Confidential concept proposal
Skyline AI
Operations Initiative
A proposed twelve-month partnership to build the operating capacity for growth without proportional administrative overhead.
Read the opportunity01 / The opportunity
Build operating capacity before complexity becomes the constraint.
Brad, our meeting made the first priorities clearer: expedite commission payout preparation, reduce the effort of recruiting, and keep a reliable count of closed transactions. These support the larger objective we discussed—protecting your time while building Skyline’s capacity for growth.
Skyline has an unusual opportunity. The brokerage can deliberately design its administrative and marketing operation around AI-supported work while the organization is still relatively lean.
Your personal production target and Skyline’s growth plan would likely create more coordination, follow-up, and administrative work under a conventional model. My proposal is to consider the operating layer before that complexity accumulates.
Business objectives identified by Brad
These growth goals from our earlier conversation remain the planning context; timing and targets can be reconfirmed as we define the first phase.
Brad’s personal production
- Reach approximately 48 closed transactions annually.
- Simplify CRM management and daily operating routines.
- Reduce time spent on coordination and administrative follow-through.
- Protect more time for relationships, clients, negotiation, and production.
- Add three properties and reduce direct rental-management burden over three years.
Skyline brokerage
- Grow from approximately 21 agents to 40 agents and a second office in one year.
- Grow toward 60 agents across Draper, St. George, Salt Lake, and Park City in three years.
- Create a meaningful Google and local-search presence.
- Improve recruiting, onboarding, agent support, and operating consistency.
- Scale without adding administrative overhead at the same rate as growth.
Systems already in use SkySlope for digital signing and file management; Lofty for CRM, website, email, and text; Canva and Canva AI for marketing and document templates; Google Calendar and shared Drive; UtahRealEstate.com as the MLS; and Google/Meta recruiting advertising with Cr8. The first step would be to understand the configuration and access available in these existing systems.
What needs to work better
- Commission preparation: reduce calculation and document-handling time between the settlement statement, Brad’s approval, filing, and the bank payout.
- Recruiting: connect the existing $1,500 employee introduction incentive, Brad’s personal network, and Google/Meta advertising through Cr8. Confirm the incentive’s eligibility and payment trigger.
- Transaction visibility: replace repeated manual counts in the MLS brokerage profile with a checked, regularly updated tracker.
- Meeting preparation: prepare a PowerPoint slide for each meeting, using approved figures and the purpose of that meeting.
Support across Brad’s work
- Brad operates several businesses. Each needs a clear scope, data boundary, cost allocation, and responsible person.
- Short-term rentals could benefit from scheduling, service-call preparation, vendor coordination, and task tracking.
- People handle signs, listing visits and cleaning, photography appointments, furniture, shopping, errands, and gift fulfillment. AI could help organize the requests.
- Personal and household assistance would be arranged separately from the Skyline engagement.
annually
within one year
within three years
02 / The approach
A business design project.
Not a software setup.
Claude Cowork may be useful. It is one candidate workspace, not the architecture or the default for every workflow.
The initiative begins by understanding how work moves through the business. Only then should a task be assigned to a person, an AI workflow, an existing platform, or an outside vendor.
Process first. Tools second.
For each recurring activity, we would define the trigger, source information, authority, approval, system of record, output, and exception path before selecting or connecting technology.
- 01Trigger
- 02Source information
- 03Authority
- 04Approval
- 05System of record
- 06Output
- 07Exception path
- 08Technology
In the proposed model, the AI Chief of Staff is a software-based coordination layer—not Jim or a human employee. It would organize approved work, coordinate specialized AI functions, prepare decisions, and escalate exceptions. Brad would retain final decision authority.
My proposed role would be to architect, implement, and govern the system so it can improve as Skyline grows. This would not make me Brad’s day-to-day assistant.
03 / The architecture landscape
No single AI is presumed to run Skyline.
I would evaluate the available landscape and recommend the smallest reliable system for each workflow. We could extend what Skyline already owns, connect specialist services, use different frontier models for different jobs, or build a controlled layer where commercial tools fall short.
Select by outcome, accuracy, access, security, integration, cost, and reversibility—not by loyalty to one vendor.
Frontier foundation models
Reasoning, language, vision, coding, and multimodal work.
OpenAI GPT and reasoning models · Anthropic Claude · Google Gemini · xAI Grok · Meta Llama · Mistral · DeepSeek · Amazon Nova · Cohere Command · Qwen
Managed model & agent platforms
Secure deployment, routing, scaling, identity, and controls.
OpenAI API and AgentKit · Microsoft Foundry · Google Vertex AI Agent Builder · Amazon Bedrock and AgentCore · Anthropic API · NVIDIA NIM · Hugging Face · Cloudflare Workers AI
Business AI workspaces
Research, drafting, analysis, projects, and team adoption.
ChatGPT Business or Enterprise · Claude Enterprise, Cowork, and Code · Gemini Enterprise · Microsoft 365 Copilot and Copilot Studio · Grok Business or Enterprise · Perplexity Enterprise · Glean
Agent orchestration
Specialized agents, handoffs, memory, and controlled execution.
OpenAI Agents SDK · LangGraph and LangSmith · LlamaIndex · Google Agent Development Kit · Microsoft Agent Framework and Semantic Kernel · CrewAI · PydanticAI · AutoGen
Plugins, tools & connectors
Permissioned access to business systems and external actions.
Model Context Protocol servers · ChatGPT apps and plugins · Claude connectors · Microsoft Copilot connectors · REST and OpenAPI tools · Webhooks · Custom functions and internal APIs
Workflow automation & RPA
Deterministic workflows, approvals, retries, and legacy systems.
n8n · Zapier · Make · Workato · Microsoft Power Automate · UiPath · Pipedream · Relay.app · Gumloop
Search & web intelligence
Current information, research, monitoring, and extraction.
OpenAI web search · Perplexity · Exa · Tavily · Bing Search · Google grounding · Firecrawl · Apify
Knowledge & retrieval
Grounded answers from approved company information.
OpenAI file search and vector stores · Azure AI Search · Vertex AI Search · Amazon Bedrock Knowledge Bases · Pinecone · Weaviate · Qdrant · Elasticsearch or OpenSearch · pgvector
Data & systems of record
Operational memory, analytics, reporting, and governed context.
PostgreSQL · Snowflake · Databricks · BigQuery · Airtable · Notion · Google Drive · Microsoft SharePoint · Amazon S3 · Cloudflare R2
Documents & contracts
Extraction, classification, review queues, and signatures.
Mistral Document AI · Google Document AI · Azure Document Intelligence · Amazon Textract · LlamaParse · Unstructured · Nanonets · Rossum · DocuSign · Adobe Acrobat Sign
Voice & communications
Calls, transcription, conversational voice, messaging, and routing.
OpenAI Realtime · ElevenLabs · Deepgram · AssemblyAI · Google Cloud Speech · Azure Speech · Twilio · Vapi · Retell AI · Aircall · Dialpad
Creative & marketing production
Brand-controlled copy, image, video, audio, and campaign assets.
Canva · Adobe Creative Cloud and Firefly · OpenAI image and video tools · Midjourney · Runway · Descript · HeyGen · Synthesia
Real-estate operating platforms
CRM, IDX, lead nurture, agent support, and client intelligence.
Lofty and the Lofty API · Follow Up Boss · kvCORE · Rechat · RealScout · Homebot · CINC · MoxiWorks · Sierra Interactive
Transactions & back office
Files, signatures, compliance workflow, accounting, and payments.
SkySlope · Dotloop · DocuSign · Lone Wolf Transactions · Brokermint · QuickBooks · Xero · Ramp · Stripe · Plaid
Engineering & operations
Custom development, hosting, monitoring, and incident response.
Codex · Claude Code · GitHub Copilot · Cursor · GitHub · Vercel · Cloudflare · AWS · Microsoft Azure · Google Cloud · Datadog · Sentry
Evaluation, security & governance
Testing, traces, access controls, auditability, and risk management.
OpenAI evaluations and tracing · LangSmith · Braintrust · Arize Phoenix · Weights & Biases Weave · Patronus AI · Lakera · Microsoft Purview · Cloud-provider guardrails · SSO and audit logs
Representative landscape as of September 2026—not a commitment to purchase or deploy every product. Availability, licensing, data handling, integration access, and total cost would be verified before any selection.
04 / The proposed AI staff
Specialized functions. One governed operating system.
I would introduce the functions gradually, beginning with the recurring work that consumes the most attention and creates the clearest return.
| AI role | Representative responsibility |
|---|---|
| AI Chief of Staff | Daily operating brief, approval queues, cross-system coordination, exception escalation, and a PowerPoint slide prepared for each meeting from approved information. |
| Lead & CRM Coordinator | Captures and classifies leads, maintains next actions, identifies neglected opportunities, and prepares follow-up. |
| Listing Operations Coordinator | Tracks each listing lifecycle, organizes assets and deadlines, and initiates status-dependent tasks. |
| Marketing & Social Coordinator | Uses existing Canva and Canva AI workflows to prepare social posts, flyers, agent introductions, approved document templates, and content calendars for review. |
| Client Experience Coordinator | Prepares client and agent gift selections, budgets, reminders, review requests, and relationship touchpoints; people approve purchases and handle delivery. |
| Recruiting & Agent Success | Tracks employee introductions, Brad’s network, and Google/Meta campaign leads with Cr8; prepares follow-up, meeting materials, and onboarding handoffs while Brad builds relationships. |
| Brokerage Administration | Prepares commission summaries with brokerage fees, agent payout, and cap position; reconciles closing counts and flags discrepancies for human review. Banking and payment release remain with authorized people. |
| Property Operations | Could support short-term rental scheduling, service-call preparation, vendor coordination, and property task lists under a separately agreed scope; physical execution remains with people. |
What this looks like in practice
First example: from settlement statement to payout approval
I propose starting with the commission process you described. AI could extract and organize the inputs; approved formulas would calculate amounts. Brad reviews the packet, and authorized staff release payment. SkySlope’s existing commission and CDA features would be assessed before adding a custom Word/PDF step; the edition and available access still need confirmation.
- 01Collect the settlement statement and SkySlope transaction record
- 02Check agent split, brokerage fees, and current cap position
- 03Calculate amounts using approved rules; flag missing or conflicting inputs
- 04Prepare an editable Word summary and approval PDF
- 05Present brokerage fee, agent payout, and cap totals to Brad for approval
- 06File the approved version; authorized staff release the bank payment
- 07Reconcile payment status and update the closed-transaction tracker
The final agent roster, workflow order, and technology stack would be established collaboratively during discovery. The aim would not be to create as many agents as possible, but to address the most valuable recurring burden first.
05 / Proposed year-one roadmap
Build in controlled stages, prove reliability, and expand according to value.
Architecture & foundation
See the operation clearly before changing it.
- Interview Brad and key team members.
- Inventory systems, data, roles, and recurring work.
- Map the commission approval process, agent split and cap rules, recruiting handoffs, and transaction-count definitions.
- Establish financial, time, and operating baselines.
- Define human and AI authority, approvals, and data access.
- Test a sample commission approval packet and closing tracker against verified records before production use.
Core deployment
Build the highest-return workflows and prove reliability.
- Build and test the highest-return workflows.
- Prioritize commission preparation, closed-transaction tracking, and the three-channel recruiting pipeline.
- Create a reusable PowerPoint meeting-slide format and strengthen existing Lofty and Canva workflows.
- Connect approved systems and create documented exception paths.
- Train the team and measure reliability before increasing autonomy.
- Review results monthly and redirect effort where value is greatest.
Scale & optimization
Expand only where the system has earned trust.
- Expand proven recruiting and onboarding workflows, client experience, and brokerage administration; consider property support under a separate scope.
- Strengthen local-search and content production.
- Support second-office readiness and repeatable operating standards.
- Optimize costs, accuracy, speed, and team adoption.
- Conduct the year-end operating and performance review.
Likely first production workflows
Begin where the return is visible.
Commission approval
A checked summary of brokerage fees, agent payout, and cap position, delivered as an approval PDF with an editable source.
Closed-transaction tracker
A reconciled count by period and agent, with agreed rules for transactions versus sides, replacing repeated manual MLS counts.
Recruiting follow-through
One pipeline for employee introductions, Brad’s network, and existing Google/Meta campaigns, with a clear next action and relationship owner.
The proposed sequence is intentionally adaptive. The highest-return workflows would be considered first, regardless of which product ultimately powers them.
06 / Governance before autonomy
AI should earn authority through demonstrated reliability.
Human ownership
- Licensed real estate activity and broker supervision.
- Negotiation, pricing judgment, and fiduciary decisions.
- Final compliance review and legal determinations.
- Sensitive client or personnel communications.
- Client funds, physical property work, and relationship ownership.
Operating controls
- Least-privilege access to systems and data.
- Human approval before external execution unless expressly authorized.
- Action logs, source-of-record discipline, and reversible workflows.
- Documented exceptions, escalation paths, and credential controls.
- Periodic privacy, access, cost, and reliability reviews.
Autonomy ladder
Any expansion of AI authority would follow demonstrated evidence.
-
01AI prepares
Human reviews every output.
-
02AI executes
With explicit human approval.
-
03Bounded action
Within defined rules and limits.
-
04Autonomous
Considered after evidence and monitoring.
How success could be measured
Measure operating results, not automation volume.
Time
- Brad administrative hours per week
- Routine matters requiring Brad
- Time recovered for clients and growth
Revenue operations
- Recruiting leads, meetings, and joins by source
- Candidates with a defined next action
- Advertising and incentive cost per productive recruit
Delivery
- Time from settlement statement to approved payout packet
- Commission discrepancies caught before payment
- Reconciled closing counts and reporting freshness
Economics
- Administrative cost per transaction and agent
- Growth without proportional headcount
- Adjusted operating profit above baseline
07 / Next step
Turn the meeting into a focused first phase.
Confirm the commission workflow, recruiting handoffs, and reporting definitions before selecting the first implementation.
Review a redacted settlement statement, the current payout summary, agent split and cap rules, and an example approval. Confirm whether closings are counted as transactions or sides, how recruiting leads are handed off, and which systems can supply authorized exports or integrations.
If both parties agree to proceed
-
01
Finalize the professional services agreement and confidentiality and data-access terms.
-
02
Provide the initial system inventory, operating materials, and financial baseline information.
-
03
Begin interviews, workflow mapping, and technology assessment.
-
04
Confirm the first production workflows and begin the initial deployment.