Map the process first, then fund the AI
Why process intelligence decides whether your AI budget pays back
Most enterprises still fund AI initiatives the way they funded automation a decade ago: pick a process that looks slow, apply the new technology, and hope the numbers improve afterward. It is an understandable habit. It is also, based on the evidence accumulating through 2026, one of the biggest reasons AI investments fail to return their cost. This guide explains why the order matters, why the whole market is converging on process-first, and what a process-first AI business case looks like.
Key takeaways
- Roughly 95 percent of generative AI pilots delivered no measurable profit-and-loss return, according to a 2025 MIT study of more than 300 deployments.
- The failures are less about weak models and more about pointing AI at processes that were never mapped or measured first.
- Three competing vendors, Celonis, Salesforce, and SAP, have independently concluded that AI needs process intelligence as a prerequisite, not a companion.
- Mapping a process first reveals what you are actually about to automate and usually surfaces the quick wins that fund the rest of the programme.
Why do 95 percent of AI pilots fail to deliver ROI?
A 2025 MIT study that reviewed more than 300 public AI deployments found that roughly 95 percent of generative AI pilots delivered no measurable return on profit and loss, with only a small number producing real, bookable value. The report’s explanation was not that the models were weak. Most of the budget went to visible, high-excitement use cases like sales and marketing copilots, while the clearer returns sat in unglamorous back-office processes. Our own reading, working alongside SAP and Salesforce transformation teams across ANZ, is that back-office processes tend to be measured and documented in far more detail than customer-facing ones, which is exactly why they are a safer place to point automation first.
Why is the whole market converging on process-first?
| Vendor | Move | What it signals |
|---|---|---|
| Celonis | Named a Leader in Gartner’s inaugural 2026 Magic Quadrant for Process Intelligence, and took its forums to Sydney and Melbourne in July 2026, with Zespri presenting savings. | Positions a real-time “context layer” as the prerequisite for AI agents. |
| Salesforce | Signed a definitive agreement to acquire process intelligence vendor Apromore. | Gives Agentforce a way to map real workflows before automating them. |
| SAP Signavio | Named a Leader in the same 2026 Gartner Magic Quadrant, its fourth consecutive year. | Treats process mining as a mandatory input to agentic AI, not a companion. |
A widely read July 2026 landscape analysis described this as a shift from static, after-the-fact process reporting to real-time, agentic process intelligence, arguing that discovering and redesigning a process before automating it is the direct fix for the industry’s high pilot failure rate.
Why does the order matter more than the technology?
None of this is really about any one vendor’s roadmap. It is a simple operating fact: an AI agent, a Salesforce Agentforce flow, or an SAP automation rule can only be as good as the model of the process it is given. If that model is a two-year-old SOP document or a manager’s best guess at how a queue actually clears, the AI will faithfully automate the exceptions, the workarounds, and the bottlenecks along with the intended process. Mapping the process first is not a governance nicety. It is how you find out what you are actually about to automate, and it is usually the fastest way to find the quick wins that fund the rest of the programme.
What does a process-first AI business case look like?
For ANZ organisations running SAP, Salesforce, or a mix of both, treating process mining as the first line item changes the shape of the business case. Instead of funding a pilot and hoping for ROI, you fund a short, scoped process discovery on one high-volume, high-friction process, using system event log data you already generate. Within weeks that tells you whether the process is fit to automate, needs redesign first, or is not worth automating at all. The automation budget then follows the evidence: it is spent where the data shows a clear, measurable return, with a human review gate on any step that carries compliance or financial risk. This is how an AI programme avoids becoming another entry in the 95 percent.
The GoWide view
GoWide runs a focused process discovery diagnostic on a single core process, typically over two to three weeks, using data already available in SAP or Salesforce. The diagnostic helps identify where the strongest opportunities for automation and AI exist before you commit further budget. If you have a high-friction process that needs closer examination, get in touch with GoWide to discuss a suitable diagnostic.
How Gowide helps
Scope a fixed-scope process discovery diagnostic
Gowide runs a focused process discovery diagnostic on a single core process, typically two to three weeks, using data you already have in SAP or Salesforce, to show exactly where the automation and AI opportunity sits before you commit the rest of the budget. Scope a diagnostic on your highest-friction process this quarter.
Frequently asked questions
Process intelligence is the discipline of using data from your systems to understand how a business process actually runs, including process mining from event logs and task mining from desktop activity. It gives a real, measured model of a process rather than an assumed one, which is the input AI and automation depend on.
Because an AI agent or automation can only be as good as the model of the process it is given. Deploying against an unmapped process automates the exceptions and bottlenecks along with the intended flow, which is a major reason roughly 95 percent of generative AI pilots delivered no measurable return.
A 2025 MIT study of more than 300 public AI deployments found roughly 95 percent of generative AI pilots delivered no measurable profit-and-loss return. The explanation was not weak models but budget aimed at visible use cases rather than the well-measured back-office processes where returns are clearer.
Yes. SAP Signavio was named a Gartner Leader and frames process mining as essential for enterprise AI, while Salesforce acquired process intelligence vendor Apromore so Agentforce can map real workflows before automating them. Celonis makes the same argument, so three competitors have converged on process-first.
Typically two to three weeks for a single core process, using event log data you already generate in SAP or Salesforce. The output is a verdict of go, redesign, or no-go, plus a scored view of where automation and AI will deliver measurable return.
