Automatic Anomaly Detection vs Manual Thresholds — Which Is Better for Your Spreadsheet?

Published March 14, 2026 · 8 min read · Guide

There are two ways to monitor spreadsheet data for problems. The first is to decide in advance what "wrong" looks like and set rules. The second is to let the software figure out what's unusual and tell you. This post gives you an honest comparison — including when manual rules still make sense.

The Case for Manual Thresholds

Manual threshold rules are simple and transparent. You say "alert me when accounts receivable aging exceeds 90 days" and that rule fires. You know exactly what you asked for, and you know exactly why the alert fired.

They work well when:

You have a specific regulatory or contractual limit (e.g. SLA response time under 4 hours)
The business rule is well-understood and stable (e.g. stock reorder point below 100 units)
You want auditability — someone asks "why did this alert fire?" and you can point to an exact rule

      The core problem with manual rules: they only catch what you anticipated. If you set a revenue alert at £40,000 but revenue has been gradually declining toward that number for six weeks, the alert fires once — the day it crosses £40,000 — not during the weeks of drift that could have been caught earlier.
    

What Manual Thresholds Can't Catch

Scenario 1: The Seasonal False Alarm

You set a sales alert for anything below £30,000 per day. Every Sunday, sales legitimately drop to £22,000 — it's a weekend. Your alert fires every Monday morning. Your team starts ignoring it. Then one Sunday, something is genuinely wrong: a payment processor outage drops sales to £8,000. Nobody notices because they've been trained to dismiss Monday alerts.

What automatic detection does: builds a separate baseline for Sundays. £22,000 on a Sunday is normal. £8,000 on a Sunday is a 3-sigma deviation from Sunday's own history. Emergency alert fires.

Scenario 2: The Correlated Failure

Your rule watches revenue. Revenue is normal today. But margin is slightly low AND order count is slightly high AND average order value is slightly low — all at the same time. Individually, none cross your thresholds. Together, they signal someone is offering unauthorised discounts to boost volume at the expense of profitability.

What automatic detection does: correlation analysis detects that three metrics deviated in the same direction simultaneously. That pattern scores high even though no individual metric crossed a manual limit.

Scenario 3: The Silent Sensor

Your IoT temperature sensor starts reporting exactly 21.4°C every hour. Your rule says "alert if temperature exceeds 80°C." It doesn't fire. But the sensor has failed — it's frozen on a cached value. Equipment overheating could go undetected for days.

What automatic detection does: stuck-value detection notices that 20 consecutive readings are identical when historical std-dev was 2.3°C. Emergency alert fires immediately.

Side-by-Side Comparison

Manual Threshold Rules

You must know what "wrong" looks like before you start
Seasonal patterns cause constant false alarms
Gradual trends invisible until you hit the limit
Multi-metric correlated failures missed
Sensor failures only caught if they breach your numeric limit
Requires reconfiguration as data patterns change
Setup time: 5–20 minutes per rule

Automatic ML Detection (ThresholdIQ)

No prior knowledge of your data required — upload and detect
Seasonal baselines built automatically — no false weekend alarms
Trend detection catches drift weeks before it becomes a crisis
Correlation analysis flags multi-metric failures
Stuck/zero detection catches sensor failures invisible to rules
Engine adapts as more data arrives — baselines update
Setup time: zero — click "Detect Anomalies"

When Automatic Detection Is Clearly Better

Automatic detection wins decisively in these situations:

You're new to the data. First time monitoring a dataset, a new KPI, or data from an acquired business. You don't know what thresholds are appropriate yet — automatic detection shows you what's unusual before you have to guess.
Seasonal or cyclical data. Anything with day-of-week, hour-of-day, month-of-year patterns. Sales, energy consumption, web traffic, staffing. Static thresholds produce constant noise; automatic detection understands the cycle.
Multi-metric monitoring. If you're watching 10+ columns, writing and maintaining 10+ rule sets is time-consuming and fragile. Automatic detection monitors all of them simultaneously with no configuration.
Operational data with failure modes. IoT sensors, data pipeline outputs, system metrics. Automatic detection catches stuck values, zeroes, and outlier patterns that rules-based monitoring misses.

When Manual Rules Are Still Useful

Honest answer: manual rules make sense when you have a specific contractual, regulatory, or operational limit that must not be crossed regardless of context — and where auditability matters more than sensitivity. Examples: GDPR breach notification timelines, credit limit exceedance, minimum cash balance covenants.

ThresholdIQ's automatic engine is not a replacement for those. It's a complement. The automatic engine finds what you didn't know to look for. Manual rules enforce the limits you already know matter.

The right long-term approach for most Finance and Operations teams is: start with automatic detection to understand your data, then layer in specific rules for your known hard limits once you know where to put them.

Getting Started

ThresholdIQ's automatic detection requires no configuration at all. Upload any structured file — Excel, CSV, JSON or XML — and click "Detect Anomalies." The engine:

Infers your time column, metrics and dimension groups automatically
Runs all 9 detection methods across every metric
Returns a severity-graded anomaly log in seconds
Shows you exactly which detection method flagged each anomaly in the Signals tab

Try Automatic Detection Free — 7-Day Unlimited Trial →

No credit card required. Your data never leaves your browser.