I recently stumbled across a Tiktok video that reminded me that the steps required to create robust enterprise software aren’t always intuitively obvious.
The video is about how to build a retaining wall. Easy, right?!
Well, no — as the video explains, there’s a big difference between throwing something up and building a wall that will actually work for years to come: concrete, a crushed-stone footing, loose gravel backing for drainage, filter fabric against mud, a perforated drainage pipe, plastic mesh Geogrid to stop the ground from sliding, rebar to handle the tension, etc.
What does this have to do with enterprise AI? Well, let’s say you use the latest AI coding environments to vibe-code a new work solution — and it works even better than a vendor’s because it’s custom and adapted to the way you work. Congratulations!
But be careful. That version is probably the equivalent of the first wall in this video. It looks great, but may have fundamental flaws that will undermine its usefulness in the future.
In the enterprise, with AI, it’s now (astonishingly) easy to create working prototypes. But maintenance, governance, compliance, confidentiality, ethics, etc is still just as hard (maybe even harder: while AI errors are slowly getting rarer, they’re also getting more difficult to spot).
And one big recent change: hacking has becoming super-powered. Really secure software is difficult to build, and requires constant monitoring. There’s a reason security experts such the UK National Cyber Security Centre say things like “take the minimum amount of responsibility for yourself, preferring that the service provider manage as much as possible.”*
Building something is one kind of effort. Running it responsibly inside an enterprise is another. The larger, more regulated, and more interconnected the business becomes, the wider that gap gets.
What do you need to think about when building a solution?
1. Business context. Enterprise AI is only as useful as the business context it understands. Building a model that processes data is one challenge. Building one that understands the semantic relationships between a vendor, a purchase order, an approval hierarchy, a compliance requirement, and a cash position is a very different undertaking. That kind of context does not come automatically from a foundation model. It has to be built, maintained, and continuously aligned to how the business runs.
2. Maintenance and change. The work does not end when the capability goes live. In many ways, that is when the harder work begins. Models drift. Data structures change. Policies evolve. Regulations shift. Security issues emerge. Every AI capability a team builds must be updated, monitored, troubleshooted, and improved over time. Many teams underestimate that ongoing cost when they are focused on getting the first version working.
3. Governance and accountability. Enterprise AI raises operational questions that go far beyond the model itself. Who is responsible when the AI recommends a course of action that leads to a bad outcome? How is access managed across a global workforce? How does the capability meet data residency, audit, and regulatory requirements across different jurisdictions? Building functionality is only one part of the problem. Building the governance around it is often where real complexity lives.
4. Integration into execution. AI capability only creates value when it fits into live business workflows and leads to action. That means connecting recommendations to decisions, decisions to processes, and processes to systems of record. This is where many internal build efforts become harder than expected. The challenge is not only generating insight. It is embedding that insight into how the business operates.
5. Innovation. Outside providers have a focused view over a market and the new opportunities for innovation that internal teams may struggle to keep up with.
Note that none of this means the answer is always to buy. There are legitimate cases where building is the right choice. A workflow may be unique enough, the data may be proprietary enough, and the internal capability may be strong enough that a custom approach creates a real advantage.
SAP provides AI capability in the environment where the business runs. It is already connected to business data, process logic, controls, and enterprise operating realities. The best way to provide a tailored solution is build on platforms that can handle everything that you don’t know as well — for example, the SAP Business AI Platform, with AI Core providing easy access to over 70 different AI models, so you can adapt to the latest opportunities and costs, and SAP Joule Studio, an agentic development environment that brings together the best of business intent with information about your environment to create solid, secure extensions and agents.
See more at https://www.sap.com/products/artificial-intelligence/joule-studio.html
*https://www.ncsc.gov.uk/collection/cloud/understanding-cloud-services
Leave a Reply