The path to AI is paved with data
The path to AI is paved with data: lessons from Dreamin’ in Data Chicago
By Barry Sheehan, Showoff
Last week I flew into Chicago for Dreamin’ in Data — a two-day community event at the Gleacher Center pulling together architects, admins, consultants and data leaders from across the Salesforce ecosystem. Showoff was proud to sponsor, and I was on our stand for most of the two days, which meant a lot of brilliant conversations with people stopping by and only a snatched window here and there for the keynotes and a couple of sessions I managed to drop into.
Even from that vantage point, one thing came through loud and clear: the conversations that matter right now are not about AI. They are about the data underneath it. This blog is a quick reflection on what made the event different, the two analogies I keep coming back to since I got home, and what they mean for any organisation staring down an AI strategy.
A refreshing change from the vendor circus
Most enterprise tech conferences follow a familiar script. A keynote. A sizzle reel. Three new product names that did not exist last quarter. A demo where everything works first time. You leave entertained, mildly overwhelmed, and not entirely sure what you are supposed to do on Monday morning.
Dreamin’ in Data was not that. It was peers talking to peers. Architects sharing the mess they actually live in. Consultants admitting what they got wrong on the last migration. Speakers using real screenshots from real orgs, sticky notes and all. There was no headline product launch, no countdown to a new SKU, no breathless pitch about how everything changes today. It was, refreshingly, just people being honest about the work.
One of the sessions I caught a chunk of was Allison Park from Salesforce on Enterprise Architecture and Strategy. Her opening analogy was the bit I have not stopped thinking about: enterprise architecture is urban planning. Without it, you get sprawl — every team building what they need, when they need it, with no shared services, no standards, no scalable foundation. With it, you get streets, schools, utilities and a city that can actually grow. The same, she argued, is true of data.
I also managed to slip into the back of Amy Oplinger Singh’s session, The 7 Deadly Sins of Data Migration. I only caught part of it but one line travelled with me back to the stand: data migration is 80% conversation and 20% click-and-drag. The failures, she argued, are not technical. They are human. Lust for too much data. Sloth on the mapping document. Pride that skips the sandbox. The room was nodding along — the kind of nodding that tells you everyone has been there.
The path to AI is paved with data
On the flight home I was comparing notes with my Showoff colleague Gordon Jackson, who had been with me on the stand for chunks of the event and had picked up his own threads from the sessions he managed to catch. He said something that summed up the whole event in a single line:
“The path to AI is paved with data.”
— Gordon Jackson, Showoff
It is the kind of sentence that sounds obvious until you hold it up next to what most organisations are actually doing. Boards want agentic AI. Vendors are happy to sell agentic AI. But the substrate underneath — the customer records, the product data, the case histories, the integrations between the half-dozen systems nobody has fully mapped — is, in most cases, not ready.
Allison’s framing kept landing on the same point. From the slides I saw, the message was that until you can inventory your data sources, understand what connects to what, and see where the gaps are, putting an AI layer on top is putting a Ferrari engine in a car with no chassis. Salesforce’s own Data 360 framing made the same case: agents need trusted context, and context comes from data that is governed, connected and clean.
Amy’s seven sins were the operational version of the same idea. Forward design, was the phrase that stuck: clean data first, automation second, AI third. Skip any step and the whole stack wobbles.
The window cleaners across from the Gleacher Center
On the second morning I was waiting for coffee outside the venue and looked up. Across the street, three figures were suspended from the roof of a glass tower, cleaning the windows one floor at a time. I took a photo — partly because the scale was striking, partly because something about it nagged at me for the rest of the day.
The view from outside the Gleacher Center: three window cleaners working a Chicago glass tower from the top down.
Here is what was bothering me. The whole operation hung on one thing: the anchor at the top of the building. Not the harnesses, not the buckets, not the squeegees, not the skill of the people on the ropes — all of those matter, but they are downstream. If the anchor fails, none of the rest of it matters. The anchor is the trust the whole job is built on.
That is the bit a lot of organisations are getting backwards with AI. We assume good foundations get built from the ground up — lay the data, then the integration layer, then the apps, then the agents on top. Sometimes, though, the anchor has to be set first, from above. Leadership has to decide that data quality, data ownership and data governance are non-negotiable before the project starts, not a thing to be retrofitted once the agent demo lands flat in front of the board.
Top-down trust is unfashionable to talk about. It sounds slow. It sounds like governance committees and steering groups and quarterly reviews. But the window cleaners do not start by improvising a harness halfway down. They start by checking the anchor. So should we.
What this means if you are starting an AI project
Pulling the threads together, three things from Chicago are worth taking back to your own organisation:
- Map before you build. Allison’s point about urban planning applies whether you are running a five-system org or a fifty-system one. Do the architecture review. Inventory what you have. Find the gaps. A conceptual map your business leaders can actually read will get you further than a perfect logical diagram nobody opens.
- Treat data work as the project, not the prep. Amy’s Golden Rule — 80% conversation, 20% click-and-drag — applies here too. The migration, the mapping, the data quality work is not the warm-up before the AI project. It is the AI project. The agent is the last 20%.
- Set the anchor at the top. Decide, at leadership level, that the data foundation is the priority. Not a workstream. The priority. Then let everything else hang from it. This is the bit that lets your data teams push back when someone wants to import fifteen years of dormant leads by Friday.
None of this is glamorous. None of it makes a good launch video. But it is the difference between an AI strategy that holds, and one that swings in the wind.
A final thought
What I liked most about Dreamin’ in Data was that nobody on stage was selling certainty. They were sharing what they had learned by getting it wrong, and what they would do differently next time. That kind of honesty is rare in this industry and worth showing up for.
If your organisation is thinking seriously about AI — Agentforce, Claude, internal copilots, whatever flavour — the most useful thing you can do this quarter is probably not to pick a model. It is to look hard at your data, your architecture and your governance, and make sure the anchor at the top is one you would trust your business to hang from.
At Showoff, that is the work we love. If you want a hand mapping where you are, where you want to get to, and what to fix first, we should talk.
Ready to set the anchor?
If you are building toward AI on Salesforce and want a sensible read on your data foundations before you commit to a roadmap, get in touch with the Showoff team for a no-pressure data and architecture review.