Key Takeaways:
- AI amplifies whatever data quality you already have, good or bad. Fixing it upstream is critical before any AI initiative can succeed.
- Automation is how you enforce data quality at scale, especially as SAP modernization adds new complexity to your data landscape.
- The biggest risk isn’t moving too slowly on AI — it’s automating flawed processes and compounding the problems you already have.
When people think about AI, they tend to think about content generation — creating images, writing documents, summarizing meetings. But as TDWI Research Fellow Donald Farmer put it during a recent webinar we did together, “automation is a long-term component of AI, and much of the business success that we get from AI comes from the ability to automate processes which today are complex, manually driven, and relatively inefficient.”
That framing stuck with me. Because the relationship between data automation and AI gets to the heart of a challenge I see constantly in conversations with SAP customers: AI readiness isn’t primarily a technology problem. It’s a data quality problem, and automation is one of the most powerful tools we have for solving it.
Why Has Data Quality Been a Top Priority (and Top Obstacle) for Decades?
Data quality has topped enterprise technology challenges lists for years. So why haven’t we cracked it?
For a long time, companies have been able to work around their data quality problems. A bad number shows up in a report, you fix it manually. You inherit multiple ECC systems from acquisitions, each with different data structures, and you manage it with spreadsheets and tribal knowledge. The dysfunction is real, but it’s been manageable.
The problem is that AI doesn’t adapt the way humans do. It doesn’t know your workarounds. It doesn’t have 20 years of institutional memory to contextualize a quirky field value. As Donald put it, when you layer AI on top of inconsistent data environments, “you’re automating the dysfunction.” And that’s not a good thing.
What Does “Data Quality at the Point of Capture” Mean for SAP Environments?
One of the most valuable ideas from our conversation was what Donald called a “shift left” approach: moving data quality upstream, to the point where data first enters your systems, rather than trying to fix it downstream.
This matters enormously for AI agents in particular. An AI agent is designed to act autonomously and at speed. If your data quality process is slow — if it relies on catching errors in a report after the fact — you’re either holding the agent back while you clean data, or you’re letting it act on data that’s wrong. Neither is acceptable.
As AI becomes more embedded in operations, the concept of Agentic-Ready Data becomes critical. You need clean, governed SAP data that’s built for confident decision-making and autonomous processes, not just reporting.
The solution is to validate data before it ever gets posted to your system. In our Precisely Automate platform, that means checking against SAP rules and business logic at the point of entry, when problems are easier and less costly to fix, rather than weeks later when the damage is done.
What makes this harder than it sounds is the complexity of SAP itself. A material master record has around 300 data elements on average, and creating one typically involves six to eight people across the organization.
Automating that process — with all its validation logic, approvals, and cross-functional dependencies — isn’t trivial. But it’s exactly where the payoff is highest, because the manual alternative introduces enormous room for error at every step.
WEBINARWhen Automation Meets AI Readiness: Building a Trusted Data Foundation
Join TDWI Research Fellow Donald Farmer and experts from Precisely as we examine how automation of data and processes, especially in SAP environments, directly addresses the readiness challenges that most often derail AI at scale.
Is “Human in the Loop” Enough to Solve AI Data Problems?
“Human in the loop” has become shorthand for solving every AI concern. But it doesn’t always hold up under scrutiny.
The whole promise of AI is speed and scale that exceeds what humans can do alone. If you put a human in the loop for every AI decision, you’ve either created a bottleneck that defeats the purpose, or a workload no human can handle.
That said, humans absolutely need to stay involved — especially in regulated industries or any process where auditability matters. You can’t name your AI agent Jerry and say Jerry signed off on the invoice. That won’t fly with a regulator, and it won’t fly in court.
The real question is where humans add the most value. Take invoicing: an agent could go across your enterprise, find everything that needs to be billed to a specific customer, and compile the invoice. A human reviews and approves it. That’s not a bottleneck, but a good process design. The bottleneck is the manual collection work the agent just eliminated.
So “human in the loop” isn’t wrong. It’s just incomplete. The goal is to redesign processes so humans focus on judgment and accountability, while automation handles the repetitive data work that doesn’t require it.
How Is SAP’s Evolution Changing the Automation Equation?
The SAP landscape is shifting in ways that directly affect both automation and data quality, and not all of them are making things simpler.
For a long time, SAP was a walled garden. The only way to automate processes inside it was through custom ABAP development: expensive, brittle, and hard to maintain. What we’ve built at Precisely is a no-code/low-code alternative that works within that environment without requiring deep SAP development expertise.
The move to cloud and to interfaces like Fiori and GUI for HTML brings more open APIs and standardized protocols that ease integration. SAP’s Business Data Cloud is, I think, an acknowledgment that companies need to connect SAP data to everything else they run on — Salesforce, HR systems, analytics platforms — and that the full data landscape matters for AI.
But here’s what surprises people going through ECC to S/4HANA migration: process complexity doesn’t disappear just because the technology modernizes. A material record that had 300 fields in ECC? Still 300 fields (or more) in S/4HANA. The organizations that manage this well treat automation and data quality as part of the migration itself, not something to revisit afterward.
Where Is the Automation and AI Relationship Heading?
The conversation around automation and AI has never been louder, but the direction it’s heading is more practical than most headlines suggest.
I’ve been in technology long enough to have heard predictions about which innovation would eliminate entire job categories. The internet didn’t do it. Big data didn’t do it. I don’t think AI will either, at least not in the way that many forecasts suggest.
What AI will change is the nature of the manual work that remains after automation. Even well-automated processes still require humans to look things up and apply institutional knowledge. AI has the potential to absorb a lot of that grunt work — not to replace the judgment, but to surface the right information so the judgment is faster and better-informed.
A concrete example: in our Automate Evolve platform, we’re working on AI-assisted auto-complete for master data creation. You enter a small amount of information, the system reviews your historical records, pre-fills the remaining fields, and gives you a confidence score. That’s not AI replacing the process, it’s instead accelerating it and reducing manual error.
The thing AI does exceptionally well is encode institutional knowledge at scale. I think about a conversation I had with a large building materials company. When I asked how they handled a specific integration between SAP and another system, they said, “Oh, we use Susan.” Susan was a 20-year veteran who was the only person who understood how those systems connected. She was the integration layer, manually moving data between spreadsheets and writing the macros to generate reports.
Susan is remarkable. But Susan is eventually going to retire, and 20 years of process knowledge walks out the door with her. That’s a problem automation and AI together can genuinely solve. Not by replacing Susan, but by encoding what she knows into a process that can be monitored, audited, and maintained after she’s gone.
What Should You Do Next?
Donald closed his presentation with five practical steps that hold up as a real roadmap:
- Audit where your engineering time goes. Find out who’s doing data work, how long it’s taking, and where the biggest inefficiencies are.
- Identify a high-value data domain. Look for a use case where improving data quality has a meaningful downstream impact: customer onboarding, product master data, or financial close are common starting points.
- Review your governance programs. Make sure your existing processes account for AI requirements, not just operational ones.
- Map your agent intentions. If AI agents are on your roadmap, define what you want them to do and what success looks like before you start building.
- Focus on one automation investment. Pick the area where automation will free up the most capacity or deliver the highest downstream value, and start there.
One thing I’d add: don’t let the promise of AI paralyze you.
No-code/low-code platforms have evolved enormously over the past decade, and a lot of what people are hoping AI will eventually do, mature automation can do today. The organizations getting real results aren’t always waiting for the perfect AI-powered solution — they’re the ones who automated their SAP data processes, cleaned up data at the point of capture, and built a foundation that can support whatever comes next.
Want to go deeper on building a trusted data foundation for AI? Watch the full on-demand webinar, When Automation Meets AI Readiness: Building a Trusted Data Foundation.
