 30-minute strategic call to explore your current challenges and see how a product delivery approach could transform your Foundry initiative.

Schedule your discovery call

If you want me to help you optimize the SEO title, meta-description, or add JSON‑LD schema for this post, let me know. Otherwise, we can move on to other things: daily freelance mission search, improving your internal processes, or simply discussing your next goals. 😄

What are we doing now?

L’article Why Product Delivery Beats Traditional Project Management in Palantir Foundry Implementations est apparu en premier sur Studio Initiative.

1. Start with Conversations, Not Just Catalogs

Too often, data governance begins with inventorying assets—creating a catalog of tables, fields, and sources. While knowing what data you have is important, it misses the point if you don’t also understand how people actually use that data. The most effective data governance efforts I’ve seen start by sitting down with analysts, engineers, and business stakeholders to ask: What data do you trust? What keeps you up at night? Where do you waste time chasing down answers?

Example: At a manufacturing client, we discovered that the finance team didn’t trust the sales forecast data—not because it was inaccurate, but because they couldn’t see how it was calculated. By opening up the calculation logic and inviting finance to co-design the governance rules around assumptions and refresh cycles, trust increased dramatically. The catalog was useful, but the conversation was transformative.

2. Make Policies Human-Readable (and Human-Centered)

Governance documents often read like legal contracts—dense, intimidating, and full of « shalls » and « must-nots. » If people can’t understand the rules, they can’t follow them, and they certainly won’t feel ownership over them. Good governance translates policy into plain language, with clear examples of what it looks like in practice.

Example: Instead of a policy stating « Data shall be classified according to sensitivity levels, » we created a simple guide with pictures and scenarios: « If your data contains customer names and purchase history, treat it like a locked file cabinet—only share with those who need it, and always lock it when you’re away from your desk. » We posted these guides in team spaces and saw a noticeable drop in accidental data exposure.

3. Embrace Transparency as a Two-Way Street

Transparency isn’t just about making data lineage visible or publishing quality metrics. It’s about creating feedback loops where people can question, challenge, and improve the governance itself. When people see that their input leads to real changes, they become advocates rather than reluctant compliance subjects.

Example: We implemented a monthly « data office hours » session where anyone could bring up concerns about data definitions, access requests, or dashboard confusion. One month, a junior analyst pointed out that our customer churn metric was being calculated differently across teams, leading to conflicting reports. We traced the discrepancy to a legacy field that no one had documented. Fixing it not only improved consistency but also gave the analyst a sense of impact—and encouraged others to speak up.

4. Celebrate the People Behind the Data

Finally, we must remember that data doesn’t appear out of thin air. It’s collected, cleaned, and maintained by real people—often working behind the scenes. Acknowledging their effort builds camaraderie and reinforces that data governance is a shared responsibility, not a policing function.

Example: Our team started a monthly « Data Hero » shout-out in our newsletter, highlighting someone who went the extra mile to improve data quality or help a colleague understand a complex dataset. It was surprising how much this simple recognition boosted morale and encouraged collaborative problem-solving across departments.

Conclusion

Data governance succeeds when it serves people, not the other way around. By leading with empathy, fostering open dialogue, and making transparency a lived practice—not just a policy—we create environments where data is trusted, understood, and put to good use. The technology is important, but the human side is what makes it work.

Next time you’re drafting a data governance framework, ask yourself: How does this help the person using this data do their job better? Start there, and the rest will follow.

L’article The Human Side of Data Governance: Building Trust Through Transparency est apparu en premier sur Studio Initiative.

1. Speak the language of business : Forget technical terms like « pipeline » or « ML model ». Focus on ROI, risk reduction, and growth opportunities. Example: « This project will reduce inventory costs by 15% » instead of « We’re optimizing the ETL ».
2. Start small with a proof of concept : Propose a limited-time, limited-budget pilot project to demonstrate value quickly. A quick win builds credibility for more ambitious initiatives.
3. Use clear, impactful visualizations : A simple dashboard with 3-4 key KPIs is worth more than a 50-page report. Use colors, trends, and clear comparisons.
4. Align with strategic goals : Show how your data project directly supports the company’s annual priorities (e.g., international expansion, cost reduction, customer satisfaction improvement).
5. Involve management from the definition : Don’t present them with a finished project. Co-create the scope with them so they take ownership from the start.

By applying these principles, you’ll transform management from passive observers into active champions of your data projects. Your role isn’t to build technology—it’s to facilitate the adoption of data solutions that create business value.

L’article How to Get Management Buy-In for Your Data Projects: 5 Proven Strategies est apparu en premier sur Studio Initiative.

We have all experienced these rituals that resemble corporate theater. The 9 AM stand-up where everyone recites their little speech. The retrospective where we stick colorful post-its only to change nothing in the end. The sprint planning that lasts three hours and results in totally fanciful estimates. We apply the Scrum guide rules with religious fervor, convinced that the methodology will save the project. Spoiler: it is not the methodology that saves projects. It is the quality of human interactions and the clarity of objectives.

The problem is that agile has become a product. Certifications that cost a fortune, frameworks that pile up, coaches selling digital transformation at a high price. As a result, we end up with teams pretending to be agile while keeping the worst aspects of waterfall. We still plan six months ahead, change our minds at the last minute, and ask teams to deliver within unrealistic deadlines. Except now we call them sprints, so it is supposed to be different. It is not different. It is just more expensive and more exhausting.

I see this regularly in data and product teams. They undergo processes designed for pure software developers, without adaptation to their reality. A data team does not work like a frontend team. Tasks are not always decouplable into homogeneous small batches. Some analyses take the time they take. Some explorations lead in unexpected directions. Forcing all of this into two-week sprints with story points is organizational masochism.

This is where Scrumban comes in. And no, it is not just a buzzword to sound modern.

Scrumban takes what really works in Scrum, namely the regularity of rituals and the visibility of work in progress, and combines it with the fluidity of Kanban. No more rigid sprints. No more mandatory estimates that make no sense. No more artificial demonstrations just to say we did a demo. Instead, we have a continuous flow, work in progress limits that prevent the team from scattering, and rituals that only exist if they bring value.

The beauty of Scrumban is its flexibility. You keep your stand-up if you need it, but you shorten it to ten minutes maximum. You keep planning, but it becomes continuous rather than calendar-based. When a task is ready, you take another one. Period. No artificial stress linked to the end of a sprint. No forced demo of a half-finished feature just because it is Friday and the sprint is ending.

For data and product teams, it is often a relief. These teams need to explore, iterate, change priorities when new data calls the analysis into question. Scrumban gives them this permission without guilt. The Kanban board remains visible, transparent to everyone. Blockages are identified quickly. But the team is no longer a slave to an artificial rhythm that does not match their actual work.

So when to choose Scrumban over pure Scrum? As soon as your work is unpredictable, exploratory, or subject to frequent interruptions. If you spend your time saying this sprint is doomed, we had an emergency, Scrumban is for you. If your tasks vary enormously in size and estimating takes longer than doing the task itself, stop torturing yourself. If your team is mature and does not need a strict framework to move forward serenely, free them from these unnecessary constraints.

Pure Scrum makes sense when you are delivering software with predictable releases, a stable roadmap, and a team that needs protection from interruptions. It is a valid tool, but it is not the only possible answer.

At the core, the real question is not Scrum or Scrumban or Kanban. The real question is: what allows this specific team to deliver value sustainably? Sometimes the answer will be Scrum. Often, for data and product teams today, it will be Scrumban. And sometimes it will be something you invent yourself, because no existing methodology perfectly matches your context.

Pragmatism is the only true agile. Everything else is just corporate religion.

And you, is your team still trapped in rituals that serve no one? What would stop you from experimenting with something else starting next week?

L’article Why Agile Has Become Corporate Theater and What to Do About It (Spoiler : Scrumban) est apparu en premier sur Studio Initiative.