Shift Living — Phase 0 Proposal

Executive Summary

Four sequenced moves across Phase 0

SaaS Consolidation

Redundant tools removed, migrated, or folded into HighLevel and Claude. ~$4k/yr saved. Excel/Project stay pending three tied workflows.

Company Knowledge Base

Scoped AWS Bedrock pilot over steward-selected authoritative folders. Sensitive data excluded; every answer cites its source; delivered to Claude via MCP.

Claude Training

Built on NBER adoption research: founder mandate first, one named steward per department, real Shift work only, a four-question validation standard taught with every skill.

AI Impact Surveys

Baseline in week 1, results measured in weeks 4 and 8 — tracked against NBER benchmarks.

The Starting Point

From the July 2026 AI Audit & Strategic Plan

Today's reality

A dozen-plus overlapping tools across the stack
No single AI platform — ChatGPT, Gemini, and Perplexity used side by side
No owned knowledge base; answers live in multi-hundred-page Drive docs
Prior tool rollouts without owners or training failed to stick
US investor PII and Costa Rican personal data need one data-handling policy

~$4kannual savings from removals

~$18kestimated annual software spend after Phase 0

18assumed users across departments

4moves sequenced across Phase 0

System Inventory — Remove / Replace

Potential $4k saved per year

Application	Cost	Recommendation
DocuSign	$360/yr	Move to DocuSeal (e-signatures)
Calendly	$120/yr	HighLevel built-in calendar
Flodesk	$120/yr	HighLevel email design
Texting Betty	$1,788/yr	HighLevel SMS automation
Squarespace	$228/yr	HighLevel landing pages
Follow Up Boss	$696/yr	Already replaced by HighLevel
Perplexity / ChatGPT	$240/yr each	Consolidate to Claude Team
Gemini	in Workspace	Turned off org-wide in admin
ILovePDF	$60/yr	Train on Claude instead
Todoist	$60/yr	Redundant to ClickUp

Key Consolidation Rationale

DocuSign → DocuSeal

DocuSeal Pro offers MCP access at $240/yr/seat (since April 2026); DocuSign's Claude connector does not work as of 8/10/26
Customers include Allstate, LEGO, Intuit; trails only in native Salesforce/HubSpot integrations
HighLevel signing rejected: its API does not support pre-filling fields
Switching cost: counsel's sign-off on e-signature validity for ~280 attorney-run reservation agreements and PSAs; export completed envelopes before cancelling

One AI platform: Claude Team

One data-handling policy for US investor PII and Costa Rican personal data
One admin console for provisioning and offboarding
One training curriculum; one place for Connectors and Projects
Team plan does not train on customer data; already in use at Shift
Personal Perplexity/ChatGPT retired at the mandate week; Gemini turned off in Workspace admin

Explore — New Additions

About $1k/yr in new subscriptions, plus usage-based costs

Application	Cost	Rationale
DocuSeal	$240/yr	Pro e-signatures with MCP access
Recall.ai	$0.50/hr rec + $0.15/hr trans	Utilization-based meeting recording via one email-alias bot
Twilio	~$13.80/yr base	Best-in-class CPaaS; integrates with HighLevel and Claude; US + Costa Rica numbers
Lovable	$252/yr	Rapid web-app prototyping; move to AWS at product-market fit
Cloudflare	Free	Keep DNS separate from web hosting so email survives a HighLevel outage
Knowledge Base	$3.6k–$8.4k/yr	Bedrock GraphRAG (~$700/mo) vs PromptOwl ContextNest ($299/mo); 2–3 day setup

Keep — Integrate with Claude

"Keep" apps all have a Claude Connector available; connection still needs testing

HighLevelClaudeQuickBooksClickUpSlackGoogle WorkspaceZapierZoomCanvaMeta AdsLinkedInCapCut ProFinal Cut ProRevitNavisworksAutoCADWhatsAppKaseya

To Be Determined: Microsoft Excel + Project

Three workflows stay tied to Microsoft: the Construction Manager's PMBOK-style Excel templates and Project schedule checks, the official credit-card expense report template, and farm crew's harvest/compost/planting logs. Shift keeps at least one seat (~$270/yr: Business Standard + Project Plan 1) while those workflows retire. License count and owners to be established in the account-ownership review.

AI-Native System Interfaces

All interfaces tie back to one intelligence and automation platform

Claude — intelligence & automation platform

Knowledge Base (Bedrock)
MCP connection

HighLevel · CRM & Marketing
Claude Connector

DocuSeal · E-signatures
Claude Connector

ClickUp · QuickBooks · Slack
Claude Connectors

Drive · Recordings & transcripts
Sources & context

Enablement in this phase: no investor/HR/legal/personal data in prompts until the usage policy is approved. Contract templates, brochures, SOPs, and steward-selected KB folders are in scope day one.

Access & Use — Estimated Annual Cost

~$18k software subscriptions + knowledge base and usage-based costs (18 users)

$4,320Claude Team — $20/m × 18 users

$3,675AEC Collection — Revit + AutoCAD + Navisworks + Forma

$3,564HighLevel Unlimited — $297/m, needed for API access

Per-user subscriptions (annualized)

Google Workspace	$1,512	Starter $7/m × 18
Slack	$1,566	Pro $7.25/m × 18
ClickUp	$1,512	Unlimited $7/m × 18
QuickBooks	$456	Simple Start $38/m × 1
DocuSeal / Canva / CapCut	$600	$240 + $180 + $180
Zoom / Lovable	$412	$160 + $252

Usage-based

Meta Ads: $0.50–$3.50 CPC + $7–$25 CPM
Recall.ai: $0.50/hr recording + $0.15/hr transcription
Twilio: $0.1031 per SMS
Free tiers may keep early costs well under these figures; most lines grow with headcount

Training & Change Management — Principles

NBER WP 33777 (Humlum & Vestergaard, 25,000 workers): adoption peaks when encouragement, an enterprise tool, and training are combined. Any one alone underperforms.

1 · Mandate before material

Founder states in writing and at all-team that AI-assisted work is the baseline (the Shopify pattern). Robert Althuis sponsors.

2 · Standard tool only

All training on the company Claude Team account. Personal ChatGPT/Perplexity retired at the mandate week.

3 · Real Shift work

Participants bring the live email, contract, brochure, or reconciliation. Nothing hypothetical.

4 · Generation + oversight together

The validation standard is the second half of every module, not a separate compliance session.

5 · A steward or no launch

Prior rollouts (Monday.com, Asana, Todoist, early Claude seats) failed on no-owner. One owner per department before session one.

6 · Measure reorganization, not hours

Track daily use, new tasks taken on, validation compliance, and where recovered time went.

Prioritization

Cohorts staggered around real calendars; leadership modeled visibly

Cohort A
Weeks 1–2

Force multipliers & future stewards. Grecia Yepez Ortiz, Catalina Velásquez, Ivana Alvarez, Micaela Rebatta Mendoza, Valeria Arias, Oksana Koval — six people touching the most workflows; several already use AI daily. They become department stewards.

Cohort B
Weeks 2–3

Department leads. Raymond Brealey, Brian Bratton, Andrew Mercer, Sidra Gross, Stephanie Valenzuela, Aarón Araya Yannarella, Alexa Gonzalez, agency POC, Dustin Hein, Daniela Villalobos Palma (narrow Spanish track: expense capture, nursery records), IT freelancer (admin only). Brokerage sessions scheduled around showings.

Leadership
Parallel · 45 min

Matthew Blesso, Robert Althuis, Ginger Rivadeneira. Review-and-approve of EA-drafted replies, AI decision briefs, and modeling encouraged use publicly. Worker perception of encouragement is what drives adoption.

Curriculum

Validation standard taught in every module

Module 0

Mandate & setup — 45 min, all staff, day 1. Provision Claude, retire personal accounts, live demo, one-page interim data rule, #ai-help channel.

Module 1

Projects & shared knowledge — 90 min, day 1. Connectors, Claude Projects per workstream, validation standard introduced.

Module 2

Role tracks — 2 × 90 min. Workflows per role: sales (increase selling time), executive support, people ops, finance, design/construction, marketing, admin/IT. Ends with one real deliverable passing validation.

Module 3

New-task roles — stewards, 90 min. Quality review, integration, steward-written usage policy replacing the interim rule (counsel-reviewed).

Module 4

Reinforcement — ongoing. Biweekly show-and-tell, monthly steward review, Modules 0–2 folded into hiring onboarding.

Performance Targets vs NBER Benchmark

Survey reuses Humlum & Vestergaard wording (WP 33777, rev. Mar 2026); baselines in week 1, repeat weeks 4 and 8

Ever used Claude for work

Daily use

Saving > 1 hr/day

Time reallocated

New AI-related tasks

KB use, % of staff

Target Wk 8NBER benchmark

Measured, not self-reported

Deliverables failing validation: < 20% of dept samples by wk 4, < 10% by wk 8 (steward sample)
Reallocation of saved time: 85% baseline → 90% target (survey + steward log)
Personal software accounts still in use: 0 by wk 4
Week 8 targets exceed the NBER benchmark: a small, fully-trained knowledge team should outperform the population average
Surveys built with CogBias.ai to avoid cognitive bias in question design

What happens next

Week 1

Founder mandate in writing and at all-team; baseline survey; Claude Team provisioned; stewards named

Weeks 1–2

Cohort A training begins; scoped knowledge-base pilot setup (Bedrock vs ContextNest, 2–3 days)

Weeks 2–3

Cohort B role tracks; leadership track; consolidation confirmations with the HighLevel agency

Week 4

First survey re-measure; steward usage and validation review

Week 8

Second re-measure against and beyond NBER benchmarks; re-prioritize the Strategic Plan's twelve initiatives with Robert

Open items: Microsoft license count & ownership · counsel sign-off on e-signature switch · HighLevel account ownership in writing · AWS knowledge-base pricing once folder scope is known