11%
of revenue: what the world’s 500 largest companies lose to unplanned downtime, about US$1.4 trillion a year.
Siemens, The True Cost of Downtime 2024
Industrial IoT · South Africa
A platform being built to bring readings from meters, machines and control systems at every site into one place, with alarms that reach a named person, energy priced at the tariff in force, and data faults shown as faults. Every part says whether it is live yet.
Industry data, not an Axiom result
11%
of revenue: what the world’s 500 largest companies lose to unplanned downtime, about US$1.4 trillion a year.
Siemens, The True Cost of Downtime 2024
27 h
a month of unplanned downtime at an average large plant.
Siemens, The True Cost of Downtime 2024
$125k
an hour: the typical cost of unplanned downtime. More than two thirds of industrial companies have an outage at least monthly.
ABB, Value of Reliability survey, 3,215 respondents, October 2023
5–20%
of a plant’s productive capacity lost to poor maintenance strategies.
Deloitte Insights, Making maintenance smarter, May 2017
Why now, in South Africa
Turn each card over for the numbers behind it. History is drawn solid; legislated or approved figures dashed; anything modelled is labelled as modelled, with its method.
What it does for a site
Each card carries its mark from the Architecture page. Only what is marked Live is in service today. Turn a card over to see what it looks like.
How it works
The edge or a gateway already on site sends readings to the site’s own connection.
Each series is named, given its unit and its expected interval, and armed.
Live tiles, alarms and data health for each site and across sites.
Figures leave as statements another party can check, each with its basis.
By industry
What the platform is being built to show in each industry. None of these views has yet been delivered to a client. Every picture on the back is a sample.
What the evidence says, and what it doesn’t
Predictive maintenance does not pay everywhere. McKinsey describes a programme whose false alarms wiped out its savings, and finds condition-based maintenance often gives good results at lower cost. Axiom starts with condition: measured run hours, limits and trends. It adds prediction only where a failure is costly and predictable, and says which is which.
Every figure on this page is another organisation’s published result. Axiom has not yet measured a result for a client. When it has, the figure will appear here with its own statement, which anyone can check.
McKinsey & Company, Establishing the right analytics-based maintenance strategy, 19 July 2021.
Site questionnaire
Contact details stay optional and are used only when you ask for a response.
How data moves
Follow one reading through the seven places it passes, or choose a station.
01 · Instrument · the meter on the main incomer
Your equipmentThe reading starts at an instrument the site already owns: here a class 0.5S energy meter on the main incomer, measuring apparent power through 800/5 current transformers.
What Axiom keeps here: which instrument, its accuracy class and its range, so every figure later carries the error it was measured with.
What goes wrong here. A drifting sensor reads wrong for weeks. The accuracy class and range travel with the figure, so the error stays visible.
Read only. The platform never writes to plant.
Read the IIoT pageThe reading at this point
meter M-01 · main incomer
S = 412.6 kVA
CT 800/5 · accuracy class 0.5S
Illustration. One demand reading from a fictional site, not a measurement from a plant and not a claim about speed. Field names follow the payload specification.
01
The platform observes plant. It never writes to a PLC, inverter, battery, setpoint or breaker.
02
Accepted or refused, every message gets a receipt and its bytes are kept exactly as they arrived.
03
A correction is a new entry beside the old one. Nothing sealed is edited or deleted.
04
Statements are hashed so anyone can check them in the browser without trusting Axiom.
The rule we build to
Every tile, chart and export carries the time the reading was taken and how much of the period was actually observed. When too little was observed, the number is withheld and the page says why. A gap stays a gap: nothing is filled in. Coverage grades are being built into every aggregate; the tiles below are samples of the design.
Main incomer demand
Sample412.6 kVA
14:30:00 · 1 min ago
Full. Every expected reading in the window arrived.
Cold room 2 temperature
Sample-18.4 °C
14:29:40 · 2 min ago
Partial. Some readings are missing. The figure is shown and says so; the gap is drawn as a gap.
Pump 3 run hours today
Sample— h
09:12:05 · 5 h ago
Below floor. Too little was observed, so the value is withheld rather than guessed.
Only 31% of today observed; the floor is 80%.
Tank 2 level
Sample6.42 m
14:30:00 · 1 min ago
Not gradeable. No expected interval is declared yet, so coverage cannot be judged.
live · every 60 s
Readings are arriving on their declared interval.
silent · 14 min
Overdue with no declared reason. This raises an alarm with an owner.
paused · until 30 Oct
Switched off on purpose, with a reopen date. No false alarm.
offline · since 06:10
The source cannot be reached at all. This raises an alarm.
Modules
A workspace gets the modules it needs and nothing else. Each one says where it stands today, using the same marks as the public Architecture page.
Connections
No rip and replace. A gateway already on site can send today; Axiom's own edge, specified, is to read the control systems a site already runs.
Beside your SCADA
Nothing here replaces the control system. Many SCADA packages can link sites too; the difference is the job, the reader and the record.
AI analyst
Specified, not yet built. Ask in plain language; the answer is to list the readings, schedules and documents it used, carry the coverage of the data behind it, and decline when the data has a gap, naming the gap. Costs are to be priced at the tariff in force when the energy was used, and nothing is to leave the workspace until a person has reviewed it.
Sample period: 6 October 2026, 07:30–08:05
Who reads what
Data from many sites, turned into information each reader can decide from, on the same record. Turn a card over to see what each reader sees.
Security and privacy
Nothing Axiom runs can write to a PLC, inverter or breaker.
Row-level security is switched on and forced on all 50 of 50 tables in the public schema (checked 8 October 2026). An independent test of it comes before a second client.
Required for owners, administrators and Axiom staff; a code before any password reset.
A Cloudflare Turnstile check on sign-in and on the public forms.
The database is in Frankfurt, in the European Union. Every provider outside South Africa is listed in a public transfer register, with the ground each transfer relies on, or a note where that ground is still being put in place.
Every connection to the platform uses TLS, and each token works for one connection only.
Every message is kept as it arrived. The interval record is append-only by design, with corrections added beside the original; it is being built.
AI reading of documents only where an owner has switched it on for the workspace.
Planned: an independent penetration test before a second client workspace goes live. None has been done yet.
Built in the open
The public Architecture page shows every part of the platform, each with its computed mark and the documents behind it, so a sceptical engineer can check what exists before believing what is promised.
See the Architecture pageCounted from the public Architecture page’s own data. 7 + 21 + 35 + 0 = 63 parts. As of 15 October 2026 · revision 17.
Start with one site
There is no sample workspace full of invented readings. A pilot connects one feed at one site, and you look at your own plant.
Tell us about the site in the questionnaire.
Choose one site, one feed and the question it should answer.
A gateway already there, or the edge, sends to the site’s own connection.
Each series named, given its unit and interval, and armed.
Your own readings, with their times and coverage, in your own workspace.
Illustration. No timing is promised.
Illustration of the pilot steps. No timing is promised here; the scoping call sets it.
Straight answers
No. Axiom reads from the control system you already run, from outside its network, and changes nothing in it.
No. There is no write path. Dispatch and demand advice is advice: a person on site decides and acts.
Its source shows Silent and an alarm goes to its owner. Figures over that period say how much was observed, or are withheld if too little was. A planned shutdown is declared as a pause, so it raises no false alarm. Alarms and coverage are being built: until they are live, this is the design, not the service.
Yes. A pilot connects one feed at one site into your own workspace. There is no sample workspace of invented readings.
Today, anything that can send JSON over HTTPS, including many gateways already installed. Axiom’s edge for OPC UA and Modbus, and an MQTT endpoint with a certificate per device, are specified and being built.
In Frankfurt, in the European Union. The public transfer register lists every provider outside South Africa, the ground each transfer relies on, and any ground still being put in place.
The AI analyst is specified to answer only from your records and to list them, and to decline across a gap in the data, naming the gap. It is not built yet.
Only on an issued measurement and verification statement. No tile, forecast or AI answer shows a saving on its own.
Sources