Real-world systems with measurable results

44 requirements, two live client projects

Baseline hours are VFP’s own build-cycle budgets. Agent figures are measured runtime. Full derivation below.

“When we started, there was no playbook for what we were trying to build: an AI engine that could take structured requirements and configure a Salesforce org on its own. Intuitive Intel didn’t just take direction; they helped us figure out what ‘done’ looked like, then built toward it. Six months later, we’ve defined and tested against a real benchmark: 17 core Salesforce component types at a 90%+ automation rate. They’re a partner who stayed in the room with us through every round of testing. If another Salesforce partner asked whether this was worth doing, I’d say yes, but go in knowing you’re helping define the category as you build it.”

The numbers

Through August 2026, the agent was exercised on real VFP client requirements, much of it run by VFP’s own consultants rather than by Intuitive Intel.

Accuracy

Across the on-site week, we scored twelve sessions run by four operators across five Salesforce orgs: 355 configuration items in total. The engagement’s success target, set at the outset, was 70%. On the harder of our two measures, the agent came in 18.7 points above that.

Where it stands today

The engine is installed and rolling out across VFP’s delivery team. Roughly ten of the twenty consultants who do configuration work are up and running, with the rest of the team-wide rollout underway. Nothing it has built has gone into a client production org yet; every figure in this document comes from real client requirements run in sandbox and partial orgs.

VFP is now targeting the engine at two upcoming implementations, each over $1M, with a goal of cutting the go-live timeline for those clients in half.

About Campfire

Campfire is VFP Consulting’s requirements management platform, built natively on Salesforce to run implementations from discovery through post-go-live. Requirements, tasks, progress metrics, and cross-team collaboration live in one place, which is why it’s the natural home for the AI configuration agent described here: requirements go in, certified Salesforce configuration comes out, and the documentation writes itself back to the requirement it came from.

Request a demo or try it free on the AppExchange.

In their words

The setup

Valley of the Flowers serves over 250 recurring maintenance clients across three divisions (maintenance, irrigation, and installation) out of Bozeman, Montana. The office team spent the better part of a week and a half every month invoicing more than 300 clients, roughly 200 hours, all manual. Scheduling lived only in two people's heads. LMN and QuickBooks ran independently, with the built-in integration too limited to fit the actual workflow.

The catalyst wasn't tech debt. It was succession. VOTF was transitioning into an employee stock ownership program, and the founders needed the operation to run without them.

How we engaged

The engagement started with two months of forward-deployed engineering on-site in Bozeman, 20 to 30 hours a week embedded with the admin team. We worked the full operation firsthand — scheduling, invoicing, and crew coordination — to learn the business inside out before building. Process consulting from the outside would have produced clean diagrams and the wrong system; building the right one required living in it first.

That posture shaped the rest of the engagement.

What we built

What changed

What's next

Phase 4, running June through August 2026, is where we lock peak-season metrics: true monthly invoicing hours at full client load, Cassie's actual admin hours reclaimed, the foreman timesheet accuracy rate against the 90% target, and capacity unlocked for new clients without adding headcount. This page will be finalized then.

Valley of the Flowers Landscaping, Inc. 100% employee owned. Bozeman, MT.

Webpage report automation

In April, before the training, we scoped an automation of Fay’s webpage reports. The scope covered eight external data sources, a connection to the CRM, a database, branded PDF reports, a schedule tied to each publishing date, and on-demand requests from agents. We estimated 150 hours of build work. The team was spending 400 to 750 hours a year making these reports by hand.

By the end of the training, a member of the marketing team had built the core of it himself. His first version was a script that only ran on his computer. He asked Claude for ways to let agents use it, picked one and worked through it step by step. It now runs on its own and agents use it directly.

The work we expected to be hired for was done in house by the person who knew the problem best.

What the team said

Where the team started

Most of the team used Claude every day to draft emails, summarize documents and answer questions. Skills, connectors and scheduled runs were not part of anyone’s work yet. One person had built a reusable skill. Nobody had connected Claude to the tools they use every day, and nothing ran on a schedule. Cowork, where Claude works across your files and applications, had one regular user.

Nothing was holding the team back except knowing what the tools could do.

How the training worked

• Five 60-minute in-person sessions.
• Week one covered what belongs in an AI tool and what does not, before any technique.
• Weeks two through five were working sessions. Each person brought a real task and left with something that ran.
• In week four, people handed their skills to a coworker to run. Anything that only worked for its author got fixed.

What the team built