Monitoring & Forecasting
AI Forecasting Assistant
Forecasts demand, revenue or volume with explained assumptions.
Monitoring & Forecasting
AI Anomaly Detection
Unusual patterns in your transactions, traffic or operations data flagged automatically, without you setting a threshold.
Some problems do not cross a threshold you thought to set: an odd spike in refunds, a quiet drop in traffic, an unusual run of orders. This build learns the normal pattern in your transactions, traffic or operations data and flags anything unusual automatically. Each alert shows the data behind it, so your team can judge it quickly.
- Unusual patterns flagged without setting thresholds in advance
- Each alert shows the data behind the anomaly
- Sensitivity tuned with you so alerts stay useful
$5,500
20–30 working days
Optional maintenance $275/mo
What it is
AI Anomaly Detection learns the normal pattern in your transactions, traffic or operations data, then flags anything unusual automatically, without you having to set a fixed threshold in advance. Each alert shows the data behind it, and connections to your data are read-only.
Why businesses buy it
Fixed rules only catch the problems you predicted. Fraud, outages, pricing errors and supply issues often show up first as patterns nobody wrote a rule for. Spotting them early saves money and reputation. This build watches your data continuously, learns what normal looks like and flags what is not, with the evidence attached. The written scope sets the sources, metrics, alerts and checks against your history before you order, so you know what is covered.
Problems this solves
Fixed thresholds miss problems that look normal on paper
The tool learns the normal pattern in your transactions, traffic or operations data. It flags anything unusual without you needing to set a fixed threshold in advance.
Unusual transactions are only spotted during month-end checks
Anomalies are flagged as the data comes in, with up to five alerts sent to email or Slack. Your team can look into them while the trail is fresh.
Teams ignore alerts they cannot see the reason for
Each alert shows the data behind it, so people can judge whether it matters. Connections are read-only, and results are checked against three months of history before launch.
A good fit if
Your normal range shifts by season, day of week or growth, so a fixed threshold would trigger too many false alerts.
You want unusual spikes, drops or patterns caught even when you did not think to define a rule for them.
You already keep data in tools such as Google Sheets, BigQuery, PostgreSQL or Google Analytics 4, and can grant read-only access.
You want each flagged anomaly explained in plain language, not just a red mark on a chart.
Not the right fit if
You already know your exact thresholds and just want alerts on them. See AI KPI Monitor, which watches defined KPIs and alerts on breaches you set.
You want future values projected, not past or current anomalies flagged. See AI Forecasting Assistant, which forecasts demand, revenue or volume.
You want ad-hoc answers to new questions rather than continuous background monitoring. See AI Data Analyst Agent.
You need messy or duplicate records fixed, not unusual patterns flagged. See AI Data Cleaning, which cleans and standardizes datasets.
How it works
- You connect up to 3 read-only data sources, such as transaction, traffic or operations data from BigQuery, PostgreSQL or Google Analytics 4.
- The agent learns the normal pattern for each metric, including seasonal and day-of-week variation, from 3 months of your historical data.
- It checks new data continuously against that learned pattern, rather than a fixed number you would otherwise have to set and maintain.
- When something falls outside the normal range, such as a transaction spike or a traffic drop, it is flagged with a plain-language explanation of what changed.
- Flagged anomalies are delivered to the channel you choose, up to five active alerts at a time.
- You review the first weeks of flagged anomalies with us and tune sensitivity before the 30-day support window closes.
What you provide within 72 hours
Your portal lists these the moment you order. Delivery starts the working day after everything is in and your access is checked.
- Read-only access to each data source the anomaly detection should watch.
- Your own definitions of the metrics to monitor, such as transaction volume, traffic or a specific operations count.
- At least 3 months of historical data, so the normal pattern can be learned before launch.
- The list of recipients and the channel you want flagged anomalies delivered to.
- Admin access and API keys for every connected system, shared by adding our team as a user or creating a key you control, never by email.
- Your business information (company basics, hours and brand voice), shared during intake.
- One decision-maker who can confirm sensitivity settings before launch.
- An API key for your own AI model account (for example OpenAI or Google Gemini), created for this build with a spending limit.
The 72-hour rule. Send everything on your checklist, including access, within 72 hours of ordering so your build starts on time.
Timeline
Start
Intake confirmed and the metrics to monitor agreed.
Weeks 1–2
Data sources connected and the normal pattern learned from your history.
Week 3
Alert delivery and sensitivity thresholds configured.
Weeks 3–4
Detection tested against 3 months of your historical data.
Week 4
Your review and two rounds of changes to sensitivity.
Go-live
Live, handover report sent, 30 days of support starts.
| Option | Delivery time | Price |
|---|---|---|
| Standard | 20–30 working days | $5,500 |
| Express | 15–25 working days | $6,450 |
| Rush | 10–20 working days | $7,450 |
What you receive at handover
- The AI anomaly detection set-up, running on your chosen data and alerting by email, Slack or dashboard
- Read-only connections to up to three data sources
- Defined metrics, one dashboard or recurring report and up to five alerts
- A test report checking detection against three months of your historical data
- An admin guide and the written handover report
- 30 days of post-launch support
Expected benefits and how to measure them
- Can surface odd patterns that fixed rules would miss
- Helps teams investigate issues sooner
- Makes each alert easier to judge with the data attached
- Time to detect an anomaly
- Compare when past anomalies were found with when the tool flags new ones.
- Confirmed versus false alerts
- Have reviewers label each alert as real or not, and track the balance monthly.
- Issues found before month end
- Count issues found by alert rather than month-end checks, from your review notes.
Before you start: List recent anomalies, when they happened and when they were found. Note which data sources they appeared in.
These are things to track, not promised results. Outcomes depend on your data, your team and your market.
Running costs
The build itself is a one-time payment, shown before you buy. Running the detection costs roughly $20 to $150 a month in your own accounts, covering AI model usage and hosting or scheduler fees, depending on how many metrics are monitored and how often data updates.
AI usage during the build and testing is billed to your own accounts too, so the provider relationship stays in your name.
Ask about running costsCustom requirements
If you need more than 3 data sources or more than five active alerts, add a note at checkout for a written Change Quote. Your order still runs on its original delivery date. Larger or bespoke projects go through the Custom AI request form instead.
Request a custom versionKeep it running
We recommend monthly maintenance for this build. It covers monitoring, fixes and monthly change hours for sensitivity tuning or a new metric to watch. The allowance scales with the build, and bigger changes go through a Change Quote. It starts after the included support period and can be cancelled any month.
$275/mo (5% of the build price a month, includes 2 change hours)
Optional. Starts after the post-launch support period. Cancel any month.
How maintenance worksHow this compares
| Do it yourself with a SaaS tool | Hire a custom agency | Upmark Labs | |
|---|---|---|---|
| Price | A monitoring platform's anomaly feature is bundled into an expensive top tier | An agency quotes per statistical modeling project | Ours is one fixed price, shown before you buy |
| Delivery | An anomaly feature is instant to enable but noisy until manually tuned | An agency model build usually takes several weeks | Ours has a fixed delivery time shown before you buy, with faster options |
| Thresholds | A monitoring platform still needs you to set a baseline threshold | An agency build depends on what was agreed | Ours learns the normal pattern from 3 months of your own data |
| Explanation | A monitoring platform flags a spike with no context | An agency build depends on what was agreed | Ours explains what changed and by how much, in plain language |
Questions about this product
What does the AI Anomaly Detection price include?
One fixed price, agreed before work starts, covers up to three read-only data sources, up to 25 defined metrics, up to five active alerts on one delivery channel and checks against three months of history. Faster delivery options are shown in the price panel before you order, and there are no other build fees.
How is AI Anomaly Detection different from setting a threshold myself?
It learns your normal pattern, including seasonal and day-of-week variation, from 3 months of your historical data, rather than relying on one fixed number you have to set and keep adjusting. That means fewer false alerts when your normal range naturally shifts.
How long does AI Anomaly Detection take to launch?
The standard, Express and Rush delivery times are all shown in the price panel before you order, and the clock starts once your intake is complete. Delays only happen when data access or metric definitions arrive late, which moves delivery by three working days per day of delay.
What kind of anomalies does AI Anomaly Detection catch?
Spikes, drops or unusual patterns in the metrics you connect, such as transaction volume, website traffic or an operations count. Each flagged event comes with a plain-language explanation of what changed and by how much, rather than a bare alert.
How much does AI Anomaly Detection cost to run each month?
Roughly $20 to $150 a month in your own accounts, covering AI model usage and hosting or scheduler fees. The figure depends on how many metrics are monitored and how often data updates. You pay these providers directly, with no markup from us.
Can AI Anomaly Detection watch more metrics later?
Yes, as an add-on. A metric count above 25, or a fourth data source, is quoted through a written Change Quote once you send your request. Adjusting sensitivity on existing metrics fits inside the monthly change hours that come with maintenance.
Who owns the anomaly patterns AI Anomaly Detection learns?
You do. The learned patterns and every flagged event run against your own data, in your own accounts, not inside a shared vendor model. If maintenance lapses, detection keeps running; you lose the monthly monitoring and included change hours, not the build itself.
Catch the pattern you did not think to set a rule for.
Order AI Anomaly Detection and your data is watched for unusual patterns automatically, with the evidence attached. The scope above is what you get, in writing, before you pay.
Works well with
