Find your biggest AI opportunities in under 30 minutes.Book a consultation
Back[ Coventa ]Case Study

[ SERVICE OPERATIONS ]LIVE

Hub for Technical Issue Resolution

A resolution assistant for technicians and operators. Hub summarises recent work on the asset, matches ticket properties and comments against the organisation's own troubleshooting guides — including the ones that only exist as PDFs — and watches ticket trends for the failure patterns a maintenance programme should have caught.

HUB // ISSUE RESOLUTIONOpen tickets · Line 3
Hub Resolve/Open tickets · Line 3
Streaming telemetryLast sync 12s ago

Monitored assets

TicketAssetFaultOpenedMatchTrend
TK-9921Press P-07Hydraulic pressure drop14 minGuide HG-22 · 0.94
TK-9918Conveyor C-12Belt tracking alarm41 minPrior fix TK-8830
TK-9915Robot R-03Joint 4 torque fault1h 05mGuide RB-07 · 0.88
TK-9910Press P-07Sensor S-4 intermittent2h 20mRepeat pattern · 3rd
TK-9904Chiller CH-1Temp drift +2°C3h 10mPrior fix TK-7712
TK-9899Conveyor C-12Motor overcurrent5hGuide CV-11 · 0.81

TK-9921 · recommended procedure

  1. 01Recent work: seal kit replaced on P-07 (TK-9702, 11 days ago).
  2. 02Matched HG-22 §3.4 — check accumulator pre-charge (spec 90 bar).
  3. 03TK-8830 comment: 'pre-charge was 60 bar, recharged, resolved'.
  4. 04Pattern: 3rd pressure fault on P-07 in 30 days → flagged to maintenance.
Open work order

// Problem

The Problem

When a machine goes down, the fastest route to a fix is usually something the organisation already knows: what was done to this asset last week, what fixed the same symptom on a sibling machine, and which page of which manual covers the error code. None of that is reachable in the moment. Work history lives in the maintenance system, prior fixes live in free-text ticket comments nobody indexes, and the troubleshooting guide is a two-hundred-page PDF on a shared drive. So the technician diagnoses from scratch, and the asset stays down while they do.

  • Recent work on the asset is invisible at triage, so technicians re-diagnose what a colleague just fixed.
  • The most valuable knowledge — what actually worked last time — is trapped in unstructured ticket comments.
  • Troubleshooting guides exist as PDFs, which means they exist but are not usable under time pressure.
  • Repeat-failure patterns only become visible in a quarterly review, long after the third identical ticket.

// Overview

Hub sits on the ticket queue and does the retrieval a technician would do if they had the time. On arrival, each ticket is enriched with a summary of work recently performed on the associated asset, so the operator starts from current state rather than from the symptom alone. Ticket properties and free-text comments are matched against the organisation's own troubleshooting documentation — unstructured formats included — and the relevant procedure is recommended rather than searched for. Historic comments on similar tickets are mined for what actually resolved the issue, which turns closed tickets into a working knowledge base. Running underneath, trend monitoring watches for clusters of similar tickets inside a rolling window and flags them into the preventative maintenance programme.

// AI System

Why AI

Every input here is unstructured and none of it is consistently written. Ticket comments are shorthand typed by a technician with oily hands; troubleshooting guides are scanned PDFs with tables and diagrams; the same fault is described five different ways by five people. Keyword search over that corpus returns nothing useful, which is why nobody uses it. Retrieval grounded in the ontology — the asset, its components, its ticket history — plus a model that can read the prose is what makes the archive answerable. The learning loop is the compounding part: every resolved ticket makes the next retrieval better.

// Specs

Specifications

CONTEXT
Recent work summary on the asset, surfaced at triage
RETRIEVAL
Ontology-driven over manuals and internal docs, PDFs included
FEEDBACK LOOP
Historic ticket comments mined for what resolved the issue
MONITORING
Rolling-window trend detection over ticket and error volume
AUTONOMY
Recommends procedures; the technician executes and closes

// Features

Features

  1. 01Comprehensive summary of recent work on the machine or asset, surfaced before diagnosis starts.
  2. 02Troubleshooting manuals and internal documentation recommended by relevance, not searched by keyword.
  3. 03Historic ticket comments turned into an answerable record of what actually worked.
  4. 04Ticket properties and comments matched against preset organisational troubleshooting guides.
  5. 05Similar-ticket clusters over 30- and 60-day windows flagged into the maintenance programme.
  6. 06Unstructured formats handled natively, so existing documentation is used as-is.

// Architecture

Architecture

RESOLUTION FLOW

Runtime · one item, left to right


  1. 01Ticket Intake
  2. 02Asset Context Assembly
  3. 03Grounded Retrieval + MatchingRecent WorkPrior CommentsTroubleshooting GuidesError Trends
  4. 04Recommended Procedure
  5. 05Technician Execution
  6. 06Comment Write-Back

dashed = the inference step, where the system exercises judgment

System stack

Data in · decisions out

01

Sources

Where knowledge actually lives

TicketingServiceNow / MaximoAsset registry & work ordersTroubleshooting guidesPDF, scannedHistoric ticket commentsMachine telemetry

02

Ingestion

Make unstructured usable

Document parsinglayout, tables, figuresComment miningwhat actually workedStreaming telemetryKafka · 1s windowsAsset linking

03

Ontology

Objects and their history

AssetTicketProcedure · StepPartFailure modeTechnician

04AI

Intelligence

Retrieve, match, watch

Grounded retrievalvector + BM25 · citationsProcedure matchingLLM · ticket ↔ guideWork summariserrecent work on the assetPattern detector30/60-day clusteringEval suiteresolution accuracy

05Human

Human control

The technician stays in charge

Technician executesSourced recommendationsThumbs-up / down on stepsMaintenance escalation

06

Actions

Written back

Ticket comment write-backMaintenance programme flagRetrieval corpus update

Observability

Every model call traced; evals run on real cases, not anecdotes.

Governance

Entitlements enforced at retrieval; rules versioned by the organisation.

Write-back

Systems of record are written only through the approval gate.

Closed tickets return to the retrieval corpus, so the next identical fault resolves faster than this one.

// Impact

Impact

Every ticket
Opens with asset context attacheddesign intent
30 / 60 day
Windows monitored for repeat-failure clustersdesign intent

Interested in Technical Issue Resolution?

Let's see what your closed tickets already know.

Get in touch
// End of case studyHub for Technical Issue Resolution