How to Become a Frontier Firm

I still wince at the phrase “Frontier Firm,” but this is where the eye-roll stops and the work starts. Part 1 made the case; Part 2 is about concrete moves. I thought I’d outline an example company, taking one revenue-touching process, map how it might actually run, and stand up a three-agent loop—intake, task, QA—that could plug right into your real systems with real guardrails.

If this were a real pilot/experiment, we’d then run it on 10–20% of live volume, watch four numbers (throughput, defects, cycle time, human minutes saved), and widen the gate only when the data says so. I’m trying to model out how a Frontier Firm might operate. The benefit of this approach? In plain English: fewer stuck tickets, faster quotes, cleaner collections, cash sooner.

Running these kinds of experiments will help you to better understand how this new technology works, and new ways in which you can automate and improve your operating efficiencies. By the time you templatize and roll these agents to a second project or team, your people won’t be babysitting bots—they’ll be managing digital labor with unit economics your CFO can get behind.

If you told me, “Make us a Frontier Firm,” here’s the exact sequence I’d use. No theater, just moves.

1) Pick one revenue-adjacent process you already measure

Choose a process with money or real risk: quote-to-cash, Tier-1 support, collections, onboarding, trial-to-paid. Set one target and two guardrails. For example: “Cut resolution time from 14 hours to 4, keep CSAT ≥ 4.6, <2% bad automations.”

2) Map the work the way it actually happens

Sit with the doers. Watch real tickets, invoices, or deals for a few days. Sketch the true flow: inputs → decisions → handoffs → systems → ‘done’. Circle the soul-sucking steps (triage, summarizing, routing, policy checks, status nudges, repetitive drafting). Those are agent jobs. Humans keep the judgment calls, relationship moments, and hairy exceptions.

3) Build the smallest agent loop that can ship

Use a simple agent sandwich:

• Intake agent: reads the item, extracts facts, tags intent, chooses next step.
• Task agent: performs the step (drafts reply, fills form, updates CRM, generates quote).
• QA agent: enforces tone/policy/math/required fields; ships or escalates.

Week one: human approves everything. Week two: move to exceptions-only. Tune your human–agent ratio so you don’t choke throughput with approvals or spike risk with too much autonomy.

4) Give agents hands—and guardrails

Wire your real systems: CRM, ticketing, ERP, docs, knowledge base. Add safety basics:

• Role-based access, audit logs.
• Redaction/masking of sensitive fields where possible.
• Confidence thresholds that escalate on low certainty.
• A visible kill switch (Teams command or toggle) to route all traffic to humans.

If agents can’t read and write to systems of record, you’ve built a demo, not an operation.

5) Create three artifacts before prompts

• Definition of Done (DoD): One page of what “good” means. Example: Support replies must answer the question, cite source docs, include next steps, match brand voice, and never invent links.
• Evaluation set: 50–100 real examples with ground-truth answers or policies. This becomes your regression test forever.
• Escalation rules: The automatic handoff list (VIPs, legal keywords, refunds over $X, missing contract, etc.).

Now quality debates are about facts, not vibes.

6) Turn it on—for a controlled slice of live work

Start with 10–20% of volume in one team/region. Track four numbers daily:

1. Throughput (agent-completed items)
2. Defect rate (ask-backs, policy failures)
3. Cycle time (intake → done)
4. Human minutes saved (measured via sampling)

Fix the narrowest thing that could fix what’s off: a rule, a prompt, a connector—not the whole system.

7) Manage agents like a real team

• Name owners: process owner (business outcome), system owner (plumbing), QA lead (quality).
• Run a 10-minute daily standup on queue health, top defects, and one improvement to ship today.
• Hold a weekly ops review: trend the four metrics, demo a before/after, decide whether to widen the gate (more volume, more autonomy).

This is where “every employee is an agent boss” becomes muscle memory.

8) Templatize and scale the win

Two stable weeks? Freeze it as a pattern:

• Turn your prompts/specs into a reusable template (“Tier-1 reply,” “Quote checker,” “Collections nudge”).
• Publish to an internal Agent Catalog with DoD, connectors, and metrics to watch.
• Roll to the next team. Same template, local tweaks. Ship in days, not months.

9) Rebalance roles and ratios on purpose

As autonomy grows, increase agents per person and reduce approvals.

• Frontline reps → agent bosses (triage exceptions, tune patterns, author new templates).
• Team leads → process product managers (pick problems, set targets, retire bad flows).
• QA → policy engineering (writes tests, not tickets).

Consider forming Intelligence Resources (IR): a small crew that provisions agents, governs access and quality, and upskills teams.

10) Lock in the economics

Add these to your dashboard:

• Cost per successful action (APIs + infra + tools)
• Dollar value per action (revenue generated/saved, cost avoided)
• Automation coverage (% handled by agents)
• Exception rate (% needing humans)

When finance can see the curves cross, expansion stops being an argument and starts being a plan.

A quick before/after to make it concrete

Context: B2B SaaS support swamped with “How do I…?” tickets.
Before: First reply in hours; resolution in a day; reps spelunking the doc site.
After 30 days: Intake classifies to a dozen intents and enriches with CRM context; Task drafts replies with exact doc anchors and a “steps check”; QA enforces tone/policy/link validity. Week one: human-in-the-loop; week two: exceptions-only (VIP, billing/legal). Results: minutes to first reply, hours to resolution, happier customers, happier reps.

How it will feel when it’s working

• Quieter: fewer “Where is this?” pings, more “Here’s what I changed.”
• Up-stack: reviews focus on policies and outcomes, not wording.
• Ship-driven: small improvements land daily; transformation becomes a habit.

Bottom line: “Frontier Firm” may sound buzzy, but the metaphor holds. Explorers test the routes; pioneers build the town. If you run this playbook on one process next quarter, you’ll feel the frontier shift from marketing term to operating model—and you’ll have the numbers to prove it.