“AI bolted onto fragmented systems doesn’t accelerate progress. It amplifies inefficiency and risk.”
He’s right about the obstacle. And he was right to move on. Dwelling on this topic is not SAP’s interest. But it might be worth a moment of yours.
SAP has always been a tool that waits. You pick it up, use it, put it down. A human is always in the loop. The system forgives a great deal because humans compensate for what it can’t see.
An AI agent isn’t like that. It doesn’t wait. It doesn’t compensate.
For the first time in the history of enterprise software, something inside your business is going to act without being asked. Not enable. Not assist. Act.
Are you ready?
An Appetite For AI
There’s a keenness and appetite to use AI. Every CEO and CFO now has to speak to where it fits in their business. But appetite doesn’t mean the ground is ready.
Pivot spent months working on a machine learning project for one of the world’s largest tea businesses. The company wanted to predict promotional sales, multi-million pound decisions, made months in advance. The technology existed. The appetite was genuine. But when the team sat down to build it, they hit the same wall that stops large enterprises before they’ve started.
The data wasn’t ready.
Not dramatically broken. Not wrong. Incomplete in the ways that accumulate over years of people working around gaps rather than fixing them. Fields left blank because everyone knew to look somewhere else. Historical records that didn’t go back far enough. Promotional data sitting in systems that didn’t talk to each other.
The team had to go back and fix what was underneath before anything else could happen.
Firms don’t realise how far behind their systems are. Multi-billion companies’ systems around SAP are 10 to 15 years old. Even when the SAP implementation itself is brand new.
That’s not a failure. It’s normal.
The Data Problem
Pivot worked with a global consumer brands giant. It was transforming financial planning across almost 70 countries. When the team arrived, planning was being done in Excel, in 30 different ways. Each market had its own templates, its own version of the truth.
“The chance to deploy AI across their data was the stuff of dreams,” said our consultant.
It stayed a dream until the data was fixed. It took three years of embedded work. One source of truth. The project eventually processed 1.5 billion data records. The lesson is not the scale. It is the sequence. The data did not have to be perfect. It had to be consistent enough to trust.
The Process Problem Is Different.
A fast-growing drinks company Pivot worked with had built its competitive advantage on agility. It had invested in SAP. Then growth outpaced the system. Demand planning broke down. People built their own spreadsheets. Some bypassed the forecasting process. During peak periods operations ground to a halt. Those who shouted loudest got served first.
The process hadn’t been designed. It had accumulated. And it couldn’t hold the weight of what the business had become.
That is what an AI agent will inherit if you deploy before you understand what’s there. Not the process you intended. The process that exists.
What You Can Do
One answer is to walk your processes before the agents arrive. This is what that looks like.
A priority customer emails a sales rep. They need an urgent delivery. Next week. Can it be done?
Before SAP’s AI assistant Joule, that question triggered a chain of phone calls. The sales rep checked stock in S/4HANA. It wasn’t enough. They called the production planner. The planner opened a different app, checked capacity, found the schedule was full. They dug further and spotted a missing component from a supplier. They called procurement. Procurement called the supplier, got a new delivery date, updated the purchase order, called the planner back. The planner reworked the schedule. Only then could the sales rep call the customer back.
Every step needed a different SAP screen. Every handoff needed a phone call. Every answer was out of date by the time the next person read it.
That process worked. Not elegantly. Not quickly. But it worked. Because the humans stitching it together knew what to do, who to call, and how to fill the gaps between the systems.
“The chance to deploy AI across their data was the stuff of dreams”
Joule is designed to run that process end-to-end. And it can, if the process is ready for it. But look at how it works and the gaps appear immediately.
The handoff from sales to production planning is currently a phone call. Can Joule notify the planner directly? The handoff from production to procurement, currently a phone call. Can Joule route the query? When procurement updates the purchase order, does Joule propagate that through to production planning, or does someone pick up the phone?
Those aren’t technical questions. They are process questions. And until you answer them, the agent is running on a process designed for humans, not for something that acts “autonomously.”
An agent that hits an unanswered handoff doesn’t pick up the phone. It stops. Or worse, it carries on regardless.
Governance, Governance and Governance
Klein is right about the destination. The companies that lead will be those that connect AI to the way their business runs, with context, governance, and trust.
That context is the sum of every process your organisation runs, every data field it populates, every decision it makes about who is authorised to act and when.
None of that needs to be perfect. It needs to be understood.
Pivot has spent more than 19 years inside SAP estates across 70 countries. In that time, we have learned what accumulates, what breaks, and what was never properly designed in the first place. Organisations which navigate change well are rarely the ones that moved fastest. They are the ones that looked honestly at where they were starting from.
That knowledge matters more now than it ever has.
Not because the fundamentals have changed. Because for the first time, something is coming that will act on what it finds.
Pivot takes no fees, commissions or referral payments from any software or AI vendor.
Image Courtesy of SAP – Copyright: Bert Bostelmann / bildfolio
