Demand & peaks
15-minute peak, contracted capacity and the risk of exceeding it, before it happens.
Lilo plant, main meter · kW · 15 minP90 1,270 kW · 15:00–16:00
- Actual
- Forecast P50
- Likely range P10–P90
- Contracted capacity 1,250 kW
Example
See where your energy is lost and what it costs.
Wattwise machine learning learns what normal consumption is for each site, and AI explains where and why energy is lost and what it costs in lari.
A pilot on one site in about 4 weeks.
- Energy
- Loss, as heat
- Meter readings
Most plants do not know when and where energy is lost.
7 October, 09:16. Losses suddenly went up.
This is how Wattwise works at one plant, in two cases: first a measurement error that would have spoiled the monthly balance, then a real loss the balance revealed. Both cases are examples.
Wattwise spots it
Loss over the last 15 minutes is 5-10% above the norm. The alert arrives within a minute in the platform and on Telegram, with the cost already in lari.
Lilo plant · High
Loss is 5-10% above the norm · +10.9 ₾ per hour
The AI drafts a report
When the episode ends, the AI writes a report with the fact, a similar past case and its own interpretation kept apart. It does not present an assumption as a fact.
INC-2026-041HighAwaiting reviewLilo plant · 07.10 · ±6 h
- Norm, by load
- Alarm ×2
FactFrom 09:00 to 12:00 the loss averaged 85.8 kWh/h, 5-10% above the norm.
Similar caseIn March: a blown voltage-circuit fuse on workshop A’s meter.
InterpretationThe incoming meter shows the usual load; the rise matches a drop in workshop C’s meter reading. A measurement problem is more likely than a real loss.
An engineer checks and closes it
The cause turned out to be measurement, not a real loss. The engineer approves the report and records the cause. Those three hours are flagged in the balance, so the month’s balance stays correct.
INC-2026-041Closed- Cause
- Workshop C’s meter had no voltage on one phase from 09:00 to 12:00 (a blown fuse) and under-read consumption
- Real loss
- None
- In the balance
- 09:00–12:00 flagged as a measurement error
- Approved by
- Chief engineer · 14:10
The cause is saved and used when analysing future incidents.
14 October, Lilo plant
And when the loss is real
Even after the error was fixed, the balance kept showing 121 kWh a day above the norm. Wattwise traced the excess loss to workshop B’s feeder: the compressor’s cosφ had dropped to 0.78, the current rose, and the cable and transformer lost more. After a capacitor bank was installed on the feeder, the loss went back to the norm.
INC-2026-049Closed
- Cause
- Workshop B compressor, cosφ 0.78: excess reactive current
- Action
- Capacitor bank on the feeder, cosφ 0.95
- Result
- Loss 841 → 720 kWh a day
- Saving
- 121 kWh and 28 ₾ a day
Over a year that is ≈ 10,200 ₾: exactly what the balance called recoverable.
Machine learning does the numbers. AI explains them.
Two tools, each with its own job. Machine learning learns from your data what is normal and what will happen. AI explains those numbers in plain language and never makes them up.
MLMachine learning
Numbers from your own data
- A norm for each site
- Normal consumption and loss are learned from the site’s own history, by hour, day of the week and load. An anomaly is a deviation from this norm, not one shared threshold.
- A forecast whose accuracy you can see
- Every forecast comes with a P10–P90 range, shown next to how it held up in past weeks. So you know how far you can trust it. Each week the model retrains on new data.
- Why it changedOn pilot data
- A change in the forecast is broken down into factors: temperature, production plan, calendar. You see not only the number but also its cause.
Forecast vs actual · Lilo plantLast 8 weeks · MWh
7 / 8
weeks the actual landed inside the forecast range
4.8%
average miss
- Range P10–P90
- Forecast P50
- Actual
- Outside the range
Why tomorrow’s forecast is higher than usual
On pilot data+3.1%- Air temperature+1.9%
- Production plan+1.6%
- Other−0.4%
Each forecast is made a week ahead, from the site’s own history. When the actual falls outside the range, you see the reason separately.
Example
AIArtificial intelligence
Text where every claim has a source
- A report on every incident
- Fact, similar case and interpretation kept apart, as you saw above. It becomes official only after an engineer approves it.
- A plain-language assistant
- Ask a question in your own words. The assistant finds the data itself and shows the source in its answer. The same assistant is on Telegram too.
- Learns from your investigations
- The cause of a closed incident is saved in the knowledge base. When something similar happens again, the report brings it up on its own and tells you what to check first.
Why did night-time consumption in Rustavi go up?
FactInterpretation
Average night load rose from 212 to 250 kW (+18%). Most of the increase comes from compressor #2.
Night load increase · kW · 14 days
Compressor #2+25
Lighting (warehouse)+6
Other+7
Source: hourly data · 2 tools
It is probably running idle at night. An on-site check is needed.
INC-2026-041Closed · 7 October
Workshop C’s meter had a blown voltage-circuit fuse.
Cause saved to the knowledge base
3 weeks later, the same symptom
INC-2026-058Report · 28 October
Similar caseINC-2026-041: same meter, same symptom. Check the voltage-circuit fuse first.
Example
Rules for the AI
- Read-only: it cannot change data or settings.
- An incident report becomes official only after a person approves it.
- Learns from your investigations: the root cause of a closed incident informs the next analysis.
- Sees only your company’s data.
Power, peaks and budget
Live power, daily consumption and losses for every site, in real time.
Wattwise / Whole company · 3 sites
07.10 · 14:32
Current power
1,748 kW
691 kvar · cosφ 0.93
Today
20,973 kWh
▲ 2.5% vs yesterday
Loss · last hour
61.2 kWh
Above norm +10%
Today’s active and reactive power
kW kvar Yesterday
Example
Forecast & budget
Month-end consumption and cost with a likely range (P10–P90), and the chance you will exceed the budget.
kWh · OctoberP50: +1.3% vs budget
69%
Chance that October ends over budget
- Likely range P10–P90
- Forecast P50
- Budget
P50 ≈ 1,084,000 · P10–P90 ≈ 1,049,000–1,119,000 · Budget ≈ 1,070,000 kWh
Example
Also: monthly PDF and Excel reports, data-quality checks, user roles and access control.
Finance wants lari. Engineers want cosφ.
Cost in lari, not kilowatts
Losses, anomalies and budget deviations are shown as financial impact.
A budget that drafts itself
The draft budget comes from the forecast; deviations split into volume and price effects.
A combined report for all sites
A monthly report with an AI summary, in PDF and Excel.
Decisions backed by data
Which site, which machine, and what a change is worth.
Budget247,621 ₾
Volume effect+3,219 ₾
Price effect0 ₾
Month-end forecast, P50250,840 ₾
Example
All sources, shared filters
SCADA, meters and files in a shared data model.
Anomalies the moment they happen
Alerts arrive straight away in the platform and on Telegram, not at month end.
Compare sites and equipment
Periods, working and non-working hours, kWh/m² and kWh per unit produced.
Answers in plain language
The AI assistant searches, compares and explains, with sources.
| Site | kW | Today kWh | Δ |
|---|---|---|---|
| Tbilisi, Lilo plant | 1,120 | 13,847 | ▲ 2.3% |
| Rustavi, plant | 503 | 5,669 | ▲ 4.2% |
| Batumi, warehouse | 125 | 1,457 | ▼ 1.8% |
Filter: whole company · today · all sources
Example
Degradation before failure
A condition score: falling efficiency, rising base load, changes in the operating cycle.
Hourly cosφ and reactive power
Reactive balance and its effect on technical losses.
Transformer headroom
15-minute peak from 1-minute data, loading in kVA and an overload forecast.
Meter fault detection
“Frozen” values, balance residuals and missing data are shown separately, as a data-quality signal.
Workshop B · compressor
61 / 100
05075100
cosφ0.86 → 0.78
Phase imbalance2 % → 9 %
Example
Your systems stay under your control
Nothing to install. Your system sends the data itself; nothing is opened in your network and nothing is written to SCADA.
Your plant
- SCADA and Historian
- Smart meters
- CSV and XLSX files
- REST API
By agreement: OPC UA, Modbus
Outbound only, encrypted HTTPS
Wattwise
- Each company’s data is isolated from the others at database level
- Every change is logged: who, what and when
- Backups; restores are tested regularly
Never writes to SCADA and never controls equipment
Browser and Telegram
Your team
- Access roles: administrator, energy manager, viewer
- Each user sees only their own sites
Alerts in the platform and in Telegram
If SCADA sits in an isolated network, the data is sent by the Historian or by a server in the DMZ. If that is not possible either, we start with files.
Start with one site
We prove the result on one site first. Then you decide whether to expand.
Week 1
Discovery
We learn about your site, meters and data sources.
Week 2
Connection
SCADA or files are connected; tags are mapped to equipment.
Week 3
Balance & thresholds
Our power engineer sets up the balance and normal loss with you.
Week 4
First report
You receive the first report: losses, anomalies and recommendations.
First report · Lilo plant
- Loss
- 841 kWh a day
- Above norm
- 121 kWh a day, ≈ 10,200 ₾ a year
- Anomalies
- 1, a measurement error
- Recommendation
- Reactive power compensation on feeder B
Timings are indicative and depend on the data source.
Frequently asked questions
Didn’t find an answer? Write to usGetting started and price
What equipment or software do we need to install?
Usually nothing. Your SCADA, Historian or metering system sends the data over HTTPS. If that is not possible, we start with CSV or XLSX files.
Which systems and manufacturers does it work with?
Wattwise is not tied to any one manufacturer. We take data in a standard format (JSON or CSV), so any system that can export data can be connected. OPC UA and Modbus are added by agreement.
What data do we need to start?
Readings from the main and workshop meters, ideally hourly or more often, and the history of recent months if you have it. The balance can be calculated from day one; the forecast gets more accurate the more history there is. We confirm the tariff and the site diagram together in the first week of the pilot.
How soon will we see the first results?
A pilot on one site takes about 4 weeks. In week 4 you receive the first report: losses, anomalies and recommendations. The timing depends on the data source.
How much does it cost?
Pricing depends on the number of sites and data channels. Pilot terms are agreed individually. Write to us and we will send an estimate.
Data and accuracy
How accurate is the loss calculation?
As accurate as your meters. The balance uses formulas checked by a power engineer. When the numbers do not add up, the system flags a meter or ratio problem instead of reporting it as a loss.
What happens if data stops coming in or a meter gives wrong readings?
Wattwise flags it as a data-quality problem: a missing period, a “frozen” value or an unusual balance residual is shown separately, and you get an alert. Such a period is not counted as loss.
Where is our data stored?
On a secured server operated only by Wattwise. Data travels over encrypted channels, each company’s data is isolated from the others, and it is backed up regularly.
AI and machine learning
How long does it take the system to learn our site’s norm?
In the first 4 weeks of the pilot, our power engineer sets up the balance and normal loss together with you, so the first results show within that time. The forecast gets more accurate as history builds up: seeing seasonal changes takes several months of data.
Can the AI change anything in our system?
No. The AI only reads your data and offers an explanation or a recommendation. It cannot change settings, data or equipment, and an incident report becomes official only after a person approves it.
Are other companies’ models trained on our data?
No. The machine learning models are trained separately for each company, only on its own data. Your data is given to the AI assistant only to prepare the answer to a specific question, and it is not used to train general AI models.
In 30 minutes we’ll show you where your energy may be going.
Write to us. In a 30-minute call we will show you the platform and assess what data a pilot could start with. You don’t need any plant data for the meeting.
info@wattwise.geTbilisi, Georgia