Case study: 240-person logistics group, 11-month payback
A detailed account of a Jarvis deployment at a regional logistics operator — from the initial operating audit to the first measurable KPI lift.
Note: This case study is a composite based on patterns common across multiple SME deployments. Business names and identifying details are not disclosed. Financial figures represent outcomes from the described deployment configuration and operating context.
The First Question
The CFO's first question, in the first meeting, was: "When does this pay for itself?"
It's the right question. It's almost always the first question. And it's the question most AI vendors answer poorly — with references to "long-term transformation value," "productivity multipliers," and "strategic positioning," which are all real things but none of which answer the question.
"When does this pay for itself?" — The CFO, first meeting, before anyone had opened a presentation.
The answer, for this organization, was eleven months. Not a forecast. An outcome, measured against an operating baseline established before the deployment began.
The Organization
Sector: Regional logistics operator
Size: 240 employees across 3 facilities
Revenue: $28–32M annually
IT function: 2 people, systems administration alongside other duties
Primary challenge: Administrative overhead scaling with headcount, consuming margin
Regional logistics operator. 240 employees across three facilities — a main depot and two satellite locations — plus a dispatch team, a fleet management group, and a central operations function. About 85 office staff; the rest operations, warehouse, and driving. Revenues in the range of $28–32 million annually. Strong client retention, competitive pricing, good operational reputation. Not growing as fast as the principals wanted, partly because the administrative overhead of running the business had scaled with headcount in a way that was consuming margin without adding capability.
No dedicated IT function. Two people handling systems administration alongside other responsibilities. The technology stack was functional but fragmented: a transportation management system, a separate invoicing platform, shared drives for documentation, and a lot of email.
The Three Problems
Problem 1: Document Processing Overhead
Every day, the operations team processed approximately 180 to 220 documents: proof of delivery records, customs forms, carrier certificates, fuel receipts, maintenance logs. These documents arrived in mixed formats — scan, PDF, email attachment, occasionally fax — and needed to be reviewed, data extracted, matched against records in the TMS, and filed. Four people spent a combined 18-20 hours per day on this work.
The error rate was approximately 4% — meaning roughly 8 documents per day required manual correction after downstream systems flagged mismatches. Each correction took 15 to 45 minutes to resolve. The errors were not careless; they were the predictable result of high-volume repetitive work performed by humans under time pressure.
Problem 2: Client Communication Lag
Standard client inquiries — shipment status, document retrieval, invoice queries, service confirmations — took an average of 4.2 hours to receive a response during business hours. The volume was manageable but not trivial: approximately 60 to 80 incoming inquiries per day. The delay wasn't indifference; it was priority sequencing. The same people handling document processing were also handling client communications.
Problem 3: Institutional Knowledge Fragmentation
The company had been operating for 22 years. Over that time, it had accumulated a substantial body of operational knowledge that lived primarily in the heads of three or four senior operations staff. When those people were away, or when new hires needed guidance, the answer was "ask Sharon." Sharon was a bottleneck. And institutional knowledge was at meaningful risk of departure when any of those senior staff eventually left.
The Decision
The evaluation process took approximately six weeks. The decision criteria were simple and stated explicitly:
- System must process documents on company infrastructure — customs and carrier documentation is commercially sensitive
- Implementation must be achievable without dedicated IT staff
- Payback period must be calculable and defensible before deployment, not theoretical
- System must be auditable — the operations team must be able to see what the AI is doing and correct it
Three vendors were evaluated. Two proposed cloud-based solutions that failed the first criterion. One proposed an on-premise deployment with defined scope. That was the vendor selected.
Deployment: Four Phases
Phase 1 — Data Inventory and Model Preparation (Weeks 1–3)
Before any hardware arrived, the team conducted a document inventory. Every document type that flowed through the operation was catalogued: what it contained, what data needed to be extracted, what format it arrived in, what downstream system it needed to feed. This step took longer than expected — partly because nobody had previously enumerated all 34 distinct document types in the system — and produced a structured specification that governed the rest of the deployment.
The knowledge base exercise ran concurrently. Senior operations staff spent two to three hours each being interviewed about their accumulated operational knowledge. The outputs — structured notes, decision trees, routing rules, client-specific handling requirements — were organized into a searchable repository.
Phase 2 — Document Processing Pipeline (Weeks 4–6)
The document processing pipeline went live in week four with a human review requirement on every output. By week six, accuracy on the highest-volume document types was above 97%. Human review was scaled back to exception-flagged documents only. The processing time per document dropped from an average of 4.1 minutes to under 20 seconds for documents the system handled without escalation. The 4% error rate dropped to approximately 0.8%.
Phase 3 — Client Communication System (Weeks 7–9)
The client inquiry system handled standard inquiries by querying the TMS and the document archive directly and generating a response from the actual data. Average response time dropped from 4.2 hours to 18 minutes — including both automated responses and the human-reviewed queue, which moved faster because the AI had already drafted the response and pulled the relevant information.
Phase 4 — Internal Knowledge System (Weeks 10–12)
The searchable knowledge base went live for the full operations team in week ten. Staff could query it in natural language: "What are the handling requirements for Meridian Foods shipments?" "What's our standard rate for lane 12 to 14 with temperature-controlled cargo?" Onboarding time for new operations staff dropped by approximately 40% in the following quarter.
Measured Outcomes at Month 11
Total annualized benefit attributable to the deployment, conservatively: approximately $430,000 in direct labour and error costs, plus client retention outcomes. Total deployment cost including hardware, implementation, and first year of support: $381,000. Payback period: eleven months.
What Didn't Go As Expected
Document inventory took twice as long. 34 document types included edge cases nobody had considered — hybrid digital-paper forms, bilingual carrier documents with inconsistent field naming. Extended Phase 1 by about ten days.
Adoption was slower than expected. Three experienced staff continued manually reviewing documents the system had already processed confidently. Fix: show them the accuracy data. Evidence changed behaviour where instruction didn't.
Two client accounts needed template customization. Generic response templates read as impersonal for specific accounts with high communication expectations. A two-day fix that underscored: the last 10% of quality often takes as long as the first 90%.
The Compound Effect at Month 18
At month eighteen, three things had happened that weren't in the original business case. The operations team had extended the system to handle carrier performance tracking, producing insights that informed a carrier rationalization decision — estimated $85,000 in annual savings. Two of the four document processors had moved into client account management roles where there was genuine demand for their attention. And a third operating site was brought into the system in two weeks rather than the twelve weeks the original deployment took. The institutional knowledge base transferred directly. The compounding had begun.
The first deployment takes twelve weeks and two false starts. The fourth deployment takes three weeks and runs reliably from day one.