Complete Refrigeration & Air
Process MapStage 2 · optimised
2026-05-05
Process Deep Dive · Quoting

How quoting should work at CRAWA.

The optimised quoting process, redesigned from the verified Stage 1 baseline. Four humans and 3-7 days collapse to one human approval gate and same-day turnaround. The CRAWA agent replaces all mechanical assembly work; structured field capture reduces rework; the office expert's diagnosis and judgement are preserved — because that's where the real value is.

Stages 1 · Map current state → 2 · Design optimised version

§1Who's involved

Three actors, down from five. The VA's mechanical work moves to the agent. The customer still exists but is downstream — not in the assembly chain.

Field tech

Structured capture

Same 19 tradesmen. Instead of typing free-text notes into AroFlo's field UI, the tech uses a purpose-built PWA — voice or typed. Prompted fields (diagnosis, asset, failure mode, photos) replace unstructured notes. Asset history auto-loaded so the tech can see prior work on this unit.

CRAWA Agent

Quote assembly + drafting

Deterministic automation. Receives structured capture, pulls asset history and prior work on this unit, looks up parts and supplier prices, applies customer rate card and markup rules, calculates labour from historical averages, drafts narrative, formats to CRAWA template. Produces a complete draft quote. Not diagnosing — that stays with the tech and office expert.

Office expert

Single approval gate

Same people — Dave Katsaros, Kurtis Dixon, Joel Baldini, Richard Barnes. Reviews the agent's draft in a single screen: tech's structured capture, asset history, complete quote with line items, margin summary, customer context. Validates diagnosis, adjusts scope/pricing if needed, approves.

Field tech CRAWA Agent Office expert Systems

§2Process flow — the contracted network

Compare to Stage 1's five rework loops and four handoffs. This flow has one human gate, one possible rework path (customer changes), and zero timezone gaps. Grey-struck nodes show what was removed.

flowchart TB classDef tech fill:#cffafe,stroke:#0e7490,color:#0e7490,stroke-width:1.5px classDef office fill:#ede9fe,stroke:#6d28d9,color:#6d28d9,stroke-width:1.5px classDef agent fill:#f3e8ff,stroke:#7c3aed,color:#7c3aed,stroke-width:1.5px classDef customer fill:#dcfce7,stroke:#166534,color:#166534,stroke-width:1.5px classDef system fill:#e2e8f0,stroke:#475569,color:#475569,stroke-width:1.5px classDef sched fill:#dbeafe,stroke:#1d4ed8,color:#1d4ed8,stroke-width:1.5px classDef decision fill:#fef3c7,stroke:#b45309,color:#78350f,stroke-width:1.5px classDef good fill:#dcfce7,stroke:#15803d,color:#15803d,stroke-width:2px classDef bad fill:#fee2e2,stroke:#b91c1c,color:#b91c1c,stroke-width:2px classDef neutral fill:#fef3c7,stroke:#b45309,color:#78350f,stroke-width:2px classDef removed fill:#f1f5f9,stroke:#cbd5e1,color:#94a3b8,stroke-width:1px,stroke-dasharray: 5 5 Start([Customer call/email]):::system S0[Office schedules
tech site visit]:::sched T1[Tech onsite — voice-first
structured capture]:::tech A1[Agent structures notes
+ pulls asset history
+ prior work on this unit]:::agent A2[Agent assembles quote:
parts · prices · rates · labour
· narrative · formatting]:::agent O1[Office expert reviews
single screen: capture +
asset history + quote + margin]:::office D1{Approve?}:::decision A3[Agent sends quote
+ updates AroFlo]:::agent C1[Customer reviews]:::customer D2{Decision?}:::decision A4[Agent re-drafts
per changes]:::agent S1[Agent creates
scheduled job]:::sched E1([Approved]):::good E2([Rejected]):::bad E3([Lapses — agent
auto-follows up]):::neutral X1[VA assembles
line items]:::removed X2[Office → VA
handoff]:::removed X3[VA → Office
review loop]:::removed Start --> S0 S0 --> T1 T1 --> A1 A1 --> A2 A2 --> O1 O1 --> D1 D1 -->|"adjust + approve"| A3 D1 -->|"needs field info
↺ rare"| T1 A3 --> C1 C1 --> D2 D2 -->|"accepts"| S1 S1 --> E1 D2 -->|"rejects"| E2 D2 -->|"silence"| E3 D2 -->|"wants changes"| A4 A4 --> O1
Rework loops eliminated or reduced:
· D1 unclear diagnosis → reduced, not eliminated. Structured capture + asset history means clearer data for the office expert, but diagnosis still depends on the tech's judgement onsite. The expert still validates.
· D2 missing info / unknown SKU — agent has full parts database and price lists; no VA guessing
· D3 quote needs revision — no VA draft to review; agent builds to spec from deterministic rules
· VA timezone gap — agent runs in seconds, not hours

One rework path remains: customer requests changes → agent re-drafts instantly → office expert re-approves. This loop takes minutes instead of days.

New capability: lapsed quotes get automated follow-up nudges instead of sitting in silence.

§3The flow, end to end

Read top to bottom. Four lanes instead of six. Steps highlighted with show where multiple Stage 1 steps have been merged.

Step · time
Field tech
CRAWA Agent
Office expert
Systems
1Capture5–10 min
Structured capture via PWA
Voice or typed — agent transcribes and structures into diagnosis, asset, failure mode, photos. Prompted for missing fields. Doesn't need to be in AroFlo's clunky field UI.
Structures + loads context
Transcribes voice, identifies asset from job context, auto-loads asset history and prior work on this unit, prompts for missing fields
AroFlo Task exists
Created at intake
2Contextseconds
Asset history + prior work lookup
Pulls every prior task, quote, and part replacement for this asset. Shows "last time this unit broke, we replaced X for $Y". Structured query — not AI guesswork. Diagnosis remains the tech's and expert's job.
3Assemblyseconds
Full quote assembly
Parts lookup from supplier price lists · customer rate card applied · CRAWA markup rules · labour from historical averages · narrative drafted · formatted to CRAWA template
Draft quote saved in AroFlo
Status: In Progress
4Approve2–5 min
Single-screen review + approve
Sees: tech's structured capture · asset history + prior work · complete line items · margin summary · customer context. Validates diagnosis, adjusts scope/pricing if needed, hits approve.
5Sendseconds
Sends quote to customer
Email via Outlook · updates AroFlo status to Pending Approval · logs send time for follow-up tracking
Quote → Pending Approval
6Customerdays–weeks
Auto follow-up on lapses
Scheduled nudge emails if no response within configurable window (e.g. 5 business days)
Customer accepts / rejects / lapses
On acceptance → agent auto-creates scheduled job
Total lead time: tech capture to customer receiving quote in under 1 hour (limited by office expert availability, not assembly). Compare to Stage 1's 3–7 business days. Customer still takes days–weeks to respond, but automated follow-up addresses the 54% lapse rate.

§4Step-by-step detail

Each step expanded. The "Replaces" column shows which Stage 1 steps were collapsed.

# Actor Action System Duration Replaces (Stage 1)
1 Field tech Structured capture: voice or typed, agent structures in real-time. Asset history auto-loaded. Prompted for missing fields. Replaces AroFlo's clunky field UI. PWA (offline-capable) · AroFlo API 5–10 min Steps 1–2 (assess + type notes). Eliminates delayed/unclear free-text entry and end-of-day note catch-up.
2 Agent Asset history lookup: every prior task, quote, and part replacement for this unit. "Last time: replaced compressor, $4,200, 6 months ago." Structured query, not AI inference. AroFlo API (tasks, quotes, assets) seconds Part of Step 3 (office expert reads + validates). Expert still owns the diagnosis — but now has full history at a glance instead of searching manually.
3 Agent Full quote assembly: parts lookup, supplier prices, customer rate card, CRAWA markup, labour estimate, narrative, formatting. Supplier price lists · AroFlo rates · CRAWA rules engine seconds Steps 4–7 (scope decision, hand to VA, VA lookup, VA assembly). Plus rework loops 6a and 8a.
4 Office expert Single-screen review: tech's structured capture, asset history, complete quote, margin summary, customer context. Validates diagnosis, adjusts scope/pricing, approves. Dashboard · AroFlo 2–5 min Steps 3, 4, 8, 9 (validate, scope, review VA draft, send). All in one sitting.
5 Agent Sends quote to customer via email. Updates AroFlo. Starts follow-up timer. Outlook API · AroFlo API seconds Step 9 (manual send). Now automatic on approval.
6 Agent Monitors response. Auto follow-up nudges on configurable schedule. On acceptance, creates scheduled job in AroFlo. Outlook · AroFlo · scheduling automated Steps 10–11 (customer reviews + outcome). Addresses the 54% lapse rate with proactive nudges.

§5Before and after

Side-by-side comparison of what changes.

Stage 1 · Today

  • 4 humans in the assembly chain (tech → office → VA → office)
  • 5 actors total including customer and systems
  • 11 steps plus 4 rework loops
  • 3–7 business days from site visit to customer receiving quote
  • 4 handoffs — each a point of information loss
  • Timezone gap between Perth office and Indonesian VA
  • 54% of quotes lapse with no follow-up
  • Free-text notes of variable quality drive rework
  • Shadow process on Teams / phone / WhatsApp / email — no audit trail

Stage 2 · Optimised

  • 1 human in the approval chain (office expert)
  • 3 actors — tech, agent, office expert
  • 6 steps, 1 rare rework path
  • Under 1 hour from capture to customer (limited by expert availability)
  • 1 handoff — agent to office expert
  • No timezone gap — agent runs 24/7
  • Automated follow-up on lapsed quotes
  • Structured capture — voice or typed, prompted, asset history auto-loaded
  • Full audit trail — every step logged in AroFlo
What's preserved: the office expert's diagnosis validation and judgement — the step that genuinely needs human experience (is the tech's diagnosis right? what should we actually quote on? customer relationship nuance). The agent handles everything downstream of that decision: parts, pricing, labour, narrative, formatting. That's where the time was being lost.

§6What happens to the VA role

Anggaa's mechanical assembly work is fully replaced by the agent. But his system knowledge and institutional context have value elsewhere.

Work that moves to the agent
  • Parts / SKU lookup from supplier catalogues
  • Supplier price checking
  • Customer rate card application
  • Markup calculation
  • Labour estimation
  • Narrative description drafting
  • Quote formatting to CRAWA template
Potential redeployment
  • Supplier relationship management — price negotiations, catalogue updates
  • Procurement optimisation — PO batching, supplier comparison
  • Data quality — cleaning AroFlo records, closing stale quotes, updating asset registers
  • Agent training data — validating diagnosis suggestions, flagging edge cases
  • Other queue work where system knowledge adds value

§7What needs to be built

The capabilities required to make this flow work, roughly in build order.

Component What it does Depends on Complexity
Supplier price index Structured, searchable index of parts and prices from CRAWA's key suppliers Supplier catalogues / portals Medium — data ingestion + maintenance
Customer rate card engine Programmatic lookup of per-customer pricing rules, contract rates, markup tiers AroFlo client data + rate cards Medium — rule extraction + encoding
Asset history service Given an asset ID, returns all prior tasks, quotes, parts used, costs, and outcomes. Structured queries against AroFlo — no AI inference needed. AroFlo API (tasks, quotes, assets) Low — straightforward API queries + caching
Field capture PWA Mobile interface for techs. Voice or typed input, structured prompts (diagnosis, asset, failure mode, photos). Offline-capable. Asset history auto-loaded on job open. AroFlo API · speech-to-text High — UX + offline + voice
Quote assembly engine Deterministic pipeline: structured capture → parts + pricing + labour + narrative → AroFlo quote Price index · rate cards · AroFlo API (write) Medium — rules + templates
Expert review dashboard Single-screen approval surface: capture, diagnosis, quote, margin, customer context, approve/adjust All above components Medium — UI integration
Auto-send + follow-up On approval: email via Outlook API, AroFlo status update, scheduled follow-up nudges Outlook integration · AroFlo API Low — straightforward automation
Auto-schedule on acceptance When customer accepts: create scheduled job in AroFlo, assign tech, notify office AroFlo API (write) · scheduling logic Low

§8Risks and open questions

Things that could complicate the optimised flow.

Risks
  • Supplier price freshness — if catalogues aren't updated regularly, the agent quotes stale prices. Need an update cadence or live API where available.
  • Edge-case quotes — unusual scope, custom fabrication, multi-trade coordination. May need a manual override path that bypasses the agent's assembly.
  • Tech adoption — voice capture requires buy-in from 19 field techs used to typing (or not typing). Needs to be genuinely easier, not just different.
  • Diagnosis still requires humans — the agent surfaces asset history, not diagnosis suggestions. The 15k historical tasks contain too much variable-quality free-text to train reliable symptom→diagnosis inference. This improves over time as structured capture replaces free-text, but for now, diagnosis stays with the tech and the office expert.
  • AroFlo write access — the current API service account has restricted permissions. Need full write access for quotes, tasks, schedules before any of this works.
Open questions
  • What % of quotes are truly routine (agent can handle end-to-end) vs. complex (needs manual scoping)?
  • Are supplier price lists available in structured format, or only as PDFs / portal lookups?
  • How many distinct customer rate cards exist, and how often do they change?
  • What's the minimum viable version? Could we start with agent-assisted (agent drafts, human always edits) before moving to agent-assembled (human only approves)?
  • Do any customers require specific quote formats that differ from CRAWA's standard template?
  • What's the approval for AroFlo write access — is it Richard's call or does it need IT/management sign-off?
CRAWA
Process map · Stage 2 · Quoting · Optimised redesign · 2026-05-05