I was at the Salesforce Data & AI Summit in Zürich last week. The energy in the room was high. Salesforce had a strong lineup — new capabilities in Tableau, Einstein updates, real talk about agentic AI. Worth attending. But what I found most valuable wasn't on the main stage.
It was the conversations I had between sessions, over coffee, while waiting for the next talk to start. With architects, project managers, and IT leaders from Swiss and European companies. People who are in the middle of Salesforce deployments — or who just came out of one.
And what they said tells a different story than the one on stage.
The Demo Gap Is Real
Salesforce does impressive demos. At this event, the Tableau showcase was genuinely exciting — more intuitive data exploration, better AI-assisted analysis, tighter integration with the rest of the Salesforce stack. Einstein is getting smarter, faster, and more embedded across the platform.
But I kept hearing some version of the same observation from attendees: "This looks great. Now explain to me how we actually get here."
The person who said this most directly works at a global industrial company. They've had Salesforce running for two years. Their Einstein features are mostly turned off. Their Tableau deployment is used by three people in finance. Not because the tools aren't capable — but because the foundation underneath them isn't ready.
Duplicate records in CRM. Sales data that doesn't match what's in the ERP. Pipeline stages that nobody updates. Reports that leadership stopped trusting six months after go-live.
The product on stage assumes clean data, consistent processes, and high adoption. The reality in most companies is none of those three.
The Implementation Problem Nobody Talks About on Stage
The pattern I heard repeatedly: a company buys Salesforce, assigns an internal project team, brings in a consultant, spends six to twelve months in implementation — and then goes live with something that works in demo mode but doesn't survive contact with the real organisation.
Why does this happen? A few reasons came up across different conversations.
First, requirements lists that grow without boundary. One architect told me their implementation had 340 documented requirements before they went live. Three hundred and forty. Nobody stepped back and asked which twenty of those actually mattered. Everything became mandatory because nobody had the authority to say no to any business unit.
Second, the wrong people making product decisions. Salesforce decisions in many companies are made by IT or by the implementation consultant — not by the people who will use the tool every day. The result is a system that is technically correct and operationally useless.
Third, training that ends at go-live. You bring people into a room, show them how to click through screens, and call it done. Three months later, half of them have reverted to spreadsheets and email because the system is slower than what they were doing before.
The core problem: Salesforce sells a platform. What companies actually need is a working system — and those are two very different things. The platform gives you the capability. Building the system requires discipline, leadership, and someone willing to say "no" to features that aren't ready yet.
What the Tableau Data Actually Showed
One of the more revealing conversations I had was with a data analytics leader at a global organisation whose name you would recognise. They've been running Tableau for over a decade.
What struck me wasn't what they said about Tableau's new features. It was how they talked about their data. No hedging. No "well, it depends on which report you're looking at." Just: "Here's what the number is. Here's what drives it."
That confidence didn't come from a new AI feature. It came from ten years of consistently maintaining one version of the truth. One governed data model. One process for what gets measured and how.
The companies at the summit who were most excited about Einstein and the new Tableau capabilities were — almost without exception — the ones who already had that foundation. The ones who were struggling were the ones trying to layer AI on top of data they don't trust yet.
It's not a technology problem. It's a sequencing problem.
Three Things Worth Bringing Back to Your Leadership Team
If you're running Salesforce today, or considering a deeper investment in Tableau or Einstein, here are the questions that kept coming up in the conversations I had at the summit.
1. Can you run your pipeline review from Salesforce data alone? Not with verbal updates from the team. Not with a side spreadsheet to fill in the gaps. Just from what's in Salesforce. If the answer is no — that's your starting point, not Einstein.
2. Do you have a named owner for data quality? Not IT. Not the Salesforce admin. A business owner who is accountable for whether the data in the system reflects reality. Without that role, every improvement degrades within months.
3. How many requirements did your last implementation start with? If the number is above fifty and your team is still in implementation, something is wrong. Complexity is not a sign of thoroughness. It's a sign that nobody has made the hard prioritisation decisions yet.
Why This Matters Now
The pressure to adopt AI is real. Boards are asking about it. Competitors are announcing it. Salesforce is making it easier to turn on with every release.
But AI doesn't fix bad data. It amplifies it. An AI agent that reads your pipeline and gives you a forecast is only as good as the data your reps entered — or didn't enter. An Einstein recommendation based on duplicate records will give you duplicate suggestions.
The companies at the summit that were furthest ahead weren't the ones who started with AI. They were the ones who spent years getting the boring stuff right — clean data, consistent processes, a system people actually use — and are now watching AI make those investments compound.
That's the story the main stage doesn't tell. But it's the one that matters most for your roadmap.