Supporting research materials for the previously published article The Future of ERP: Licensing, Build vs Buy, and Agentic-first Architecture

This is a summary of supporting research materials for the previously published article [The Future of ERP: Licensing, Build vs Buy, and Agentic-first Architecture], which explored AI agents and enterprise software — from pricing models and build-vs-buy decisions to the skills that matter most in implementations.

Executive Summary

The enterprise software landscape is undergoing its most significant transformation since the shift to cloud computing. AI agents are fundamentally reshaping how ERP systems are licensed, implemented, and consumed. This report validates three main points about the future of ERP against current evidence from leading analyst firms, major vendors, and industry practitioners.

The Structural Problem

The per-seat licensing model was built on an assumption that no longer holds: that human labour is the primary bottleneck in business operations. AI agents operate continuously, make autonomous decisions, and consume computational resources in patterns that fundamentally don't fit the "per-seat" paradigm. As one analysis notes, "How do you charge per seat for something that never sleeps and scales infinitely?".[^1][^2][^3]

The numbers are stark. The agentic AI market is projected to reach $45 billion by 2030 (up from $8.5 billion in 2026), and McKinsey research shows that 63% of software vendors believe AI will fundamentally change their business model within three to five years. Partner Economics predicts that agent-to-agent software transactions could reduce seat counts by as much as 70%.[^4][^3][^1]

How Major Vendors Are Responding

The major ERP and enterprise software vendors are actively experimenting with alternative pricing models, though no consensus has emerged:

Vendor	Pricing Approach	Details
Salesforce	Three parallel models	Per-seat (traditional), Flex Credits (20 credits = $0.10 per action), and AELA (flat-rate unlimited AI agent usage over 2-3 years) [^5][^6][^7]
Microsoft	Bundled per-seat with AI	Core M365 pricing going up in July 2026 with Copilot Chat and security features folded in; Copilot Studio uses consumption-based Security Compute Units (SCUs) [^7][^8][^9]
SAP	Embedded agentic via Joule	Joule Studio Agent Builder with low-code/no-code capabilities; multi-agent orchestration across finance, supply chain, and HR [^10]
Zendesk	Outcome-based	$1.50 per automated resolution ($2 pay-as-you-go); customers only pay when AI solves the problem [^7]

Salesforce's trajectory is particularly instructive. The company initially launched Agentforce with per-conversation pricing, then shifted to consumption-based Flex Credits, and most recently introduced AELA (Agentic Enterprise License Agreement) — a flat-rate, unlimited-usage model. CEO Marc Benioff acknowledged that "customers have pushed for more flexibility" and that seat-based pricing "was becoming the norm" because customers wanted pricing predictability.[^11]

The Nuance: Per-Seat Isn't Dead Yet

Gartner's Jan Cook observes that in the current phase, "users are not buying AI agents to replace people, and they want some certainty in the pricing model before investing". The likely near-term equilibrium is hybrid:[^11]

Copilots (human-augmenting AI) will continue with seat-based pricing since usage is tied to humans and is generally predictable.[^11]
Autonomous workflow agents will migrate to usage-, output-, or outcome-based pricing.[^11]
Enterprise-wide AI may gravitate toward flat-rate enterprise license agreements that decouple cost from per-unit consumption.[^6]

Deloitte predicts that by 2030, at least 40% of enterprise SaaS spend will shift toward usage-, agent-, or outcome-based pricing. The transition won't be overnight — the seat-based model is evolving rather than disappearing, and Deloitte expects "a lot of experimentation and pricing variety in 2026 and beyond," with standard practices potentially taking years to emerge.[^12]

Conclusion

Per-seat licensing is structurally misaligned with how AI agents consume software. However, the transition is more gradual and messy than a clean paradigm shift — vendors are running multiple pricing models simultaneously, and enterprise buyers still crave predictability. The industry is currently in an experimental phase where hybrid models (combining seat-based subscriptions with consumption overlays) represent the practical reality for 2026–2028.

Why the Buy Preference Persists

Several structural factors make building enterprise-grade ERP from scratch impractical for most large organisations:

Compliance and regulatory burden. For financial services, pharmaceutical, manufacturing, and publicly listed companies, the compliance burden of custom ERP is substantial. Commercial ERP systems provide vendor-managed regulatory updates, built-in audit trails and controls, certifications and compliance reports, and a community of users facing the same requirements. Building custom systems requires dedicated compliance engineering, internal audit teams reviewing code, documented controls and change management, and regular security audits — costs that often exceed the licensing savings.[^13]

Scale and ecosystem effects. The global ERP market reached approximately $131 billion in 2024, with Oracle ($8.7B), SAP ($8.6B), and Microsoft ($5.4B) dominating the large enterprise segment. These vendors provide pre-built connectors to thousands of systems, marketplace apps, ISV ecosystems, and implementation partner networks that no in-house team can replicate. Gartner notes that as core ERP functionalities become increasingly standardised, "the true differentiator for ERP vendors lies in the strength and depth of their partner ecosystems".[^14][^15][^16]

Talent scarcity and total cost of ownership. Building custom ERP succeeds only when organisations can hire and retain specialised engineering talent, secure multi-year executive commitment, and treat the ERP platform as a product with ongoing investment. A practical 5-year TCO analysis shows the cost gap between build and buy is often surprisingly narrow (as little as 15–19%), but the risk profile is fundamentally different — custom builds carry significantly more execution risk.[^17][^13]

The "build core, buy context" framework. Industry thinking has converged on a clear principle: "Build what sets you apart. Buy what scales". ERP systems for finance, HR, procurement, and standard operations represent "context" (table stakes that must work reliably), not "core" (processes that differentiate in customers' eyes). As one analysis puts it: "Be honest: Are you building ERP because it creates advantage, or because your CTO wants to build something?".[^18][^13]

The Emerging Hybrid Model

The real evolution isn't pure build vs. buy — it's a composable hybrid approach. Industry data suggests an 80/20 rule: 80% of enterprise AI and ERP needs are met by purchased or subscription-based solutions, while 20% are addressed with custom-built components where deep integration or unique IP is critical.[^19]

McKinsey's framework for AI-ERP integration advises organisations to "purchase standardised agentic capabilities — embedded approval agents, pre-built data products, ERP-integrated orchestration frameworks — and reserve custom development for domain-specific logic that creates competitive advantage".[^20][^21]

This is reinforced by the practical reality that AI agents can now sit on top of existing ERP systems, connecting what was never designed to connect, without requiring rip-and-replace migrations. The composable approach — commercial ERP for the core, with custom AI agents and integrations layered on top — is emerging as the dominant pattern for 2026 and beyond.[^22][^23][^13]

Risk of AI-Native Disruption

Deloitte acknowledges that some AI-native startups are developing agentic solutions that could potentially disrupt incumbents. In the short term, simpler processes like customer service are more likely to be disrupted, but disruption could spread to more complex markets like ERP and CRM. However, Deloitte predicts this won't happen in 2026 — it will likely take at least five years or more, as "traditional SaaS providers have large footprints across complex workflows that will likely be hard to supplant".[^12]

Conclusion

Large organisations will overwhelmingly continue buying ERP from major vendors for reasons of compliance, ecosystem maturity, risk management, and talent efficiency. However, the nature of "buying" is shifting from monolithic implementations toward composable architectures where vendor platforms serve as the foundation, augmented by custom-built AI agents and integrations at the edges. The 2026 reality is "buy the platform, build the intelligence layer."

The Current Skills Transformation

The three largest ERP vendors — SAP, Oracle, and Microsoft — have all released AI modules with autonomous decision-making capabilities. According to digital transformation expert Mark Karelov, AI assistants have already automated roughly 40% of routine ERP consulting tasks, including generating business requirement documents, mapping data fields, and configuring basic workflows in D365 F&O and SAP.[^24]

The impact on the consulting workforce is already visible:

By 2030, an estimated 25–30% of traditional ERP jobs may disappear, particularly those focused on repetitive configurations or documentation.[^24]
Junior analysts and technical writers are the most vulnerable roles.[^24]
Companies aren't necessarily hiring fewer consultants — they're shifting what they need. After Copilot's release, many firms cut back on junior implementers but started paying premium rates for experts who could validate AI-generated configurations and ensure compliance.[^24]
Nearly half of all Dynamics 365 deals now include AI governance clauses.[^24]

The New Skill Profile

The emerging ERP implementation professional requires a fundamentally different skill-set from the traditional functional consultant:

1. From Requirements Collector to Strategy Partner Business analysts are no longer just documenting requirements — they are becoming strategic partners who understand value chains, redesign future workflows, and guide business teams through agentic AI transformation. The role shifts from translating business needs into system configurations to defining how human-agent hybrid workflows should operate.[^25]

2. Specification as a Core Competency The rise of "vibe coding" and spec-driven development makes the ability to write detailed, unambiguous specifications the critical bottleneck. As one practitioner notes: "In automation projects, not having a strong BA or someone who acts like one almost guarantees low ROI". The quality of AI-generated output is directly proportional to the quality of the specification — vague requirements produce vague code. Tools like Kiro.dev demonstrate how conversational AI approaches, combined with proper requirements and design documentation, produce significantly more predictable and production-ready results.[^26][^27]

3. Prompt Engineering and Context Engineering Domain knowledge now shapes system prompts, memory rules, and instructions. Clear, testable prompts reduce rework and replace many traditional handoffs between business and technical teams. Beyond prompting, BAs will design "context models" — deciding what data agents access, how fresh it must be, and which sources feed intelligent decision-making.[^25]

4. AI Governance and Validation Agentic systems can hallucinate or behave unpredictably. The new consultant must help define failure modes, acceptance criteria, safety checks, and red-team scenarios — becoming a key partner in agentic quality assurance and governance. Every AI-generated configuration should go through sandbox testing, human review, and compliance checks before going live.[^25][^24]

5. Orchestration Thinking The shift from "prompt thinking" to "workflow thinking" is fundamental. Rather than crafting single interactions, the modern implementer must orchestrate specialised agents — understanding how to decompose complex processes into agent-appropriate tasks, define authority boundaries, and build systematic evaluation loops.[^28]

6. New KPIs and Outcome Metrics Success measurement shifts from activity metrics to autonomy KPIs: percentage of decisions resolved autonomously, accuracy/quality of agent reasoning, and business-value impact of AI decisions.[^25]

McKinsey's Cross-Functional Model

McKinsey recommends that the most effective approach to agentic ERP implementation is to "bring together domain experts, ERP functional experts, and AI practitioners in short, structured working sessions". These sessions walk along target AI workflows step by step, explicitly marking which ERP tables, fields, and processes must be accurate, available, and exposed to AI agents to run and scale.[^29]

BCG reinforces this, noting that agentic AI requires "a mix of advanced technical talent such as AI-prompt engineers, AI/ML specialists, and data engineers as well as business translators who can map AI use cases to workflows. Most organisations underestimate this need". The insurer example is telling: a company that initially staffed its AI innovation team with only existing data scientists quickly realised it needed domain experts who understood claims processing in depth.[^30]

The Convergence of Domain Knowledge and AI Fluency

McKinsey's vision of the "agentic organisation" makes clear that leaders across all functions — not just CIOs — will need technology fluency once expected only of chief information officers. Early evidence shows that "employees without technical backgrounds can learn to manage agentic workflows as quickly as trained engineers". The implication is profound: the competitive advantage shifts from knowing how to configure an ERP system to knowing the business well enough to specify what the AI agent should do — and what it should never do.[^31]

As Mark Karelov summarises: "Today's AI doesn't replace consultants — it replaces those who merely translate requirements into IT systems. Our task is mastering symbiotic collaboration".[^24]

Conclusion

The future ERP implementer is not simply a traditional consultant who can also write prompts — they represent a new professional archetype combining deep domain expertise, specification discipline, AI governance capability, and orchestration thinking. Organisations that invest in developing this hybrid skill-set will have a decisive advantage in the agentic era. The critical insight for D365 F&O specifically is that functional knowledge doesn't become less valuable — it becomes the essential input that determines whether AI agents produce reliable business outcomes or expensive errors.

Cross-Cutting Themes

The Speed of Transition

There is broad agreement that the agentic transformation of ERP will happen faster than the client-server to cloud migration, but slower than the hype cycle suggests. Deloitte predicts that full replacement of enterprise applications by agents won't happen in 2026 — "it will likely take at least five years or more". However, the experimental and augmentation phase is already well underway, with Gartner forecasting that 40% of enterprise applications will integrate task-specific AI agents by the end of 2026.[^32][^4][^12]

The Governance Imperative

Across all three dimensions — licensing, build/buy, and skill-sets — governance emerges as the critical enabler and potential bottleneck. BCG's comprehensive framework covers design-phase governance (ownership, access controls, risk tiering), build-phase safeguards (kill switches, tool hardening, validation testing), and operational controls (human oversight, explainability, change management). McKinsey warns that "the scale of agentic adoption will be capped by how much oversight capacity humans can provide — making governance itself a potential bottleneck to productivity".[^30][^31]

Implications for the D365 F&O Ecosystem

For the Microsoft Dynamics 365 Finance and Operations ecosystem specifically:

Licensing: Microsoft is currently pursuing a bundled per-seat strategy (folding Copilot into higher-priced M365 SKUs) while maintaining consumption-based pricing for Copilot Studio agents. This dual approach hedges against both scenarios.[^7][^8][^9]
Build vs Buy: The Power Platform and Copilot Studio provide the "build the intelligence layer" capability on top of the bought D365 F&O platform, aligning with the composable hybrid model.
Skills: D365 F&O functional consultants who can combine deep module knowledge with specification discipline for AI agents — covering processes like intercompany eliminations, pharmaceutical batch tracking, or warehouse management — will command premium rates as the market shifts from configuration expertise to AI orchestration expertise.[^24]

References

The AI-Driven Transformation of Enterprise Software ... - How AI shifts enterprise software licensing from per-seat fees to token/consumption pricing — token ...
The End of Per-Seat Software Licensing: How AI Agents Are ... - The Future of Software Licensing: How AI-Driven Workflows Are Dismantling Per-Seat Models Introducti...
The Per-Seat License is Dead: Why AI Will Kill Traditional ... - George Brown from Partner Economics predicts exactly this, a 70% reduction in seats due to agent-to-...
Why AI could spell the end of per-seat software licences - Goodbye, per-seat software licensing? Goodbye, software as a service? Vendors ahead in the artificia...
Can usage-based pricing unlock mass adoption of enterprise AI ... - Salesforce and Microsoft are betting on usage-based pricing to scale AI agent adoption. Can this mod...
Salesforce Rewrites The Rules Of Pricing - Forrester - Salesforce’s AELA introduces flat-rate, unlimited AI agent usage. Learn how this pricing model resha...
Salesforce Now Has 3+ Pricing Models for Agentforce. And ... - Zendesk: Moved to outcome-based pricing for AI agents starting at $1.50 per automated resolution (or...
2026 Microsoft Copilot Consulting Pricing And Licensing Guide - Master the latest Microsoft Copilot pricing and licensing 2026 updates. Compare enterprise plans and...
Microsoft 365 Copilot: Licensing, Pricing, ROI - As of October 2025, Sales, Service, and Finance solutions are included in Microsoft 365 Copilot at n...
SAP Joule Agentic AI 2026: The Future Of ERP - Automate 80% of routine tasks with SAP Joule Agentic AI. Move beyond chatbots to achieve 10X speed. ...
Salesforce opts for seat-based AI licensing - The Register - : Analysts say the shift is part of a trend which offers stability, but embedded usage caps ensure s...
SaaS meets AI agents | Deloitte Insights - As AI agents pervade the SaaS market, how businesses experience and leverage software will likely ch...
Should You Build a Custom ERP? The Honest Decision ... - Should you build custom ERP or buy commercial? This guide covers the real trade-offs: vendor depende...
Gartner Predicts 2026: The Future of ERP - RamBase - Download the new Gartner report to discover how AI, ecosystems, and cloud sovereignty are revolution...
Has Oracle knocked SAP off the ERP throne? - CIO - A new study estimates that Oracle became the leading provider of ERP applications in 2024, stealing ...
How Big is the ERP Market? (2025) - Global ERP market size is $73B in 2025. Oracle overtakes SAP as #1 vendor. Complete data on vendors,...
Build or Buy Enterprise Software A Practical Decision Framework with ...
Build vs. Buy AI Agents - Use a neutral 10-point scorecard to decide which AI agents to build and which to buy—plus KPIs, TCO ...
Enterprise AI in 2025: An 80/20 Balance of Buy vs. Build - As of 2025, artificial intelligence is no longer a novelty—it's an operational necessity. With AI ag...
McKinsey Just Confirmed What We've Been Saying - McKinsey research reveals 60% of AI implementations fail due to poor ERP integration, costing Europe...
How Agentic AI Will Enable Businesses to Leapfrog the Legacy ERP ... - Agentic AI enables businesses to leapfrog legacy ERP monoliths through composable architecture. Gree...
AI Agents: Future of Business Process Automation in 2026 - Here is the paradox most companies face right now. Keeping legacy systems running costs more every y...
Build vs Buy: The Road to Composable IT by 2026 - LinkedIn - A silent but decisive shift in enterprise IT The debate isn’t new: should companies build their own ...
The Silent Revolution: How AI-Powered ERPs Are Killing Traditional ... - Mark Karelov reveals why 42% of ERP consultants will pivot roles by 2025 – and how to future-proof y...
Top 5 insights in 2025-26: How the BA role is evolving in the agentic-AI era - In this episode of our AI Agents for Business Analysts podcast series, we explore how the Business A...
Why a Business Analyst is crucial in Vibe coding - Does the BA role fit Vibe coding? If you're familiar with Vibe coding, you might wonder where the Bu...
Spec Driven Development with AI Coding Agent - Vibe coding has transformed how we approach development. The ability to describe what you want and g...
From Prompt Engineer to Agentic Engineer: Four Skills ... - That DevSecOps transformation required four layers of change that match the four skills I will share...
ERP modernization for AI: Transform your enterprise - Discover how AI and ERP integration drives agentic enterprise, domain transformation, and AI-enabled...
How Agentic AI is Transforming Enterprise Platforms - Agentic AI is redefining how businesses operate, installing intelligent virtual assistants that can ...
The agentic organization: A new operating model for AI | McKinsey - Discover how agentic organizations use AI-first workflows, empowered teams, and real-time data to dr...
What agentic AI requires from enterprise architecture in 2026 - Deploying AI agents safely and reliably demands rebuilding enterprise architecture around autonomy.