AI Adoption: Caliche Roofing
Where the hours are. The heaviest repeatable work sits in Dana's role (billing, collections, intake coverage, follow-up) and the owner's role (insurance supplements, marketing copy, job descriptions), with Marco's role close behind on pipeline reporting built from texts and memory.
By role
| Role | Repeatable work that eats hours | AI can carry now (with a check) | Stays human | Hours a week (owner to confirm) | Standard written down? |
|---|---|---|---|---|---|
| Dana, office manager | Drafting collection follow-up emails, payment reminder messages, billing summaries, payroll prep notes, intake responses to inbound calls and emails (including covering when receptionist is out) | Drafting collection and payment reminder emails from a template, summarizing open AR by age, drafting intake responses to common customer questions | Payroll decisions, any conversation where a customer is disputing a charge, releasing a lien, approving a write-off | (draft) 5 to 8 hrs, higher end when covering intake | Partially: billing process is routine but collection escalation steps are not written down |
| Owner (Ray), marketing and supplements | Drafting insurance supplement narratives, writing marketing copy, drafting job descriptions, responding to adjuster emails | Drafting supplement narratives from a scope template, drafting marketing copy for review, drafting job descriptions from a role outline | Pricing decisions, adjuster negotiation calls, hiring decisions, final sign-off on any customer-facing supplement | (draft) 6 to 10 hrs | No: supplement narrative standard lives in the owner's head; this is the dependency that must be resolved before AI touches it |
| Marco, sales lead | Drafting estimate follow-up emails, writing pipeline report from texts and memory, summarizing where each job stands, coaching notes for estimators | Drafting follow-up emails from a template, generating a first-pass pipeline summary from call notes, drafting coaching notes from conversation summaries | Close rate decisions, insurance estimate strategy, hiring or removing an estimator, any customer escalation | (draft) 3 to 5 hrs | Partially: follow-up cadence exists but is not written; pipeline report format is being written this quarter |
| Kevin and Sam, estimators | Drafting estimate cover emails, summarizing job scope for Luis, responding to adjuster information requests | Drafting cover emails from a template, summarizing scope notes into a document for the crew | Insurance estimate line items, pricing judgment, adjuster negotiation | (draft) 2 to 3 hrs each | No: insurance estimate standard is being written this quarter. Dependency note: Kevin and Sam did not adopt the CRM in 2024 because they kept paper notes instead of logging into the system. Any AI move in this role that requires clean logged data hits the same wall. The standard and the logging habit have to come before the tool. |
| Luis, production manager | Drafting crew schedule summaries, materials order reminders, daily job status notes | Drafting schedule summaries from a standard template, drafting materials reminder messages | Crew assignments, supplier decisions, any safety or field judgment call | (draft) 2 to 3 hrs | Partially: scheduling runs on routine but is not documented |
Dana, office manager
Repeatable work that eats hours: Drafting collection follow-up emails, payment reminder messages, billing summaries, payroll prep notes, intake responses to inbound calls and emails (including covering when receptionist is out)
AI can carry now (with a check): Drafting collection and payment reminder emails from a template, summarizing open AR by age, drafting intake responses to common customer questions
Stays human: Payroll decisions, any conversation where a customer is disputing a charge, releasing a lien, approving a write-off
Hours a week (owner to confirm): (draft) 5 to 8 hrs, higher end when covering intake
Standard written down?: Partially: billing process is routine but collection escalation steps are not written down
Owner (Ray), marketing and supplements
Repeatable work that eats hours: Drafting insurance supplement narratives, writing marketing copy, drafting job descriptions, responding to adjuster emails
AI can carry now (with a check): Drafting supplement narratives from a scope template, drafting marketing copy for review, drafting job descriptions from a role outline
Stays human: Pricing decisions, adjuster negotiation calls, hiring decisions, final sign-off on any customer-facing supplement
Hours a week (owner to confirm): (draft) 6 to 10 hrs
Standard written down?: No: supplement narrative standard lives in the owner's head; this is the dependency that must be resolved before AI touches it
Marco, sales lead
Repeatable work that eats hours: Drafting estimate follow-up emails, writing pipeline report from texts and memory, summarizing where each job stands, coaching notes for estimators
AI can carry now (with a check): Drafting follow-up emails from a template, generating a first-pass pipeline summary from call notes, drafting coaching notes from conversation summaries
Stays human: Close rate decisions, insurance estimate strategy, hiring or removing an estimator, any customer escalation
Hours a week (owner to confirm): (draft) 3 to 5 hrs
Standard written down?: Partially: follow-up cadence exists but is not written; pipeline report format is being written this quarter
Kevin and Sam, estimators
Repeatable work that eats hours: Drafting estimate cover emails, summarizing job scope for Luis, responding to adjuster information requests
AI can carry now (with a check): Drafting cover emails from a template, summarizing scope notes into a document for the crew
Stays human: Insurance estimate line items, pricing judgment, adjuster negotiation
Hours a week (owner to confirm): (draft) 2 to 3 hrs each
Standard written down?: No: insurance estimate standard is being written this quarter. Dependency note: Kevin and Sam did not adopt the CRM in 2024 because they kept paper notes instead of logging into the system. Any AI move in this role that requires clean logged data hits the same wall. The standard and the logging habit have to come before the tool.
Luis, production manager
Repeatable work that eats hours: Drafting crew schedule summaries, materials order reminders, daily job status notes
AI can carry now (with a check): Drafting schedule summaries from a standard template, drafting materials reminder messages
Stays human: Crew assignments, supplier decisions, any safety or field judgment call
Hours a week (owner to confirm): (draft) 2 to 3 hrs
Standard written down?: Partially: scheduling runs on routine but is not documented
The first three moves
| # | Move | Owner | Tool class | Guardrail | Human check | Measured by in 30 days |
|---|---|---|---|---|---|---|
| 1 | Marco uses a transcription and summary tool on his Monday pipeline calls and estimator coaching conversations. The tool produces a summary of open jobs, next actions, and flagged blockers. Marco reviews the summary, corrects it, and uses it as the basis for the pipeline report already being written this quarter. This move does not require clean logged data from Kevin or Sam. It works from the call itself. | Marco | Call transcription and summary tool | May not be used on any call where a customer is present without their knowledge. Summary is a draft, not a record. No customer personal data stored. | Marco reviews every summary before it becomes part of the pipeline report. Estimators see the pipeline report, not the raw transcript. | Time Marco spends writing the pipeline report each week: baseline now (owner to confirm), target cut by half. Pipeline report delivered on time: tracked against the 13-of-13 target already in the quarterly priorities. |
| 2 | Dana uses an AI drafting assistant to generate collection follow-up emails and payment reminder messages. She provides the customer name, amount, and days overdue; the tool drafts; she reviews and sends. Standard needed first: Dana writes a one-page collection escalation sequence (3 stages, tone for each) before the tool is set up. This is a two-hour task, not a project. | Dana | AI writing assistant that drafts from a short prompt and a saved template | May not send anything without Dana reading it first. May not reference legal action or lien language unless Dana adds it manually. No customer financial data stored in the tool. | Dana reads every draft before it goes out. No exceptions in the first 30 days. | Hours Dana spends drafting collection emails per week: baseline now (owner to confirm), target cut by half. Turnaround from invoice due to first follow-up: today vs. day 30. |
| 3 | Owner drafts the insurance supplement narrative standard, one page, covering scope language, tone, and the three most common adjuster objections. Then an AI drafting assistant generates first-pass supplement narratives from that standard and a job scope. Owner reviews and edits before anything goes to an adjuster. The standard must exist before the tool is used. This is the move that starts getting the supplement process out of the owner's head. | Owner (Ray), with Dana tracking that the standard is written by week 2 | AI writing assistant that drafts from a structured template and scope notes | May not send to an adjuster without owner review. May not invent line items or dollar amounts. No job-specific customer data stored in the tool. | Owner reads and edits every draft. In 30 days, the goal is not to remove the owner from review but to cut the time he spends drafting from scratch. | Hours owner spends drafting supplement narratives per week: baseline now (owner to confirm), target cut by at least one-third. Number of supplements drafted using the standard vs. from scratch: tracked by Dana. |
1
Move: Marco uses a transcription and summary tool on his Monday pipeline calls and estimator coaching conversations. The tool produces a summary of open jobs, next actions, and flagged blockers. Marco reviews the summary, corrects it, and uses it as the basis for the pipeline report already being written this quarter. This move does not require clean logged data from Kevin or Sam. It works from the call itself.
Owner: Marco
Tool class: Call transcription and summary tool
Guardrail: May not be used on any call where a customer is present without their knowledge. Summary is a draft, not a record. No customer personal data stored.
Human check: Marco reviews every summary before it becomes part of the pipeline report. Estimators see the pipeline report, not the raw transcript.
Measured by in 30 days: Time Marco spends writing the pipeline report each week: baseline now (owner to confirm), target cut by half. Pipeline report delivered on time: tracked against the 13-of-13 target already in the quarterly priorities.
2
Move: Dana uses an AI drafting assistant to generate collection follow-up emails and payment reminder messages. She provides the customer name, amount, and days overdue; the tool drafts; she reviews and sends. Standard needed first: Dana writes a one-page collection escalation sequence (3 stages, tone for each) before the tool is set up. This is a two-hour task, not a project.
Owner: Dana
Tool class: AI writing assistant that drafts from a short prompt and a saved template
Guardrail: May not send anything without Dana reading it first. May not reference legal action or lien language unless Dana adds it manually. No customer financial data stored in the tool.
Human check: Dana reads every draft before it goes out. No exceptions in the first 30 days.
Measured by in 30 days: Hours Dana spends drafting collection emails per week: baseline now (owner to confirm), target cut by half. Turnaround from invoice due to first follow-up: today vs. day 30.
3
Move: Owner drafts the insurance supplement narrative standard, one page, covering scope language, tone, and the three most common adjuster objections. Then an AI drafting assistant generates first-pass supplement narratives from that standard and a job scope. Owner reviews and edits before anything goes to an adjuster. The standard must exist before the tool is used. This is the move that starts getting the supplement process out of the owner's head.
Owner: Owner (Ray), with Dana tracking that the standard is written by week 2
Tool class: AI writing assistant that drafts from a structured template and scope notes
Guardrail: May not send to an adjuster without owner review. May not invent line items or dollar amounts. No job-specific customer data stored in the tool.
Human check: Owner reads and edits every draft. In 30 days, the goal is not to remove the owner from review but to cut the time he spends drafting from scratch.
Measured by in 30 days: Hours owner spends drafting supplement narratives per week: baseline now (owner to confirm), target cut by at least one-third. Number of supplements drafted using the standard vs. from scratch: tracked by Dana.
Rules of the road
- Data that never goes into an outside AI tool: Customer names, addresses, phone numbers, and email addresses. Insurance claim numbers and adjuster contact details. Payroll figures, employee pay rates, and any HR records. Banking information and any data covered by a vendor or subcontractor agreement. Job-specific dollar amounts tied to a named customer.
- Nothing reaches a customer without: A named person at Caliche reading it first and deciding to send it. This applies to every email, every follow-up, every supplement narrative. No automated sending in the first 90 days.
- What gets logged: Which tool was used, who ran it, and what was sent or filed. Dana keeps a simple log: date, tool, role, what was produced, who reviewed it. One row per use. This is how a mistake gets traced.
- Who owns these rules: Dana owns the log and the data rules. The owner sets the line on what data is off-limits and reviews the rules at the 30-day mark. If a new use comes up that is not covered here, it waits for the owner to say yes before it runs.
What waits
Automating any part of the estimating process (Kevin and Sam's role) is tempting and will save real hours, but two things have to come first: the insurance estimate standard being written this quarter under Priority 1, and a logging habit that Kevin and Sam actually use. The CRM drop in 2024 showed that a tool requiring consistent data entry from that role will stall if the habit is not there. The AI move in that role comes after both are in place.
Luis's scheduling role has repeatable work that a drafting tool could help with, but the Loops part of this play (owned by Luis) is not started yet. Until there is a cadence for reviewing what the schedule produces, adding a tool to that role adds noise before there is a system to catch it. This is the third or fourth move, not the first.
A customer-facing chatbot or auto-responder for inbound inquiries is a real option at Caliche's volume, but it requires a written FAQ standard and a clear escalation path. Neither exists yet. It waits.
What to check first
- Hours in the By Role table. Every figure is a draft estimate. Dana, you are closest to the actual numbers across most roles. Correct any that are wrong before the owner sees this document.
- The collection escalation sequence (Move 2 standard). Does a written version exist anywhere, even partial? If you have a mental script you already use, that is the starting point. Two hours to write it down is the estimate.
- Data rules. The list of data that never goes into an outside tool is a draft. You and the owner should read it together and add anything specific to Caliche's insurance contracts or vendor agreements that belongs on that list.