Six months ago, a data cleansing effort ahead of a group-wide ERP rollout, spanning SAP and SuccessFactors had a problem: lots of manual work, no shared direction, and no way to tell how close we actually were to done.
I stepped in trying to fix that. First, by building a single source of truth through dashboarding, so progress became visible instead of anecdotal. Then, by getting subject matter experts and business process owners around the table to define the business rules behind the data, so most of the cleansing could run automatically, leaving the team to focus on the real exceptions.
Along the way, a new management question came up that I had to work through myself: AI can fill gaps and validate data far faster than manual effort, which changes when a cleansing stream should "exit." But it also introduces new risks, like confidently wrong output, and a harder question underneath: "good enough for the ERP" and "good enough to train AI on top of it" are not the same bar.
But the part that surprised me most wasn't about AI at all. Once you force people to define the actual rule behind every field, not just the process on paper, you start seeing a second, mostly invisible layer of the organization: local exceptions everyone quietly knew about but never documented, fields treated as optional in practice despite being mandatory by design. Making that layer visible turned out to matter as much as the ERP itself.
And here's a provocation worth sitting with: even a "traditional" ERP implementation benefits from AI well beyond faster cleansing. Used deliberately, it also builds AI skills in an organization and trains a model on the actual business, turning AI into a standing partner alongside the ERP rather than a one-off tool.
I've written up the full story, including concrete examples of that hidden layer and how we approached the exit-criteria trade-offs, in the article below.
Curious how others managing similar programs are thinking about this.