“We’re integrating AI into our quoting workflow.” OK — but at what level? Are you asking a model to draft quotes that humans review and send? Or building a system where AI parses the inbound, queries your ERP, drafts the quote, and routes it through an approval workflow? Or wiring AI into your CPQ tool so it operates inside the system of record?
These are wildly different projects. Different costs — a 10x range. Different timelines — weeks versus quarters. Different ROI. Different risk profiles. The mistake most companies make is treating all of them as if they were the same thing.
Here are the four levels we use to make sure both sides — partner and client — are talking about the same depth of integration.
Level 1: AI-assisted
Humans use AI tools individually.
What it looks like: Your team uses Claude, ChatGPT, or Copilot as a personal productivity tool. The sales rep drafts emails faster. The estimator gets a starting point on a quote. The CFO summarizes a long report.
Cost: $20–$50 per user per month in tool licenses, plus a few hours of training.
Time to value: 1–2 weeks for individuals comfortable with new tools.
Risk: Low. The human is in the loop on every output.
When this is right: Almost every employee should have Level 1 capability. It’s the baseline. The mistake is thinking Level 1 is your AI strategy. It’s not — it’s individual productivity. This is also where most companies stall: in McKinsey’s 2025 State of AI, 88% of organizations regularly use AI in at least one function, yet only 39% report any EBIT impact from it. Broad usage is easy. Value lives deeper.
Level 2: AI-augmented workflow
AI is embedded in a defined workflow; humans approve.
What it looks like: A specific workflow has AI built into it. The system parses inputs, drafts outputs, routes to a human for review and approval, and learns from feedback. Examples: AI-drafted quotes routed to a human approver, AI-structured field reports posted into the ERP after human review, AI-drafted customer service responses routed to a queue.
Cost: $5K–$95K per workflow (one Value Sprint), with most in the $2K–$25K range. The deepest workflows that run past ~$95K become a multi-quarter program rather than a single sprint.
Time to value: 30–60 days to a working version, 90 days to optimized.
Risk: Moderate. Errors get caught at the human checkpoint, but you need to design the checkpoint well.
When this is right: This is where the bulk of mid-market AI value lives today. Most workflows that justify AI investment should be at Level 2.
Level 3: AI-operated workflow
AI runs the workflow; humans supervise exceptions.
What it looks like: AI handles the routine flow without per-transaction human approval. Humans review exceptions, edge cases, and a sampled portion of outputs. Examples: AI handling Tier 1 customer service inquiries autonomously with escalation paths, AI processing routine document data entry with quality sampling, AI agents executing multi-step research or coordination work.
Cost: $100K+ per workflow as a multi-quarter program.
Time to value: 4–9 months to production.
Risk: Higher. Failures happen without a per-transaction human catch. You need robust monitoring, escalation paths, and the discipline to retire agents that drift.
When this is right: When the volume of routine work is high enough that human-in-the-loop becomes the bottleneck, and the consequences of any single error are bounded. Customer service, document processing, internal helpdesk, routine research.
Level 4: AI-native operation
AI is the workflow; the business model assumes it.
What it looks like: The business has restructured around AI capability. New product lines that wouldn’t exist without AI. Service models whose unit economics depend on AI. Roles in the org chart that didn’t exist before. This isn’t “we use AI” — it’s “we couldn’t operate this way without AI.”
Cost: A multi-quarter program at $100K+, scaling across multiple quarters as the restructure plays out.
Time to value: 12–36 months for a full restructure.
Risk: Strategic. You’re betting business model decisions on AI capability holding up.
When this is right: When you’ve matured through Levels 2 and 3, you’re seeing real competitive advantage from AI capability, and you want to extend that advantage by reorganizing around it. Most mid-market service businesses won’t get here for another couple of years.
The expensive mistake: matching the wrong level to the workflow
The most expensive AI mistake we see isn’t bad technology. It’s choosing the wrong integration level for the workflow at hand. And the bill is coming due industry-wide: Fortune, citing S&P Global Market Intelligence, reports that the share of companies scrapping the majority of their AI initiatives jumped from 17% in 2024 to 42% in 2025 — and the average company abandoned 46% of its proofs of concept rather than deploying them. Most of those weren’t bad ideas. They were the right idea built at the wrong level.
Over-integration. A company builds Level 3 autonomy for a workflow that runs 20 times a month. The build cost is $150K. The savings are real, but the payback is four years. They should have built Level 2 for $40K with a nine-month payback.
Under-integration. A company stays at Level 1 — “the team uses ChatGPT” — for a workflow that runs 5,000 times a month. They miss $400K of annual value because they never moved to Level 2 or 3. They could have spent $80K to capture it.
Wrong-level-by-fashion. A company builds Level 3 agents because agents are fashionable — but the workflow doesn’t need autonomy and the company doesn’t have the supervisory discipline yet. The agents drift. Quality degrades. The team loses trust.
The right level is a function of volume, consequence, team maturity, and ROI math:
- Volume — how often the workflow runs
- Consequence of error — how much a wrong output costs you
- Maturity of the team — do you have the discipline to supervise Level 3?
- ROI math — does the cost of the build pay back in a reasonable time?
How to think about your own portfolio
Most mid-market service businesses should have:
- Level 1 for everyone — every employee has access to AI tools, basic training, and sensible policies
- 3–5 workflows at Level 2 — the highest-impact workflows in your business, each measured and producing documented ROI
- 1–2 workflows at Level 3 — once you’ve proven Level 2 discipline, one or two high-volume workflows graduate to operated autonomy
- Level 4 only when you’re operationally ready — usually after 18–24 months of Levels 2 and 3 discipline
A portfolio like this typically requires $200K–$400K of investment over 18 months and produces $500K–$1.5M of annual run-rate value. Those numbers compound year over year as the workflow library grows and the team’s capability deepens.
What level is your next workflow?
If you’ve got a specific workflow in mind and you’re not sure what level fits, that’s the right starting question for a 30-minute call. We’ll walk through volume, consequence, team maturity, and ROI math — and tell you straight what level we’d recommend and what it would cost.