Opportunities

Meridian Field Services — scored against evidence, with inspectable ROI

Steady-state net
$134,450
per year across the portfolio, after run costs
Year one, net
-$509,200
after implementation — negative is expected
Hours returned
6,771 h
per year at steady-state adoption
Evidence cited
26 chunks
across 7 opportunities

How to read this page

3 sourcesGrounded in evidence
Click any score to read its rationale, then a source chip to see the exact text it came from.
Assumption
The model inferred this rather than finding it. Never presented as fact, and always editable.
26Proposed by the model
Hours, headcount, rates and durations. The model never states a dollar amount — it has no field for one.
$45,543Computed by code
Every currency figure comes from a deterministic TypeScript function, recalculated as you edit.

Real: the scores, rationales, and coverage gaps below are the recorded output of an actual claude-opus-5 run over the evidence; every currency figure is computed live in TypeScript. Seeded: Meridian Field Services is a fictional company — its documents, people, and numbers are invented, and internally consistent. It is not a real customer.

01Dispatch / work order intake

AI intake agent: turn service@ email into ready-to-approve ServiceTitan work orders

An extraction agent reads the shared mailbox, classifies each message (new request / reschedule / status / invoice query), pulls the SOP-required fields (contact, fault description, access instructions, priority), resolves the property ID by fuzzy-matching the free-text address against the ServiceTitan property master, and creates a draft work order for one-click dispatcher confirmation. Unmatched addresses and low-confidence extractions are queued as exceptions rather than guessed. Monday-morning weekend backlog is pre-processed overnight so dispatchers start the week triaging, not typing.

7.6
weighted score

Current state

Dispatch re-keys every request by hand into ServiceTitan at 8-12 minutes per work order on 180-220 work orders a week, with property ID lookup from inconsistently formatted addresses named as the worst step. On Mondays six to eight dispatchers do nothing but intake until 11:00 against ~200 weekend emails, so Saturday requests are effectively two days late before a technician is assigned.

Scoring

8 scored · 2 assumptions

Inputs — proposed by the model

$
%
%
months
$
$

Outputs — computed, never model-authored

Steady-state net savings
$45,543
per year, after run cost, once adoption settles
Year one, net
-$50,335
after build + run cost
Payback
26 mo
cumulative net turns positive
Hours returned
1,856.4 h
per year at steady state
Three-year net
$40,751
36-month cumulative
Cumulative net position, months 1–12

1 of 8 inputs are assumptions. Weakest input: Hours per person / week at 70% confidence.

02Billing / contract administration

Contract-terms extraction so the PO number is on the work order at intake, not chased after invoicing

Run an extraction pass over the client contract set to pull billing terms — PO required yes/no, PO issuance rule, billing entity, property mapping — into a structured terms table surfaced in ServiceTitan at work order creation. Intake then blocks or flags any work order for a PO-required contract that has no valid PO, and an agent drafts the PO request to the client contact or account manager automatically instead of dispatch guessing or waiting on an AM email.

7.5
weighted score

Current state

About 40% of contracts require a PO on the invoice, but nobody in dispatch holds the contracts, so they guess or email the account manager and wait — which the team names as the reason invoices bounce. Missing or incorrect PO terms account for 244 of 611 Q2 invoice disputes, average 11.4 days to resolve, and one REIT has rejected invoices for a missing PO four months running with roughly $62k outstanding across nine invoices.

Scoring

8 scored · 3 assumptions
03Client service / dispatch

Self-serve work order status (client portal link + SMS/voice deflection) to kill "where is my technician" calls

Publish a per-work-order status link (no login friction: tokenised link sent on creation and on every schedule change) showing stage, assigned window and ETA, backed by an AI responder that answers status questions arriving by email or the dispatch line and hands off to a human when the question is not status. Pair it with proactive push when a window slips, so the client is told before they call.

7.4
weighted score

Current state

"Status of my request" is 1,247 of 4,118 Q2 contacts at a median 2.4 contacts per work order and 6.2 minutes handle time, and 94% are a dispatcher reading status out of ServiceTitan because there is no client-facing status view. 80% of contacts arrive by phone, peaking Monday 08:00-11:00 at 3.1x the hourly mean. "No update / no communication" is the top escalation theme (31 of 87), and one $410k ARR account is at renewal risk over exactly this.

Scoring

8 scored · 2 assumptions
04Dispatch / approvals and controls

Digitise the >$2,500 approval loop with AI-drafted approval packets and risk-based auto-approval

Replace email-based approvals with a routed approval task that carries an AI-drafted packet — fault description, property and contract context, prior visits for the same fault, estimated value and margin, PO status — so the dispatch lead and AM can approve from a phone in seconds. Low-risk work orders that match historical always-approved patterns for that property and contract auto-approve with post-hoc audit, and the system records the approval rule actually in force rather than the one in the stale SOP.

6.8
weighted score

Current state

The SOP requires three email approvals for work orders above $2,500 before scheduling, but the VP-level step stopped happening months ago because it was too slow and the SOP was never updated — dispatch now tells anyone who asks "two approvals." Approvals sit on the critical path between intake and scheduling, against SLAs of 30 minutes for P1 and four business hours for P2, and are recorded as loose email attachments.

Scoring

8 scored · 2 assumptions
05Field completion / billing

Billing-readiness agent: completion-note capture on site and a daily exception queue instead of a weekly export

Two moves. On site: voice-to-structured-note capture in the ServiceTitan mobile app that turns a technician's spoken summary into the labour, parts and fault-resolution fields, with an AI completeness check that tells the technician what is missing before they leave. Behind it: replace the weekly billing export with a daily agent that scores every completed work order for invoice-readiness (notes present and specific enough, PO attached, correct property, labour hours plausible against the schedule) and routes named exceptions to the owning technician or AM the same day.

6.7
weighted score

Current state

Q1 closed with roughly $310k of completed work orders never invoiced, mostly missing completion notes or a PO, and only about half is ever recovered — the rest ages past the contract window and is written off. The SOP holds back work orders lacking completion notes pending technician follow-up, but billing only exports weekly, so defects surface days after the technician has left site. The team's own whiteboard marks "NOTES?" as where leakage starts and questions why the export is weekly.

Scoring

8 scored · 3 assumptions
06HR / field onboarding

New-technician onboarding copilot and mentor sign-off tracker for the first 30 days

A mobile assistant for new technicians that answers the questions they currently cannot get straight answers to — my schedule, who my mentor is, how to fill a completion note to standard, expense and timesheet steps — grounded in the onboarding checklist and each region's actual practice. Alongside it, an agent tracks checklist and mentor sign-off completion per hire per region, chases the missing sign-off before week-2 unsupervised assignment, and flags at-risk hires (no mentor contact, high held-back work orders, unanswered scheduling questions) to the regional supervisor.

6.4
weighted score

Current state

First-year technician attrition is 38%, and exit interviews repeatedly cite two causes: chaotic first two weeks and no straight answers about scheduling. The onboarding checklist exists but is not used consistently across regions, and an HR audit found roughly a third of new technicians in NE and Southwest taking unsupervised assignments in week 2 with no recorded mentor sign-off.

Scoring

8 scored
07Regional integration / scheduling

Northeast recurring-schedule bridge: automate the Friday spreadsheet and prove multi-site logic before migration

Two-part intervention. Immediately, an ingestion agent parses the Friday spreadsheet from the NE region, maps each line to properties, trades and recurring templates in ServiceTitan, and creates the recurring work orders with a conflict check that flags crew double-booking before commit. In parallel, replay historical NE multi-site schedules through the ServiceTitan recurring configuration to reproduce and fix the double-booking failures, giving the region an evidence-based go/no-go instead of a standoff.

5.4
weighted score

Current state

The NE region still runs recurring work on its prior scheduling tool and sends dispatch a spreadsheet every Friday that is re-entered by hand. The regional lead refuses to migrate until the multi-site contract logic works, citing a January attempt that double-booked four crews in a week; leadership calls it a configuration problem but nobody has tested it. Meanwhile NE work is invisible to any ServiceTitan-based status or billing-readiness process.

Scoring

8 scored · 1 assumption

What this analysis could not determine

Recorded gaps are a deliverable, not a failure — they are the next round of discovery. An analysis that reports no gaps is not more complete; it is less honest.

  • ServiceTitan integration reality: is the API tier available for work order writes, recurring-schedule creation and status reads, and what does 'partially configured' actually exclude? All five system-touching opportunities depend on this.
  • Current baseline for time from request arrival to technician assignment, split by priority and region — the SOP states 30-minute and 4-hour targets but no measured performance was supplied.
  • Where the client contracts physically live, in what format, and how many of the ~340 properties are covered by how many master agreements — this sizes the PO-terms extraction directly.
  • Full-year unbilled figure and recovery rate: only Q1 2026 (~$310k, ~half recovered) was given, so we cannot see seasonality or trend, or how much of the leakage is notes versus PO versus other defects.
  • Technician time actually spent on completion notes per job, and connectivity/device conditions in the field — needed before sizing the voice-capture side of the billing-readiness work.
  • Whether the NE legacy scheduler's data is exportable via API or only via the Friday spreadsheet, and the structure of the multi-site contracts that broke ServiceTitan recurring in January.
  • Volume and line count of the NE Friday spreadsheet, and how many dispatcher hours the re-entry consumes — the estimate used here is inferred, not evidenced.
  • Telephony platform and call recording/transcription capability: 80% of client contacts arrive by phone, so email-only automation caps the addressable volume.
  • Structured exit-interview data and cost per technician replacement (recruiting, onboarding, unproductive ramp) to convert the 38% attrition rate into a defensible business case.
  • Dispatch headcount allocation by region and how the 14 dispatchers split between intake, status calls and scheduling — hours-per-person splits across opportunities 1 and 2 risk double counting without this.
  • Client-side appetite and technical constraints for a status portal (single sign-on, tenant vs. property manager access), beyond the three renewal conversations mentioned.
  • PE sponsor requirements on spend-authorisation controls and audit evidence, which determine whether risk-based auto-approval above $2,500 is permissible at all.
  • Whether any repeat-visit root-cause data exists (12 escalations cited 'repeat visit for same fault') — potentially a first-time-fix opportunity we could not size from this evidence.