How a SaaS Firm Underwrote $700K From One AI Workflow
A PE-backed SaaS firm's AI looked done but never reached the P&L. A drafting workflow on the client's existing tools cut 1–2 weeks of consultant work to ~2 hours — with $700K in NPV underwritten over three years on top.
$700K NPV
Underwritten 3-yr, one workflow
Jimmy Bijlani
Founder & CEO
AI Momentum Partners
Connect ↓
The Challenge
A private equity-backed vertical SaaS company ($100–150M revenue, ~40% EBITDA margin) had named AI as a core lever in its value creation plan, and the board expected AI to contribute meaningfully to margin improvement in the current fiscal year. Leadership believed the heavy lifting was already done: developer tooling was rolled out, the CTO was engaged, a chatbot was live, and there was real enthusiasm. Their hypothesis — one AMP hears constantly — was 'We're already doing AI; we just need to organize it and scale it.' But none of the AI work was connected to the P&L. There were explicit cost-takeout targets in the value creation plan and a finite PE hold clock, yet no financial model behind the AI program and no operator accountable for capturing value.
What They Built
AMP ran its diagnostic-and-sequencing playbook, then executed against the P&L. Working from roughly 75 enterprise artifacts (financials, SOPs, system data, historical project data) plus stakeholder interviews from exec to frontline, AMP mapped workflows to the level where automation potential became visible. It whittled ~18 candidate initiatives to a 7-initiative roadmap, separating a foundational layer (governance, operating model, AI PMO, intake) from prioritized use cases scored for short-term P&L impact and long-term capability. Critically, the program was built on the client's existing Azure/Microsoft environment (Microsoft 365, Power Automate, SharePoint) to preserve speed-to-value and capture margin in-year with no procurement cycle. The AI agents are model-agnostic — currently Azure OpenAI (GPT-4.1), with an orchestration layer that routes individual tasks to Claude via Amazon Bedrock where it outperforms, so the solution doesn't need rebuilding when the frontier moves. The first pilot automated professional-services documentation drafting (requirements analyses, BRDs, SOWs, meeting summaries) using those agents to pull from call transcripts, discovery notes, customer handoffs, and the historical SharePoint library, with Power Automate orchestrating the workflow.
AMP started not with tools but with the P&L. Stakeholder interviews ran from the exec team to frontline operators, and the team reviewed roughly 75 enterprise artifacts — financials, SOPs, system data, historical project data — then mapped workflows in enough detail that automation potential became visible. The diagnosis broke the internal narrative: developers had 90% adoption of Cursor and Copilot, but ~60% of the codebase ran on a legacy language and proprietary IDE that neutralized the tools; support had a live chatbot fed by a knowledge base only ~20% usable; governance was fragmented and the AI policy unseen. AMP whittled ~18 candidate initiatives to a 7-initiative roadmap, separating a foundational layer (governance, operating model, AI PMO, intake) from prioritized use cases scored for short-term P&L impact and long-term capability. Rather than introduce new platforms, AMP built on the client's existing Azure/Microsoft environment to preserve speed-to-value, with a model-agnostic agent layer — currently Azure OpenAI (GPT-4.1), routing individual tasks to Claude via Amazon Bedrock where it outperforms. The first pilot — professional-services documentation drafting, orchestrated through Power Automate — was delivered in roughly three to four weeks, collapsing one-to-two weeks of senior-consultant effort to about two hours. That ~95% time reduction is realized and measured on live documents; the $700K NPV and 5–6% Year-1 margin improvement are underwritten over three years, not yet banked.
AI Role
AI drafts highly structured, repetitive professional-services documents — requirements analyses, BRDs, SOWs, meeting summaries — from existing inputs (call transcripts, discovery notes, customer handoffs, and the historical SharePoint library), producing review-ready first drafts with citations and house-style consistency that consultants review, refine, and ship. The agents are model-agnostic by design — currently Azure OpenAI (GPT-4.1), with an orchestration layer that routes individual tasks to Claude via Amazon Bedrock where it performs better, so nothing has to be rebuilt when a stronger frontier model appears.
Infrastructure
Azure / Microsoft 365 environment (already licensed), including Azure OpenAI (GPT-4.1) • Amazon Bedrock (Claude, routed per task) • SharePoint (historical professional-services document library) • Power Automate (workflow orchestration) • Internal source data: financials, SOPs, discovery notes, customer handoffs
Integration Points
Model-agnostic agent (Azure OpenAI GPT-4.1, with Claude via Amazon Bedrock per task) pulling from call transcripts, discovery notes, and the SharePoint document library • Power Automate orchestrating the drafting workflow end to end • Human-in-the-loop review / refine / ship step before documents are finalized
Impact
Realized and measured on live documents: senior-consultant drafting of BRDs, SOWs, and requirements analyses fell from 1–2 weeks to about 2 hours per engagement.
Underwritten over three years from this one workflow, plus a projected 5–6% Year-1 EBITDA margin improvement — modeled and underwritten, not yet banked.
The drafting workflow runs in daily production today; SOWs are the next workflow, with the rest of the professional-services suite behind them.
Implementation Complexity
Built primarily on the client's existing Azure/Microsoft environment (Microsoft 365, Power Automate, SharePoint, Azure OpenAI) with a model-agnostic agent layer that routes to Claude via Amazon Bedrock where it outperforms — avoiding a platform rip-and-replace and capturing value in-year. The pilot was scoped via a three-stage filter (financial materiality, feasibility in ~3–4 weeks, change capacity) and built/validated in roughly three to four weeks. The broader program sequenced 7 initiatives across a foundational governance/operating-model layer and prioritized use cases.
Best Fit For
PE-backed services and SaaS businesses ($15M–$500M) under pressure to show AI on the P&L within the hold period — especially those with heavy, repetitive knowledge work (professional-services documentation, support, engineering) and an underused enterprise software stack they already pay for.
