Every enterprise software vendor — from SAP to Salesforce to Microsoft to Oracle — is now embedding AI into its ERP and CRM products. The pitch is consistent: autonomous agents, natural-language interfaces, “digital labour” that processes invoices, resolves customer tickets, and even changes business decisions without human intervention. If you are a Founder or CEO of a mid-market company, you’ve almost certainly received some version of this pitch in the last twelve months. The question isn’t whether AI in ERP and CRM is real. It is. The question is whether it’s ready for what your business actually needs — and whether the vendor narrative is telling you the whole story.
The Vendor Race Is Over. Everyone Has Shipped AI.
At the product level, the embed race is essentially complete. SAP launched its generative-AI copilot Joule in September 2023 and now reports 34,000 customers using SAP Business AI across 400+ scenarios (SAP, “Business AI Customers in Action,” June 2025). Salesforce evolved from Einstein to Agentforce, which went generally available in October 2024 and had closed over 18,500 deals (9,500+ paid) by Q3 FY26 (Salesforce Q3 FY26 Earnings, Dec 2025). Microsoft embedded Copilot across Dynamics 365 and announced ten autonomous agents in October 2024 (Constellation Research, Oct 2024). Oracle unveiled 50+ AI agents at CloudWorld in September 2024, then scaled to 600+ by October 2025 with an AI Agent Marketplace (Oracle press release, Oct 2025). ServiceNow, Workday, HubSpot, NetSuite, Infor, Epicor and Sage have all followed.
The differentiation is no longer whether AI exists in the product. It’s about pricing model, data platform depth, and governance — which is where the real decisions sit for a CEO.
The Results That Vendors Publish (and How to Read Them)
There are genuine results. SAP’s customer page reports Western Sugar Cooperative reduced invoice processing time by 25% and cut cost per invoice from $8 to $6. AMD reportedly cut supply-chain issue resolution time by 90%, saving 3,120 staff hours annually. Coca-Cola Europacific Partners improved forecast accuracy by 6% (all SAP self-reported, “Business AI Customers in Action”). On the CRM side, Salesforce reports that 1-800Accountant’s Agentforce deployment autonomously resolved 70% of chat engagements during peak tax season, handling 1,000+ conversations in the first 24 hours (Salesforce, Agentforce 3 press release, June 2025). OpenTable reports 70% of its inquiries handled without humans. Salesforce itself — as what Marc Benioff calls “customer zero” — cut its support workforce from 9,000 to roughly 5,000 (Benioff on The Logan Bartlett Show, per CNBC, Sept 2025). Microsoft’s own study of Copilot in Dynamics 365 Customer Service found a 12% reduction in case-handling time across 6,500 users (Microsoft, April–July 2023 study).
These numbers are real — but they carry an important caveat. Every one of these is vendor self-reported or comes from a vendor-commissioned study (including the Forrester TEI of Agentforce, commissioned by Salesforce, which projected 195% ROI). We should read them as what’s achievable in the best-case, not what’s typical.
Curious where AI in ERP and CRM could actually move the needle in your business — and where it would be a distraction? That’s the conversation a Business Diagnostic starts with.
What the Independent Research Actually Says
This is where the picture changes materially, and where I think CEOs and Business Leaders need to pay close attention. The vendor keynotes and the independent evidence are telling two quite different stories about AI in ERP and CRM right now.
McKinsey’s “State of AI in 2025” (November 2025, 1,993 respondents across 105 countries) found that while 78–88% of organisations are using AI in some form, only 39% report any enterprise-level EBIT impact — and for most, that impact is under 5%. Only about 6% qualify as “AI high performers” with meaningful profit impact. Those high performers share a defining trait: they are 3.6 times more likely to have fundamentally redesigned their workflows, not simply added a copilot on top of an unchanged process.
S&P Global’s “Voice of the Enterprise: AI & ML Use Cases 2025” survey (1,006 respondents, Oct–Nov 2024) found that the share of companies abandoning most AI initiatives jumped from 17% to 42%, with the average organisation scrapping 46% of proof-of-concept projects before they reached production. MIT’s Project NANDA study (July 2025) reported that 95% of enterprise GenAI pilots deliver no measurable P&L impact — though it should be noted this was based on 150 interviews and 300 public deployments, and the study itself uses a “directionally accurate” caveat.
Gartner, meanwhile, predicts that over 40% of agentic AI projects will be cancelled by end of 2027 due to escalating costs, unclear business value, or inadequate risk controls (Gartner press release, June 2025). They also warned explicitly about “agent washing” — the practice of rebranding chatbots, assistants, and RPA scripts as autonomous agents. Gartner estimates only around 130 of the thousands of vendors claiming to offer “agentic AI” are genuine.
The Three Things Vendors Aren’t Diagnosing for You
We work with mid-market businesses across manufacturing and services, and I see a consistent pattern in how AI in ERP and CRM conversations go wrong. The vendor sells the capability. The implementation partner sells the timeline. But nobody diagnoses the three things that actually determine whether any of this works.
First, Data Foundation. The uncomfortable truth beneath every failure statistic is that the AI model is now effectively a commodity — the differentiator is whether your master data, CRM records, and transactional data are clean, unified, and governed. Salesforce’s own Data Cloud (a prerequisite for Agentforce, estimated at roughly $108,000/year minimum) exists precisely because agents pulling from duplicate or stale records produce unreliable outputs. We have seen this pattern across industries: AI amplifies the environment it enters. If that environment is messy, AI makes the mess faster.
Second, the Transaction-versus-Decision divide. Nearly every proven case study today — invoice matching, case summarisation, tier-1 ticket deflection, draft generation — is transaction automation. That is genuine and bankable. The far larger promise (in my words: the keynote promise) — agents that reprioritise a supply chain, change a pricing decision, or reallocate capital — is where pilots stall. Decisions require trust, context, and accountability that current governance cannot yet underwrite. Buy the transaction automation now; treat decision automation as a roadmap, not a purchase.
Third, the pricing signal. When Salesforce prices Agentforce per conversation ($2/conversation, plus Flex Credits at $0.10/action) while SAP, Oracle, NetSuite and Workday deliberately bundle AI at no extra charge, they are making opposite commercial bets. Consumption pricing signals confidence that agents create measurable value — but it also transfers volume risk to the buyer and charges you for conversations the agent fails on (you pay per attempt, not per success). “Free” bundling is a defensive play to protect platform loyalty at a moment when a significant share of enterprise buyers are actively evaluating switches. For a CEO, the pricing model tells you whether the vendor is selling outcomes or defending turf.
What a CEO Should Actually Do Next
The path forward isn’t to reject AI in ERP and CRM — the capability is real and the transaction-level gains are proven. The path forward is to diagnose before you deploy. Audit your data quality before signing any AI add-on. Pick one narrow, high-volume, measurable transactional use case with a named owner and a baseline metric captured before you go live. Model the true total cost of ownership — including the data platform, implementation, and consumption overruns — not just the headline per-conversation price.
And critically, measure your own outcomes. If you can’t state today’s cost-per-transaction for the process you want to automate, you are not ready to evaluate whether AI improved it. The 95% who show no P&L impact and the 42% who abandon their projects share this trait: they deployed before they diagnosed.
The AI is no longer the bottleneck. Your Readiness is. And that’s a Business Diagnostic conversation, not a technology purchase.
Book your FREE Business Diagnostic today. Limited slots available: 100dayrenew.com