SG&A Procurement in the AI Era

Dan Bleicher, Partner, Inside Consulting

Published May 19, 2026

AI and large language models represent one of the most tangible near-term opportunities to reduce SG&A costs and improve operational efficiency that mid-market and PE-backed companies have seen in a generation. But realizing that opportunity requires more than deploying tools. It requires the same procurement discipline that the best operators have long applied to direct spend.

We explore why SG&A indirect spend remains chronically undermanaged, where AI genuinely delivers savings today and where it falls short, how new AI tool categories are reshaping the SG&A spend map, the real economics of the make-versus-buy decision in an era of LLM-assisted rapid development, the governance requirements for internally built tools, and how to manage AI vendor contracts in ways that most procurement teams have not yet adapted to.

We present a maturity model and decision framework to help leadership teams assess where they stand and prioritize action.

There’s a window of opportunity for management teams to leverage these new tools and secure cost advantages. Contact Inside Consulting to explore this opportunity.

AI and LLMs are not a distant promise. They are already reducing SG&A costs in measurable ways for organizations that approach adoption with rigor. For CFOs and operational leaders at mid-market and PE-backed companies, this is among the most accessible value-creation levers available today. Unlike capital investment or headcount reduction, procurement-driven SG&A improvement is directly tied to EBITDA, executable early, and builds infrastructure that compounds over time.

Yet the same organizational dynamics that have historically allowed SG&A to escape rigorous procurement scrutiny are now complicating the AI opportunity. Tools are proliferating faster than governance can keep up. Teams are building internal applications without cost-benefit discipline. Vendor agreements are being signed without the legal and pricing review they require. The companies that capture the most value from AI will be those that treat it not as a category apart from procurement discipline, but as the newest and most important reason to apply it.

This paper is a playbook for that work. It covers the full arc: understanding the SG&A visibility gap, evaluating what AI genuinely delivers today, navigating the make-versus-buy decision honestly, governing internally built tools properly, and managing AI vendor agreements in ways that most procurement teams have not yet adapted to.

The Direct vs. Indirect Attention Gap

In most organizations, procurement energy concentrates on direct spend: raw materials, manufacturing inputs, cost of goods sold. This focus is rational. Direct spend is large, visible, and closely tied to product economics. But it leaves a significant portion of the cost base undermanaged. SG&A indirect spend, encompassing software, professional services, marketing, contingent labor, facilities, and IT, rarely receives the same sourcing rigor, category strategy, or contract discipline.

For mid-market companies, this gap is particularly pronounced. Procurement functions, where they exist at all, are typically built around direct categories. SG&A spend is often managed departmentally, with Finance approving budgets but not actively negotiating or consolidating vendor relationships. The result is a fragmented, underpowered cost structure.

Visibility Is the Root Problem

The core problem is not that SG&A spend is too high in any single category. It is that no one has a complete, normalized view of what is being spent, with whom, and under what terms. SG&A invoices are coded inconsistently across cost centers. Software subscriptions accumulate on individual expense reports. Professional service agreements are renewed at the department level without reference to enterprise-wide spend. Contract terms and expiry dates go untracked.

Without a consolidated view, the organization cannot quantify what it actually spends with any given vendor across the enterprise, cannot identify which categories have the most leverage, and cannot distinguish where competitive sourcing would be productive from where it would be disruptive.

What This Costs Mid-Market Companies

Research from the Hackett Group consistently finds that top-performing procurement organizations deliver 20 to 30 percent lower cost structures in managed indirect categories compared to peers. McKinsey has documented that indirect spend savings of 10 to 20 percent are achievable in most mid-market companies through category strategy and vendor consolidation alone. For a company with $50 million in SG&A, that represents $5 to $10 million in annual savings potential, much of it directly additive to EBITDA.

Most of that opportunity sits untouched because no one has been assigned to find it.

The promise of AI in SG&A is real and measurable in specific, well-defined use cases. The following categories represent where early movers are capturing genuine, documented value today.

What AI and LLMs Can Credibly Deliver Right Now

• Contract review and redlining. Deloitte’s 2024 future of legal work research found that legal teams using AI contract review software achieve 25 to 50 percent faster contract processing times, with time savings reaching 75 percent or more for high-volume, repetitive tasks such as NDA reviews and compliance checks.
• Spend categorization and anomaly detection. LLMs can normalize and classify AP data at scale, surfacing vendor consolidation opportunities and payment anomalies that manual review consistently misses. This capability is particularly valuable in mid-market companies where SG&A spend is fragmented across departments and cost centers with no normalized view.
• Vendor benchmarking and RFP drafting. AI tools can accelerate the sourcing cycle by generating market benchmarks, drafting initial RFP documents, and summarizing vendor responses for comparative evaluation. This compresses sourcing cycle time and allows procurement teams to manage more categories simultaneously.
• Invoice auditing and accounts payable automation. According to APQC benchmarking data, the cost to process an invoice ranges from $1.77 for top-performing organizations to $10.89 for bottom performers, a gap driven primarily by automation adoption. Levvel Research puts average manual processing cost at $10 to $15 per invoice, dropping to $2 to $3 with automation, a reduction of over 70 percent.
• Back-office workforce productivity. In Finance, HR, and Legal functions, LLM-assisted tools are reducing time spent on routine drafting, summarization, data extraction, and analysis. McKinsey’s June 2023 analysis found that developers using GitHub Copilot completed coding tasks 56 percent faster, with the productivity impact on software engineering functions estimated at 20 to 45 percent of current annual spending.

The Promise vs. the Typical Year-One Reality

The opportunity is real. The timeline is frequently not. The gap between what AI vendors promise and what organizations actually realize in year one is well-documented, and understanding it is important for honest planning.

The primary friction points are consistent across industries: change management and user adoption lag behind tool deployment; data quality issues surface only after implementation begins; integration complexity with legacy ERP and finance systems is consistently underestimated; and AI-generated outputs require more human review in early deployment than vendors typically represent. Organizations that close this gap faster treat AI tool adoption as a procurement and implementation project, not a technology deployment.

The New AI Spend Categories

AI has created a set of SG&A spend categories that did not exist five years ago and that most procurement frameworks have not yet incorporated. These include:

• LLM API subscriptions and consumption fees. OpenAI, Anthropic, Azure OpenAI, and Google Gemini all offer API-level access with token-based pricing. These are not flat subscription fees. They scale with usage in ways that are difficult to forecast and easy to underestimate.
• AI-assisted coding tools. GitHub Copilot, Cursor, and Windsurf are increasingly standard in engineering organizations. Per-seat pricing is predictable, but usage patterns vary significantly and consolidation across teams is rarely evaluated.
• AI-augmented SaaS with embedded consumption pricing. Many traditional SaaS vendors are introducing AI features on consumption models layered on top of existing seat-based agreements. Finance teams often do not know these charges are accruing until they appear on invoices.
• AI-enhanced professional services. Consulting firms, law firms, marketing agencies, and staffing companies are increasingly using AI tools in their service delivery. The cost structures and billing models for these services are changing, and procurement teams need updated frameworks for evaluating and negotiating them.

Shadow AI and Tool Proliferation

Individual contributors and department heads are purchasing AI tools on expense reports and departmental budgets with no central visibility. This is not primarily a cost crisis. It is a governance gap. The consequence is that organizations are paying for multiple overlapping tools, cannot negotiate enterprise pricing because they lack consolidated volume data, and cannot evaluate which tools are actually being used versus which have been abandoned.

The SaaS sprawl problem of the 2010s is repeating itself, faster, with higher unit economics and more sensitive data implications.

Pricing Models That Break Traditional Budgeting

Token-based and consumption pricing is structurally incompatible with annual budget cycles. A development team that increases its use of an LLM API by 3x in response to a new project does not trigger a procurement review. The cost simply appears in the next invoice. The fix is not complex, but it requires intentionality: usage tracking, spend caps negotiated into vendor agreements, and defined approval thresholds for consumption above a baseline.

The make-versus-buy decision has always been central to procurement strategy. LLM-assisted development has changed its economics significantly, but has not eliminated it. If anything, the temptation to build has become so low-friction that the decision is being made, implicitly, thousands of times a day across organizations, without any formal analysis.

What Vibe Coding Actually Is

Vibe coding is the informal term for LLM-assisted rapid prototyping: using tools like Cursor, GitHub Copilot, or direct API access to generate functional software through natural language prompts, often with limited formal programming knowledge. The barrier to building an internal tool has dropped dramatically. A finance analyst, a marketing manager, or an operations director can now produce a working application in hours or days that would previously have required weeks of developer time.

This is genuinely useful. And it is also generating a new category of organizational risk that most companies are not yet managing.

The Promise vs. the Reality of LLM-Assisted Development

The promise is compelling: custom tools, perfectly tailored to your workflow, built faster and cheaper than any vendor could deliver them. The reality is more complicated. LLMs are excellent at generating plausible-looking code. They are significantly less reliable at generating code that is secure, well-documented, maintainable, and production-ready. The gap between a working prototype and a tool that can be safely deployed at organizational scale is larger than most non-technical stakeholders appreciate, and the gap is not obvious from looking at the prototype.

LLM-generated code frequently contains security vulnerabilities that are invisible to non-expert reviewers. It is often brittle: functional under expected conditions, prone to failure at the edges. It is almost never documented in ways that allow a second person to maintain or extend it.

Vibe Coding Is Not Free: The Hidden Cost Stack

The single most common misconception about internally built tools is that they are free or nearly free because no vendor invoice is involved. This is not accurate. The true cost of a vibe-coded internal tool includes all of the following:

• Human labor. The person doing the vibe coding is not free, whether that is a full-time employee redirecting time from primary responsibilities or a contractor engaged for the purpose. Hours spent prompting, debugging, testing, and iterating are real labor costs that are rarely tracked or attributed to a project budget.
• LLM API and tool costs. Extended development sessions using Cursor or GitHub Copilot, or direct access to LLM APIs, accumulate costs that are typically expensed individually, never aggregated, and never compared against the alternative of buying a vendor solution.
• Cloud infrastructure. Vibe-coded tools frequently spin up compute instances, databases, storage buckets, or other cloud resources. Those AWS, Azure, or GCP charges appear on the next invoice and are almost never traced back to the originating project.
• Maintenance and iteration. The original builder is almost always the only person who fully understands how the tool works. When they move to a different role or leave the organization, the tool becomes a black box requiring expensive reconstruction.
• Security and compliance remediation. When LLM-generated code is deployed into production, security vulnerabilities are often discovered later through audits or incidents. Remediating these post-deployment is substantially more expensive than building securely from the start.

The Make vs. Buy Framework

Before any significant internal tool is built, a disciplined organization should work through the following decision factors:

When Building Actually Makes Sense

There are legitimate cases for internal development. Highly proprietary workflows with no viable vendor analog are a genuine example. Tools that represent a source of competitive differentiation can justify the investment. Simple internal utilities with a short intended lifespan, low data sensitivity, and a clear deprecation plan are often not worth the procurement overhead of a vendor engagement.

But these cases are narrower than the current default behavior suggests. Most internal tools that organizations are building today are substitutes for vendor solutions that already exist, are cheaper on a fully-loaded basis, and are better maintained and supported than anything a single employee can build and sustain.

Even when the make decision is correct, the way most organizations currently build internal tools creates a separate and significant problem. The speed of LLM-assisted development encourages individuals to build and deploy before anyone else has reviewed, tested, or approved the tool. This produces tools that work for the builder and frequently fail for everyone else.

The One Person in a Room Problem

When a single person builds a tool to solve a problem they personally experience, the result reflects their interpretation of the problem and their assumptions about how the tool will be used. Absent structured stakeholder input, internal tools routinely duplicate features that already exist in other systems, fail to account for edge cases obvious to other users, create data handling practices that violate IT security policies, and generate ongoing support burdens for teams never involved in the design.

Who Needs a Seat at the Table

For any internally built tool that will be used by more than one person, touch organizational data, or become part of an operational workflow, the following stakeholders should be engaged before development begins:

• End users. Not just one representative. A cross-section of the people who will actually use the tool, including those whose workflows differ from the builder’s.
• IT and Security. Any tool that accesses organizational systems, stores data, or connects to external services must be reviewed for security requirements, authentication standards, and data classification implications.
• Finance. Build decisions with meaningful cost implications, including infrastructure spend, labor allocation, and vendor displacement, require Finance visibility before commitments are established.
• Legal and Compliance. Tools that handle customer data, employee data, financial data, or any regulated information category require legal and compliance review before deployment.
• The process owner. The person accountable for the workflow the tool supports needs to sign off on the problem statement, the proposed solution, and the ongoing support model.

A Lightweight Governance Checklist

• Problem statement sign-off. Is the problem clearly defined and agreed upon by the people who will use the solution?
• Make vs. buy decision documented. Has a vendor alternative been evaluated? Has the fully-loaded 3-year cost been estimated for both options?
• Data classification review. What data will the tool access, store, or process? Has IT reviewed the data handling requirements?
• Cost estimate with 3-year horizon. Has the full cost stack, including labor, infrastructure, maintenance, and security, been estimated?
• Rollout and support plan. Who owns the tool after it is built? Who provides support and manages updates?
• Exit and deprecation criteria. Under what conditions will the tool be retired? What is the plan for data migration and user transition?

Most procurement and legal teams have developed mature processes for evaluating and negotiating traditional SaaS agreements. AI vendor contracts require a different set of review priorities that most standard procurement frameworks do not yet address.

Data Privacy and Model Training Terms

Many AI vendor agreements, particularly those with consumer-oriented or freemium pricing tiers, include clauses that permit the vendor to use customer inputs and outputs to train or fine-tune their models. For companies handling sensitive customer data, financial information, or health-related information, this is a material risk. Legal and procurement teams need to review AI vendor agreements specifically for model training clauses, data retention policies, and the conditions under which data is isolated from general training pools.

Enterprise agreements with major AI vendors typically offer stronger data protection terms, but these terms require active negotiation and are often not reflected in standard online agreements.

Model Versioning and Output Quality Risk

AI model updates can change output behavior materially. A workflow built on a specific model version today may produce substantially different outputs when the vendor updates or deprecates that model. This creates operational risk that has no analog in conventional software procurement.

Procurement should negotiate for advance notice of model changes, the ability to remain on prior model versions during transition periods, and clear communication about deprecation timelines.

Consumption Pricing Controls

LLM API pricing is usage-based. There is no natural ceiling on spend unless one is explicitly negotiated and technically enforced. Practical controls include: spend caps at the account and project level, automated alerts when consumption approaches defined thresholds, tiered approval requirements for consumption above a baseline, and monthly reconciliation of actual AI spend against budget.

Consolidation Opportunity

Most mid-market organizations that have not conducted an AI and SaaS rationalization review in the past 12 months are paying for multiple overlapping tools. Inside Consulting’s experience in SG&A rationalization engagements consistently surfaces 20 to 35 percent cost reduction opportunities in organizations that have not previously conducted this analysis.

The following maturity model assesses SG&A procurement capability across five domains. Score each domain from 1 (ad hoc) to 4 (optimized). Total scores: 5 to 10 indicates significant opportunity; 11 to 16 indicates developing capability; 17 to 20 indicates leading practice.

Organizations scoring below 12 across these five domains have a significant, near-term EBITDA improvement opportunity in SG&A procurement. The question is not whether the opportunity exists. It is whether the organization has the bandwidth and methodology to capture it systematically.

Inside Consulting is a veteran-led, mid-market management consultancy founded in 2005. We work with PE sponsors and C-suite leaders to execute rapid, step-change profitability improvements, with fees at risk tied to results.

Our SG&A procurement engagements typically follow a consistent arc: a spend diagnostic that establishes full visibility across categories and vendors; a category strategy that prioritizes opportunities by impact, feasibility, and organizational readiness; a vendor rationalization that consolidates fragmented relationships and captures negotiated savings; and a governance design that sustains the savings through ongoing M&A and organizational change.

On the AI and vibe coding front specifically, we bring both the procurement methodology and the technical literacy to evaluate make-versus-buy decisions honestly, assess the true cost of internally built tools, and design governance frameworks that allow innovation without unmanaged risk.

Our model is hypothesis-driven and accountability-oriented. We do not charge for assessments that do not lead to clear action. We do not collect fees on hypothetical/indicative savings. We audit our results and demonstrate the savings hit the P&L.

AI represents a genuine SG&A cost reduction opportunity. It is already delivering measurable results in contract management, spend analytics, AP automation, and back-office productivity for organizations that approach it with discipline. The companies that build procurement rigor around SG&A and AI spend in the next 12 to 18 months will have a structural cost advantage that compounds over time.

The opportunity does not require a large investment or a long runway. It requires clear ownership, honest analysis, and execution discipline. The window is open. If you’re ready to explore the opportunity, let us know.

(Prefer Plain Text:  Click here)

About the Author

Dan Bleicher Partner, Inside Consulting

Dan is a Partner at Inside Consulting with more than 11 years of management consulting experience, including McKinsey & Company. He specializes in procurement transformation, supply chain optimization, and operational value creation across healthcare, pharmaceuticals, and advanced industries. Dan graduated from the United States Naval Academy and earned an MBA from Dartmouth’s Tuck School of Business.

Connect on LinkedIn →