Process Improvement
Report Quality Drift Detection
12 engineers decliningAI monitors report quality scores across rolling 7/30/90-day windows. Detects gradual decline before it becomes a performance issue, and highlights engineers whose reports consistently provide actionable insights.
Quality Declining — Intervention Recommended
Report notes have shortened from avg 42 words to 8 words per field. Deviation explanations now single-word ("fine", "ok", "standard"). Decline correlates with schedule density increase on 3 July (moved from 4 to 6 jobs/day).
Suggested action: Reduce to 5 jobs/day for 2 weeks. Schedule coaching session on report value.
Photo documentation rate dropped from 94% to 61%. Deviation sections increasingly answered "no" without notes despite 3 confirmed callbacks in the period where deviations were later identified.
Suggested action: Ride-along review. Possible disconnect between field observations and reporting discipline.
Report entropy score has dropped sharply — identical phrase "Completed to manufacturer spec, no concerns" appears in 89% of notes across all job types over the past 12 working days.
Suggested action: Gentle prompt via agent — "Your notes help future engineers. What did you notice today that was different from the last time you did this job type?"
+ 9 more engineers below threshold (view all)
Actionable Insight Leaders — Top Contributors
Consistently notes root causes in deviation fields. 4 of his observations have been incorporated into updated procedure guidance this quarter. Notes average 67 words with specific part numbers and conditions.
Nominated for "Insight Contributor" badge. Reports used as training exemplars.
Identifies environmental factors (hard water, coastal corrosion, property age) that correlate with failure modes. Notes have driven 2 scheduling algorithm improvements for property-type-aware duration estimates.
Photographs non-obvious access issues and annotates with measurements. His documentation has reduced return-visit rates for Hive installations in new-build properties by identifying a recurring wiring configuration issue.
Fleet average report quality: 68/100. Top 5% of contributors generate 34% of all procedure improvement insights.
Travel Time Integrity Monitoring
142 routes analysed todayCompares actual travel time against routing engine estimates. Flags unexplained gaps between jobs and identifies routes where the engine systematically underestimates travel time.
Systematic Routing Underestimates
Routes where actual travel consistently exceeds estimate by >25%, across multiple engineers over 30+ days.
Estimate: 12 min. Actual avg: 26 min. Affected 847 journeys over 90 days. Cause: school zone speed restrictions not in routing data (Highdown School, 08:00–09:15 and 15:00–16:00).
Recommendation: Add time-of-day modifier for this segment. Estimated schedule improvement: 34 fewer breaches/week.
Estimate: 17 min. Actual avg: 26 min. Affected 312 journeys. Cause: single-lane roadworks (Thames Water) active since March 2026, no end date published.
Recommendation: Override route time until roadworks complete. Check utilities roadworks register monthly.
Estimate: 18 min. Actual avg: 25 min. Affected 156 journeys. Cause: new housing development access road creating unmarked junction queuing.
Unexplained Time Gaps — This Week
Journeys where actual elapsed time exceeds expected travel + reasonable margin by >15 minutes, with no traffic incident or depot stop recorded.
Total gaps flagged this week
Across 5,000 engineers, 24,840 daily journeys
Explained after review
Unlogged depot visits, customer no-access waits, fuel stops
Unexplained — requires investigation
No traffic, no depot, no logged reason. Avg gap: 28 min
Trend: unexplained gaps are 2.7% of total journeys (down from 3.4% last month). 14 engineers account for 41% of all unexplained gaps — flagged for manager attention.
Recommendations for Investigation
3 activeAI-generated recommendations based on cross-referencing engineer reports, parts data, callback rates, and job outcomes. Each recommendation includes the evidence chain and proposed action.
Part Number Confusion — Vaillant PCB Range
High impactRaised 3 days ago · 47 engineers affected · Estimated cost: £18,400/month in wasted visits
Evidence
- Part number mix-ups between VL-PCB-8735 (ecoTEC Pro) and VL-PCB-8730 (ecoTEC Plus) account for 67% of all PCB-related return visits across the fleet.
- 23 engineers have independently noted in report fields that "part numbers are hard to read on the box" or "print is smudged/faded."
- The confusion rate for Vaillant PCBs (4.2%) is 6x higher than the fleet average for part mix-ups (0.7%). No other vendor has a comparable error rate.
Recommended Actions
Inspect depot stock immediately — Verify label legibility on all VL-PCB-8735 and VL-PCB-8730 boxes currently in Reading, Oxford, and Bristol depots (estimated 340 units).
Raise with Vaillant UK — Formal quality notification regarding batch #VL-2026-Q2 label defect. Request replacement labels or corrected packaging for affected stock.
Interim mitigation — Deploy colour-coded stickers at depot level (green = 8735/Pro, blue = 8730/Plus) and enhance agent parts warning to show a visual comparison image during confirmation.
Worcester Bosch Greenstar 30i — Casing Screw Corrosion
Medium impactRaised 7 days ago · 312 engineers reporting · Step time overrun: avg +8 min
AI detected that "Remove boiler casing" on Greenstar 30i units aged 10+ years overruns by 8 minutes in 74% of cases. 312 engineers have independently noted corroded/seized screws. Recommendation: add penetrating oil spray to standard van stock and update procedure to include 2-minute pre-spray soak at step start. Projected saving: 1,240 engineer-hours/month.
Hive Install — New-Build Wiring Configuration
Medium impactRaised 12 days ago · 89 engineers affected · Callback rate in new-builds: 18% (vs. 4% in existing properties)
Hive installs in properties built 2022+ by Taylor Wimpey and Persimmon have a 4.5x callback rate. Root cause identified from engineer reports: S-Plan wiring with non-standard thermostat loop routing. Recommendation: add new-build S-Plan wiring check to Hive install procedure step 1 and create a 5-minute video training module.
Training Effectiveness Measurement
Last 90 days · controlled comparisonMeasures actual operational impact of training by comparing engineer performance metrics before and after attendance, controlled against non-attendees on equivalent job mixes.
Advanced Fault Diagnosis (March 2026)
412 engineers attended · 3-day classroom + practical
+8.3%
First-time-fix uplift
(vs. non-attendees, p<0.01)
-12 min
Avg diagnostic time
(complex repairs only)
-34%
Callback rate (28-day)
(from 14% to 9.2%)
£127k
Estimated monthly saving
(callbacks avoided × £120)
Recommendation: Expand to remaining 4,588 engineers. Priority: those with FTF rate below 88% on complex repairs.
Customer Interaction Excellence (April 2026)
286 engineers attended · 1-day workshop
+0.4%
CSAT improvement
(not statistically significant)
+1.2%
Upsell conversion
(not significant, p=0.34)
0
Safety impact
(no measurable change)
-£84k
Net cost
(training cost, no offset)
Recommendation: Do not expand. Review course content and delivery method. Consider replacing with AI-agent-delivered micro-coaching during upsell phase.
Gas Safety Regulatory Refresher (May 2026)
1,840 engineers attended · Mandatory annual · 1-day
-62%
Near-miss incidents
(30 days post-training)
+18s
Avg gate dwell time
(more thorough checks)
-4 min
Avg job completion
(confidence effect)
£0
Safety incidents
(zero since training)
On track. Remaining 3,160 engineers scheduled through Q3. Notable: gate dwell time increase fades after ~45 days — recommend AI agent reinforcement prompts at 6-week mark.