Enterprise AI roadmaps are full of GenAI initiatives. Few of them make it out of pilot mode, and the reasons rarely come down to the model, the budget, or leadership support.
That pattern emerged repeatedly during our Integrate Forward Event with SnapLogic, hosted at the Siemens Auditorium in Erlangen, featuring some of our largest global enterprise customers in the region. Across organizations and industries, the same gap persists: the distance between where data and processes currently stand and what AI at scale demands.
The practitioners in the room were the ones running these transformations. What follows are the sharpest lessons from the day, each with something concrete to act on before your next planning cycle.
1. Point-to-point, manual integration doesn’t scale
Bachem is a leading Swiss biotechnology and pharmaceutical company specializing in the development and manufacture of peptides and oligonucleotides. Paul Endras, Senior Data & AI Platform Manager at Bachem, argues that organizations need to shift to an “integration factory” mindset, treating integration as a repeatable production discipline (standards, reusable patterns) rather than one-off projects, with clear data/system ownership assigned from day one so accountability doesn’t evaporate when something breaks.
Buying a good platform is necessary but not sufficient. You also need skills, ongoing stakeholder education, and a willingness to redesign business processes around the tool.
In regulated environments like GMP pharma, this discipline is audit survival: you need real end-to-end data lineage. And the punchline tying it together: integration maturity is the prerequisite for AI maturity. You can’t bolt predictive analytics or chatbots onto a data landscape that’s still held together with Excel.
2. Data readiness first: the foundation AI initiatives actually stall without
Jakub Szpetkowski (Data and AI Transformation Lead), Amelie Soares (Senior Data & AI Consultant), and Loic Kuntz (Data and Machine Learning Engineer) from Data Reply argue that when AI initiatives stall, the cause is rarely budget, buy-in, or the wrong tool. It’s data that isn’t mature enough for the use case at hand.
Scaling further requires a real foundation first: clear data ownership, accountability for quality, structured portfolios, and governance that’s actually followed, not just documented. Fixing these gaps up front makes every later use case cheaper and faster to deliver, which is what justifies the investment.
Ranking a use case portfolio by business value alone is a trap. Without scoring technical feasibility and data quality too, you risk funding projects that can’t scale. And asking if the company is AI-ready is the wrong question. Evaluating maturity domain by domain gives you an actual plan instead of a vague verdict.
Finally, building every pipeline independently is a maintenance trap. Reusable, configurable sub-pipelines on a visual, no-code platform speed up execution, simplify debugging, and make onboarding new developers far easier.
3. Flip the order: adopt first and deploy second
The panel “AI in the Workforce” brought together leaders like Carolin Franz (SVP and Head of Human Resources EMEA, APJ, and China, Siemens Healthineers), Robin Mager (VP Digital Transformation Office, Siemens Energy) and Gustav Rek (VP AI, Perbility) amongst others.
Moderated by Jeremiah Stone (CTO, SnapLogic), the discussion explored how AI is reshaping jobs and skills, and what organisations need to turn adoption into lasting change.
Teams that succeed tend to build adoption into the plan from day one, rather than treating it as an afterthought. Robin Mager pointed to a Copilot rollout that paired broad access with hands-on workshops focused on real business challenges, using early learnings to shape a wider deployment. The result was strong uptake and sustained usage.
The panel returned to one thread a lot. Leadership models the behavior itself at every level of the organization, and treats visible experimentation, including the occasional miss, as part of how teams actually learn. The panel put it plainly: “Failures are how you learn.“
Tip: Before your next tool rollout, draft an adoption plan with the same rigor as the deployment plan. Include specific workshops, specific business problems, and specific owners. Three real use cases from your own workflows are a good sign you’re ready to roll out.
4. Treat MCP and agent orchestration as a governance problem
MCP is a protocol designed to ensure the inference or response from a language model is accurate for the system to which it is talking. Registering existing APIs as MCP tools requires little effort. Federated identity and authorization require real work, especially once an agent’s context starts propagating through 10, 15, or 20 connected systems. Nine out of ten AI systems get this wrong today.
Teams running complex multi-consumer architectures, like the one described above, are asking the same question from experience: the right model is zero trust at every hop, with governance explicitly defined at the agent layer. The MCP specification itself stays quiet on the topic.
Before you register your first APIs as MCP tools, map out who is accountable for authorization at each step, covering the agent, the MCP layer, and the underlying system. If you can’t answer what happens when an agent’s access needs to cascade through 15 systems, you’re not yet ready for production traffic.
Wherever you are on the journey, we will meet you there
The path from point-to-point integration to a fully agentic enterprise is a sequence, and every stage on it is a legitimate starting place. What matters going forward is whether the foundation underneath you can actually support what comes next.
Is your organization AI-ready? Let SnapLogic help you get there. Take a self-guided platform tour or book a personalized demo to talk with our experts.
