It starts with a battery.
Site 7 is a solar-hybrid mini-grid about four hours from your operations office. It was commissioned eighteen months ago, serves 340 connections, and has been performing well. But over the past three weeks, its lithium-ion battery bank has been quietly degrading — losing about 8% of usable capacity. Nobody notices. The operations manager responsible for that site is also responsible for 19 others. The degradation hides inside daily fluctuations of charge and discharge cycles, invisible in the spreadsheet she updates every Friday.
On Day 22, the battery fails. The genset kicks in at 100% runtime. Someone in the community sends a WhatsApp message. The technician gets dispatched, drives four hours, arrives without the right replacement parts because nobody knew exactly what had failed or how badly. He drives back. Procurement takes three days. He drives out again. Total elapsed time from first sign of degradation to resolution: 38 days. Total cost: somewhere between $3,200 and $7,800 when you add up the replacement battery, two round trips, excess diesel, and lost revenue.
This is not a hypothetical. This is Tuesday.
The Sector-Wide Warning Signs
Nigeria's mini-grid sector is at an inflection point. The Distributed Access through Renewable Energy Scale-up (DARES) programme, a $750 million initiative backed by the World Bank, aims to deliver clean power to 17.5 million Nigerians through solar mini-grids and standalone solar systems. More than 200 mini-grids were deployed in 2025 alone, and the pipeline is accelerating. Twenty-eight more were scheduled for completion in Q1 2026.
But deployment speed means nothing if the assets we are deploying cannot be sustained. NAPTIN (the National Power Training Institute of Nigeria) has warned that power sector reforms will stall without a skilled workforce capable of maintaining the infrastructure being built. The institute, in collaboration with JICA, has been developing training curricula for maintenance personnel — an acknowledgment that we are building faster than we can maintain.
And the evidence from the field confirms the concern. A Premium Times investigation that visited 40 mini-grid sites found systems "installed and commissioned without proper documentation — commissioning test results, as-built single line diagrams, key equipment manuals, maintenance spares, or operations and maintenance schedules." Dozens of government and donor-funded mini-grids were found operating far below capacity or lying abandoned entirely.
The pattern is consistent: we commission well, but we maintain poorly. And the cost of that gap is compounding.
What Reactive Maintenance Actually Costs
Let me walk through the economics that most fleet operators already feel but rarely quantify.
The multi-visit problem. In the current operating model, technician first-visit resolution sits around 40%. That means for every incident, there is a better-than-even chance your technician arrives, diagnoses the problem, and drives back for parts. At remote sites — and most mini-grid sites are remote by definition — each round trip costs $200-400 in fuel, vehicle wear, and staff time. Two to three visits per incident is the norm, not the exception.
The diesel trap. When a battery bank fails or an inverter goes down, the genset picks up the load. Diesel is already a dominant share of OPEX for hybrid systems — the SEforAll CAPEX/OPEX benchmark study found fuel can account for over 50% of ongoing costs in hybrid configurations. During an unplanned battery failure, diesel runtime goes to 100%. At current Nigerian diesel prices north of N1,800 per litre, an extended outage-response cycle can burn through $600-1,500 in excess fuel — compared to $100-200 if the disruption is caught early and managed intelligently.
The invisible revenue loss. Every hour a site runs on degraded capacity or is offline for repair is an hour of lost kWh delivered. For a site serving 300+ connections, that translates to $200-500 in lost tariff revenue per incident. Multiply that across a fleet of 20 or 50 sites, and the numbers become material to your P&L.
The compliance drag. For operators participating in results-based financing through NEP or DARES, performance data during outages must be manually compiled for IVA submissions and compliance reporting. When the data is fragmented across VRM exports, WhatsApp threads, and spreadsheets, compiling a single compliance report can take days of analyst time.
Add it all up, and a single undetected battery degradation event costs $3,200-7,800 by the time it is resolved. For a fleet of 20 sites experiencing even four such incidents per year, that is $256,000 to $624,000 in avoidable cost annually.
The Status Quo Tool Stack
Most operators I know are running some version of the same setup: Victron VRM (or FusionSolar, or SparkMeter Koios) for telemetry, WhatsApp for coordination, and Excel for tracking.
VRM is excellent hardware monitoring software. It was built for that purpose and it does it well. But it was not built to be a fleet operations platform. Its alert system will rate-limit emails if a site hovers near a warning threshold — a sensible anti-spam feature that also means critical alerts can go silent at exactly the wrong moment. Alerts arrive in your inbox alongside 47 other notifications, most of them false positives. The signal-to-noise ratio makes it nearly impossible to catch a slow degradation trend.
WhatsApp is fast, but it is not searchable, not structured, and not auditable. The diagnosis your technician texted about Site 7 last month? It is buried in a group chat between photos of someone's lunch and a forwarded news article.
Spreadsheets capture whatever someone remembers to enter. They do not capture what no one thought to look for.
The result: each failure produces zero structured training data. The lesson learned at Site 7 never transfers to Site 14 — even when the failure pattern is identical. The same $5,000 mistake repeats across your fleet because there is no system learning from it.
What Predictive Maintenance Makes Possible
Now consider the alternative.
Day 3: an anomaly detection system flags that Site 7's battery bank is showing an unusual capacity decline — not a threshold breach, but a trend that deviates from the battery's baseline behaviour. The pattern matches a known degradation signature.
Day 4: a work order is automatically generated. The system recommends specific replacement parts based on the diagnosed failure mode. The technician is dispatched with the right parts, the right tools, and a diagnosis already in hand.
Day 5: one visit. Three hours on site. Battery bank replaced. Site back to full solar operation.
Total cost: $800-2,200. Time from first anomaly to resolution: 8 days instead of 38. One technician visit instead of three. Minimal diesel burn. Full data capture for compliance reporting. And — critically — the failure pattern is now encoded in the system, so when Site 14's battery starts showing the same signature six months later, the detection happens on Day 1 instead of Day 3.
This is not science fiction. The U.S. Department of Energy has documented that functional predictive maintenance programmes deliver a 25-30% reduction in maintenance costs, 70-75% elimination of breakdowns, and 35-45% reduction in downtime. Companies like Omnidian are already monitoring over 1,500 MW of solar assets using machine learning to detect performance anomalies before they become failures.
The technology exists. The question is whether our sector will adopt it before the maintenance backlog becomes unmanageable.
The Learning Loop: Why Every Incident Should Make the Next One Cheaper
Here is what I think is the most underappreciated argument for predictive maintenance in the mini-grid context.
Nigeria is building hundreds of sites across dramatically different operating environments. A site in Borno faces different dust loading during harmattan than a site in Bayelsa dealing with humidity and salt air. Research on dust soiling in West Africa has shown that harmattan dust can reduce solar generation by more than 50% in some locations. Battery cycling patterns differ based on community load profiles. Inverter stress varies with grid interaction modes.
Every one of these sites is generating telemetry data. Every failure is a data point. But right now, that data evaporates. It sits in disconnected VRM instances, unstructured WhatsApp messages, and individual technicians' memories. When a technician leaves, his diagnostic intuition leaves with him.
A proper predictive maintenance system creates a learning loop: every incident feeds back into the model. Every resolved failure makes the next detection faster and the next dispatch more precise. Over time, the system learns your fleet — not just individual sites, but patterns across sites, across seasons, across equipment vintages.
This is especially powerful for the DARES pipeline. As hundreds of new sites come online, they will face the same failure modes that existing sites have already encountered. The operators who have been capturing and learning from their operational data will commission new sites with a maintenance playbook already written. The operators who have not will be starting from zero every time.
The Cost of Waiting
The Premium Times investigation found lead-acid batteries dying under Nigeria's heat, users disconnecting because of limited capacity, and communities losing trust in systems that were supposed to transform their access to electricity. Every abandoned mini-grid site is not just a financial loss — it is a setback for community trust that the next developer will have to overcome.
The SEforAll benchmark study recommended dedicating 20% of project budgets to maintenance. That is a start, but throwing more money at reactive maintenance is like buying a bigger bucket for a leaking roof. The question is not just how much we spend on maintenance, but whether we are spending it before the failure or after.
Reactive maintenance shortens asset lifespan by 30-40% and increases energy consumption by 15-20%. For a sector where the business case already depends on assets lasting 15-20 years, accelerating degradation through poor maintenance is an existential risk, not an operational inconvenience.
We are in a window right now where the fleet is growing fast but is still small enough to instrument properly. The operators who build predictive maintenance into their operating model today — who treat every site as a node in a learning network rather than an isolated installation — will be the ones still operating profitably in 2030. The ones who do not will be writing off assets at year five and wondering what went wrong.
The DARES programme represents the most ambitious bet Nigeria has ever placed on decentralized energy. The capital is flowing. The sites are being built. But capital expenditure without operational intelligence is just hardware waiting to become e-waste.
So here is my question for the mini-grid operators, developers, and funders reading this: if we know that reactive maintenance is the single biggest threat to mini-grid sustainability, why are we still treating predictive maintenance as a future investment rather than a present necessity?
I would genuinely like to hear how others in the sector are thinking about this.
Arome is a CTO based in Abuja building AI systems for the energy sector.