Energy software for assets nobody can go and look at
Your kit is spread over hundreds of miles, behind locked gates, on a link that fails for days. Every question costs a van trip unless the software answers it first.
What makes energy assets hard
5 things that decide this
- 01A truck roll is the unit of waste. Every question your software cannot answer becomes a person driving to a site to look at a gauge, and half of those trips find nothing wrong.
- 02Design as though you will never touch the device again. Remote diagnostics, watchdogs and over-the-air updates decide whether a fleet is fixable or whether each fault costs a visit.
- 03Sensors drift, so a reading is evidence rather than truth. Keep raw values apart from derived ones and hold a calibration record per device, or a slow error becomes a business mystery.
- 04A gap in the data is not a reading of zero. Conflate them and you either hide a genuine fault or raise an alarm every time a site loses signal.
- 05Load is seasonal, and the worst day is the one where everything peaks together. Test against the winter peak rather than the annual average.
Distance turns every small question into a cost
In most trades a question is cheap. Someone walks over and looks. Here the same question means a van, a driver, a day, and often a locked gate at the end of it.
That changes what software is for. A dashboard is not the point. The trip it saves is. Judge a build by how many visits it removed, and how many of the rest arrived with the right part on the van.
TankAware works this way for a Canadian site operator. Sensor data streams from remote tanks on AWS. Alerts flag an odd reading before it turns into a failure. Inspections are captured on a phone that expects no signal. Sutherland Excavating Ltd. reported 40% fewer manual errors and 30% less overhead after launch.
- 01Count the visits that found nothing. That number is your business case.
- 02Send the diagnosis with the dispatch, so the technician brings the right part.
- 03Buffer at the gateway for days of outage, not minutes, and backfill with the original timestamps.
Relevant work
40%
fewer manual errors, reported by Sutherland Excavating Ltd.
30%
less operational overhead, reported after TankAware launched
20%
faster maintenance turnaround, reported on TankAware
5
permission tiers with per-organisation isolation
- SenseRaw value, device, time taken.
- HoldStore and forward through the outage.
- BackfillOriginal times survive the reconnect.
- CorrectCalibration reinterprets, never rewrites.
- TriageOne condition, one alert, ranked.
- DispatchWith the diagnosis attached.
A project that skips step five produces an alert stream operators stop reading within a month. After that the system is technically live and practically switched off.
Where we are useful
Remote asset telemetry
IoT pipelines on AWS with predictive alerting, built on TankAware for sites where the link drops for days. Raw and derived values stored separately so history stays trustworthy.
Field capture without signal
Scheduled inspections captured on React Native with photos, a named signer and audit logs. The time of work is recorded at the point of work.
Contractor coordination
A deficiency becomes a work order, reaches an external contractor through their own portal, returns with evidence and ends in a payment. Shipped on TankAware.
Multi-tenant access
Five permission tiers with each organisation isolated under its own subdomain. Operators, site managers and contractors need different views of the same asset.
Most of this is not AI, and saying so helps you
Vendors in this sector attach AI to work that is arithmetic. Predicting consumption from temperature history is regression with decades behind it. Scheduling under constraints is operations research. Both are useful and neither needs a model.
Machine learning earns its place in two spots. Separating causes in noisy data, and triaging alerts so the ones that reach a person are worth reading. That is a narrower claim than the market makes. It is also the one that survives your engineers.
Ask any vendor which part of their product is a model and which is a rule. The answer tells you what you are buying and what you will be maintaining.
- A forecast from weather and past usage is arithmetic. Call it that.
- Alert triage on top of a certified detection method is honest. Replacing the method is not.
- Every model needs a plan for the day the data shifts under it.
Systems running unattended across sites
“TankAware has revolutionized how we manage petroleum sites. The real-time data and automation have exceeded expectations.”
Blake Sutherland · President, Sutherland Excavating Ltd.
The stack behind TankAware
Built with
Field and device work
Security
Questions buyers ask us first
01How do you monitor sites with no reliable connection?
Buffer at the gateway for days of outage and backfill on reconnect with the original timestamps. Mark backfilled data as such, so nobody mistakes a late batch for a burst of activity. The rule underneath is that absence is its own state, distinct from a reading of zero and from a device reporting itself unavailable.
02Can software cut the number of site visits?
That is the main reason to buy it, and the number to track is visits that found nothing wrong. Remote readings answer the routine questions, and a triaged alert with a diagnosis attached means the technician who does travel brings the right part. Sutherland Excavating Ltd. reported 20% faster maintenance turnaround on TankAware.
03Why do our alarms get ignored?
Because most of them have been wrong, and people learn that fast. Never put a raw device alarm in front of an operator. Deduplicate so one condition raises one alert, suppress patterns that are usually benign, and report how often an alert was right as a monitored number. Treat a drop in that number as an incident.
04Do you integrate with the systems we already run?
We build the operational layer around your system of record rather than replacing it, and yours wins any conflict about a value. Scheduled file exchange is still normal in this sector and is not a warning sign. Name the systems early, because their access model usually sets the schedule.
05Is AI actually useful for energy operations?
In two narrow places, and it is oversold everywhere else. Separating causes in noisy readings is a real statistical problem, and alert triage on top of a proven detection method is genuine value. Forecasting demand from temperature and history is arithmetic, and calling that AI weakens a pitch rather than strengthening it.
06How do you scope an energy build?
Scoping calls are free. Where the work starts inside your existing codebase, there is a paid two-week diagnostic ending in a fixed price rather than a range. We will not put a schedule on it before we know how many device generations are in your estate.
