Copilots make your people faster. They don't make your portfolio cheaper. Real EBITDA shows up when an agent runs the workflow, not when a person prompts a chatbot. We start where the economics actually compound: L3.
Almost everyone has reached L1, chat and copilots. Almost no one has reached the tiers where AI actually moves EBITDA. The distance between those two curves is the entire opportunity.
The two curves cross before L3. Past it, the value is real and the field is nearly empty, which is exactly why we anchor there.
Chat and copilots are a real first step, but the lift is capped, and it never reaches the P&L. Letting people "play with AI" feels like progress. It isn't the same thing as banking it.
Same workflow, different question: who does the work. That's the line between a faster headcount and a cheaper function.
A company is in L2 if its data platform, function-level AI deployment, and governance posture match the L2 definition. A BOD consultant completes the classification in a two-hour working session. No survey, no scoring rubric, no hedging.
AI makes individuals faster, drafting, summarizing, coding, but people still do all the work and nothing is integrated. A real first step, and the place ~80% of the mid-market stalls.
AI grounded in your own data through retrieval, but a human still drives every step. Answers become cited and trustworthy instead of generic.
An AI agent runs a whole workflow step; a human reviews. Cost-per-task replaces cost-per-seat, the point where a function gets faster and cheaper at once.
Agents run end-to-end and humans only edge-gate the high-risk decisions. Fleets of agents work in parallel against backlogs you couldn't hire against.
Agents orchestrate other agents at organization scale. Largely aspirational in 2026, real today only in narrow domains like coding and customer experience.
The industry-standard "climb the stack" story tells portfolio companies to start at L1, build to L2, hope to reach L3, and dream of L4. That story keeps the integrator employed. It doesn't deliver EBITDA. Blue Orange Digital inverts it.
Enter at L3, expand down to the L2 substrate (and re-incorporate L1 humans at the escalation edge), then up to L4–L5 orchestration.
L1 productivity uplift is real but bounded. L2 RAG answers questions but doesn't redesign work. L3 is where a function gets faster and cheaper at the same time, where cost-per-task replaces cost-per-seat. L3 is where the economics compound.
An L3 agent without a clean L2 substrate hallucinates. A workflow agent without L1 source-of-truth data lies confidently. We re-engineer the substrate as the agent demands it, not as a 12-month prerequisite. That sequencing is the difference between transformation theater and shipped systems.
We say what we'd build, why, and what it returns. We implement alongside your stack, shadow-run before we migrate, and attribute cost per agent so the EBITDA line is defensible in an IC memo.
Chained FP&A agents draft the close; finance reviews the edge. Cost moves from seats to tasks.
Grounded draft-and-review becomes agent-run tier-1, priced on resolution instead of headcount.
Claude Code runs the change in CI; engineers review the PR. Output scales with agents, not headcount.
Representative of the bands we underwrite, calibrated to $50–500M middle-market companies, not specific client results.
How consulting engagements compound: diagnose, build, operate, repeat across the portfolio. Orchestration-agnostic, cloud-portable, built for the middle market specifically.
A two-hour, six-dimension diagnostic that classifies the tier and writes the 90-day plan. A free read for PE deal teams on a target.
Diagnoses · L1–L5Ships the L2 substrate and the first production L3 workflow agents into a function in 90 days, gated, observable, grounded in your data.
Builds · L2–L3The Agent Ops control plane for L3–L5: per-agent cost attribution, evals, and audit. The layer that makes orchestration safe to run.
Operates · L3–L5The standing partnership across the portfolio, repeating the playbook company by company, and reporting EBITDA back to the fund.
Compounds · PortfolioCumulative implementation cost, run cost, time to tier, and EBITDA lift, published, defensible, ready for an IC. Notice where the curve bends: L3 and above.
| From → To | Impl. Cost | Time | Run Cost (add) | EBITDA (cum.) | Revenue (software) |
|---|---|---|---|---|---|
| Nothing → L1 | $150–350K | 3–6 mo | $80–200K | 50–150 bps | N/A |
| L1 → L2 | $350–750K | 6–9 mo | +$400–900K | 150–450 bps | 1–3% |
| L2 → L3 | $750K–1.75M | 9–12 mo | +$1.2–2.8M | 350–950 bps | 4–11% |
| L3 → L4 | $1.5–3.5M | 12–18 mo | +$2.5–6M | 750–1,850 bps | 9–26% |
| L4 → L5 | $3–8M+ | 18–24+ mo | $10M+ | 1,450–3,350+ bps | 19–51% |
Calibrated to $50–500M revenue middle-market companies. Ranges, not promises.
L3 is where the economics compound. L4 is where they inflect. L5 is where exit multiple moves, and where the partnership becomes R&D rather than implementation.
BOD Maturity Memo, rev 3Not a deck of maturity adjectives, a diagnostic an IC can act on: the tier you're in, the tier to target, the six-dimension read, and the 90-day move.
Illustrative sample, every readout is specific to the company's own stack and data.
Straight answers, drawn from how we run the framework.
Because "using AI" almost always means L1, copilots that make individuals faster. That uplift is real but bounded: it's people-dependent, it decays with adoption, and it never changes unit cost. The headcount and handoffs are still there.
EBITDA moves when an agent runs the workflow and cost-per-task replaces cost-per-seat. That's L3, and it's where almost no company operates yet.
L1 productivity uplift is bounded, and L2 answers questions without redesigning work. L3 is the first tier where a function gets faster and cheaper at the same time, where cost-per-task replaces cost-per-seat. That's where the economics compound.
So we anchor at L3, then expand down to build the L2 substrate the agent needs and re-incorporate L1 humans at the escalation edge, and up to L4–L5 orchestration. We re-engineer the substrate as the agent demands it, not as a 12-month prerequisite.
Tier is a diagnosable fact, not a self-assessment. A company is in L2 if its data platform, function-level AI deployment, and governance posture match the L2 definition.
A BOD consultant completes the classification in a two-hour working session, no survey, no scoring rubric, no hedging.
Two hours, with read-only access to your data and AI stack. You get a six-dimension diagnostic readout, an opportunity register, a 90-day plan, and durability flags.
It's read-only and time-boxed; credentials are deleted at engagement close.
Our published bands are ranges, not promises, calibrated to $50–500M revenue middle-market companies. Reaching L3 from L2 typically runs 9–12 months for a 350–950 bps cumulative EBITDA lift; L3→L4 is 12–18 months for 750–1,850 bps.
L3 is where the economics compound, L4 is where they inflect, and L5 is where exit multiple moves.
No. The Edge suite is orchestration-agnostic and cloud-portable, and we run your agents across whatever platforms you actually have, we're not a reseller.
We're opinionated for the middle market, Databricks as our default data platform, Claude primary, Temporal for durable execution, but those are starting points, not requirements.
Both. The framework classifies and moves an individual portfolio company, and an Edge Assessment gives PE deal teams a read on a target's AI maturity for the IC memo.
Because the tiers and the Edge product suite are consistent, the same playbook repeats across a portfolio cohort.
Two hours. Read-only access to your data and AI stack. A six-dimension diagnostic readout, an opportunity register, a 90-day plan, and durability flags. Credentials deleted at engagement close.
