The $15.7 billion question: Can Salesforce tame the DataFam?
Why Tableau's improvisational data culture and Salesforce's rigid orthodoxy may be fundamentally incompatible
The man in charge of merging Tableau's "DataFam" with Salesforce had barely stepped on stage at April's conference before launching into his cheery greeting. "Hello! Good morning! How's it going! It's great to see everyone here…" — before he could have recognized a single face in the crowd.
Tableau President and CEO Ryan Aytay has got not a second to spare to coax the DataFam and its famously improvisational ways with data into the Salesforce fold, one of the few remaining homes of “a single version of the truth.”
The DataFam’s watching, and so is the boss, Salesforce founder and CEO Marc Benioff. Benioff wants Tableau to justify his $15.7 million 2019 purchase of Tableau. The DataFam wants Tableau to work the way it always has plus do whatever nice tricks Salesforce may add. And Aytay wants everyone to stay on board until his mission’s accomplished and he leaps into a bountiful new challenge.
DataFam roots
The DataFam — so named for the hordes who years ago swarmed through Tableau conferences — presumably still love Tableau’s free and easy way of data discovery. Questions
and answers flow one after the other. Stories take shape to account for the multiple factors afoot — some starting with a spark and failing fast, followed by more, failing or sparking even more inquiry.
Trouble is, that easy flow of question and answer doesn’t quite mesh with Salesforce — which is one of the few enterprise data vendors that still subscribe to the old “single version of the truth” dogma. You might have assumed that these two methods of data divination — the free-flowing play and the down-to-business seriousness — have found a balance. Each has been around long enough to have worked things out. In fact, the Salesforce-Tableau compatibility became a big question back when the acquisition was announced. I never heard a satisfying answer aside from the obvious payout for the Tableau founders and other senior people.
What does Aytay think? The one time I got to ask any question that came close was last year at a “fireside chat” with other industry analysts attending the annual conference.
The single most important piece of evidence I know of on that is what I heard him say last year to me in a group of industry analysts. I asked him: “For a company that's built its brand around data storytelling, shouldn't Tableau invest more in helping people actually tell stories?"
His reply was troubling. He said that the problem with storytelling is that stories are told differently every time. Well, that's what stories are supposed to do. It’s also how data discovery works. That chaos is how we sort through complex causation.
The competing bananas stories
Imagine a grocery store, CloudFlow Market. September banana sales dropped 23% compared with August. The data was clear, but the explanations multiplied like fruit flies.
The produce manager's story: "It's the weather. August was hot, people wanted smoothies. September cooled off, banana demand dropped naturally."
The marketing director's story: "We moved the bananas from the front entrance display to aisle 7. Impulse purchases disappeared. Location is everything in retail. Even our “Banana Daiquiri Fridays” failed even after I extended it over the weekend and the alcohol-control board inspector took my bribe.”
The operations analyst's story: "Our supplier switched from Ecuador to Alaskan hothouse bananas in early September. Customers noticed the taste difference and bought less."
The store manager's story: "It's the competition. The new Whole Foods opened three blocks away in late August. They're stealing our health-conscious banana buyers."
The sustainability coordinator's story: "It's the Cavendish panic. That Bloomberg article about banana fungus wiping out plantations has customers going cold turkey. They're emotionally divesting from bananas. I've seen customers avoiding the banana section and others taking photos of the banana display ‘for posterity.’”
Each story had supporting data. Weather reports showed a 12-degree temperature drop. Security cameras confirmed reduced foot traffic past the banana display. Customer complaint logs mentioned "different tasting bananas." Loyalty card data revealed some regular customers hadn't returned.
But the stories couldn't all be true. Or could they?
Banana Court
At FreshCart Market — which has no “agentic AI” — the regional manager resolves multiple stories by bringing the managers together in a “court,” where they consider every factor against daily sales data.
To resolve the multiple banana stories, the team mapped every factor against daily sales data. The weather story crumbled first; banana sales stayed strong during cool August days. The location theory weakened when they found that other relocated produce hadn't dropped.
The breakthrough came when the analyst cross-referenced supplier data with customer complaints. The new Guatemalan bananas weren't just different — they were arriving overripe because of a longer shipping route. Customers were rejecting them at checkout or not buying any at all.
The final story: Supplier change caused quality issues, which location change amplified by reducing impulse purchases that might have overlooked the quality problems. Competition made things worse by giving disappointed customers an alternative.
By October, FreshCart had negotiated faster shipping with the Guatemalan supplier and moved bananas back to the entrance. Sales recovered to August levels.
Enter Agentic AI
Over in the next town, CloudFlow Hub has also seen a slump in banana sales. But instead of a court, the store manager feeds data to its vendor’s version of agentic-AI.
It correlates weather data with daily sales; analyzes foot traffic around different stores; cross-references supplier changes with quality metrics and customer comments; compares competitor opening dates with customer defection; and compares statistical models to weigh each factor.
Imagine it: “Supplier change is 0.4 correlation; location change is 0.3 correlation; interaction effect is 0.6 correlation…” And so on.
But then, as if to annoy Aytay and his team, competent agentic AI would generate the same messy collection of competing hypotheses - it would just test them all simultaneously and assign probability weights.
Good luck to Ryan Aytay and Salesforce. May Aytay’s own story, the Data Fam’s, and Tableau’s stories find one happy version of the truth — after they sort through the multiple versions.