TapCub is live — analytics, insights and live chat, free on one platform
Glossary

What is ad-hoc analysis? Definition, example and reuse rules

Ad hoc analysis is a one-off, question-driven investigation of data that no existing report answers. Definition, a three-way comparison, a recomputable channel example, and rules for turning a query into a scheduled report.

Su He Reviewed by Lin Zhou Also called ad hoc reporting Last updated Nov 16, 2026 Glossary

Quick answer

Ad hoc analysis is a one-off, question-driven investigation of data that answers a specific business question no existing report covers. The expensive part is starting from zero every time; good ad hoc work leaves a reusable metric or segment definition behind.

What is ad hoc analysis?

Ad hoc analysis is a one-off investigation of data to answer a specific question that no existing report covers. TechTarget defines it as a business-intelligence process designed to answer a specific business question; Amplitude's glossary calls it a spontaneous, one-time investigation into data. The output is an answer — a chart, a table, a sentence — not a permanent dashboard. Ad hoc reporting is the output side of the same activity: the one-off report built to answer the question. The two terms overlap in practice; this entry uses ad hoc analysis for the investigation and ad hoc report for its artifact.

The term comes from Latin ad hoc, "for this." Cambridge Dictionary defines ad hoc as "made or happening only for a particular purpose or need, not planned before it happens." An ad hoc query is written for this question, now: it is not scheduled, and nobody planned it last quarter.

Three criteria separate ad hoc work from its neighbors:

  • One specific question. "Why did signups drop last week?" — not "how is the business doing?"
  • No pre-built report covers it. If the weekly dashboard already answers it, you are reading a report, not doing ad hoc analysis.
  • The output is an answer. It may be reused later, but it was not built as a permanent asset.

The cost of ad hoc analysis is not running the query — it is starting from zero every time. Good ad hoc work leaves a reusable definition behind. In the types of data analytics, it is usually diagnostic: something changed, and you need to know why.

Ad hoc analysis vs scheduled reporting vs exploratory analysis

Ad hoc analysis sits between two neighbors. Scheduled reporting answers the same questions on a calendar; exploratory analysis looks for questions you have not asked yet. Ad hoc work usually happens in self-service BI tools or a SQL console, by the person closest to the question. The last row of the table is the thesis in miniature: a scheduled report pays its cost once, while ad hoc analysis pays it every time unless the definition is saved.

Ad hoc analysisScheduled reportingExploratory analysis
TriggerA specific question arrives nowA calendar: daily, weekly, monthlyCuriosity, a new dataset, an unexplained pattern
Question shapeOne known questionThe same questions every cycleUnknown; you are looking for questions
OutputAn answer: chart, table, sentenceA recurring report or dashboardHypotheses, segments, new questions
ReuseLow, unless you save the definitionHigh: built once, run on scheduleMedium: findings feed later analyses
Who runs itAnalyst or ops person closest to the questionBI team or an automated scheduleAnalyst or data scientist
Main costRebuilding the definition from zero each timeMaintaining reports nobody readsTime spent without a decision

Three if-then rules keep them apart:

  • If the same question repeats on a schedule, it belongs in a scheduled report.
  • If you do not know the question yet, start with exploratory analysis.
  • If one specific question needs an answer now, run an ad hoc analysis — and decide afterwards whether to save it.

How to run an ad hoc analysis

Microsoft Learn's Business Central documentation lists three situations that call for ad hoc analysis: you don't want to run a standard report, no report exists for the need, or you want to iterate quickly. Five steps, each leaving an artifact:

  1. Write the question in one sentence. "Why did signups drop last week?" Artifact: the written question, with the metric and the comparison window.
  2. Pick metric, dimension and time window. Signup rate by channel, week A versus week B. Artifact: a definition — event, aggregation, filters, window — that someone else could rerun.
  3. Query it. SQL console or BI tool; one row per channel per week. Artifact: the result table. Keep the query text with the result; a screenshot cannot be rerun.
  4. Decompose the change. Split the total into parts that sum back to it, as in the example below. Artifact: a contribution breakdown, not a screenshot.
  5. Decide whether to reuse. Save the definition, schedule it, or archive the answer. Artifact: a saved metric or a written decision.

The artifact in step 2 is the one most teams skip, and it is the one that makes step 5 possible. For the broader workflow, see how to analyze data.

Worked example: why did signups drop last week?

Sample data. Week A: 25,000 sessions, 1,200 signups. Week B: 21,000 sessions, 900 signups. The question is where the −300 came from. The blended signup rate fell from 4.8% (1,200 ÷ 25,000) to 4.29% (900 ÷ 21,000), so this is not only a traffic-mix story.

ChannelWeek A: sessions × rate = signupsWeek B: sessions × rate = signupsΔ signups
Organic10,000 × 5.0% = 50010,000 × 4.5% = 450−50
Paid search8,000 × 5.0% = 4005,000 × 4.4% = 220−180
Social5,000 × 4.0% = 2005,000 × 3.6% = 180−20
Referral2,000 × 5.0% = 1001,000 × 5.0% = 50−50
Total25,000 → 1,20021,000 → 900−300

The deltas sum to −300, and paid search alone accounts for 180 ÷ 300 = 60% of the drop. But "paid search fell" is not yet an answer: did traffic shrink, or did the channel convert worse? Split each channel with the identity Δsignups = Δsessions × rate_A + sessions_B × Δrate:

ChannelTraffic effect: Δsessions × week-A rateRate effect: week-B sessions × ΔrateTotal
Organic0 × 5.0% = 010,000 × (−0.5%) = −50−50
Paid search(−3,000) × 5.0% = −1505,000 × (−0.6%) = −30−180
Social0 × 4.0% = 05,000 × (−0.4%) = −20−20
Referral(−1,000) × 5.0% = −501,000 × 0 = 0−50
Total−200−100−300

Traffic explains 200 ÷ 300 = 66.7% of the drop; rate explains 33.3%. The paid-search check: (5,000 − 8,000) × 5.0% = −150 from lost sessions, and 5,000 × (4.4% − 5.0%) = −30 from the lower rate. The answer: a paid-traffic cut, not a site-wide conversion problem — check the campaign before redesigning the signup form. This is diagnostic analytics in practice.

The query behind the first table, one row per session with a boolean signup flag (PostgreSQL SELECT syntax):

SELECT channel,
       date_trunc('week', started_at) AS week,
       COUNT(*) AS sessions,
       SUM(signup::int) AS signups,
       SUM(signup::int)::numeric / COUNT(*) AS signup_rate
FROM sessions
GROUP BY channel, date_trunc('week', started_at)
ORDER BY channel, week;

When to turn an ad hoc analysis into a scheduled report

The expensive part of ad hoc analysis is not the query — it is rebuilding the definition from zero each time. The asset is the saved definition (metric, segment, filters, window), not the screenshot. Three promotion rules:

  • Asked three or more times. Save the metric definition — event, aggregation, filters, window — so the next answer reuses it instead of re-deriving it.
  • The same definition appears in different questions. Promote it to a saved metric or segment. A definition copied into five queries is five places to get wrong.
  • A stakeholder waits for it weekly. Schedule it. An ad hoc query that runs every Monday is a scheduled report with extra steps; give it the structure of a data analysis report.

If none of these holds, keep it ad hoc and review the pile periodically — monthly is enough. Retire definitions nobody asks for; one-off reports tend to get re-run until they clog the reporting cycle.

Save five things with every reusable answer: the written question, the metric definition, the segment or filter definition, the time window, and an owner with a review date. That list is what the next person inherits — and what stops the next analysis from starting at zero.

Common mistakes

MistakeWhat happensFix
No written questionThe analysis drifts; nobody can reuse the answerWrite the question in one sentence before you query
Filters differ between the two weeksChannel deltas no longer sum to the totalFreeze metric, dimension, filters and window; reuse the same definition
One-off promoted without reviewA permanent dashboard nobody readsApply the reuse rules above; set a review date
Raw detail exported past the retention windowThe export is empty; the detail is already deletedCheck retention first; daily aggregates survive

Each of these is the same failure in a different costume: the definition was never written down, so the next person starts from zero.

Ad hoc analysis in TapCub

Take the signup drop from the worked example into TapCub. The goal is not only to answer it once, but to leave definitions behind so next week's version of the question starts from step 3, not from zero.

  1. Ask in one sentence. On the Ask tab of All-platform analytics › Operations › AI Assistant, type "signups last week versus the previous period". The question is parsed by rules first, and a model is called only when confidence is low or it is a follow-up. The numbers come from an analysis query run on this project, and the answer shows the query it ran. See ask the AI assistant.
  2. Split the change. The Why did it change tab shows each dimension value's contribution, its change divided by the total change, across channel, referrer domain, country, device, entry page and platform, one or two levels deep. On the sample data, paid search is −180 of −300, or 60%. It does not separate traffic from rate; for that second table, click Open in analyzer and put sessions and signup rate by channel next to each other.
  3. Save what you defined. Click Save as insight (the assistant never saves on its own), and save signup rate as a formula metric in All-platform analytics › Behavior › Metrics & SQL so every later chart uses the same definition. For a cross-tab the built-in models do not give you, the SQL tab of All-platform analytics › Behavior › Behavior analysis runs one SELECT against the project's v_events and v_sessions views, so a query like the one above can live next to the saved metric. See query data with SQL.
  4. Promote the question that keeps coming back. In All-platform analytics › Operations › Monitors, add a Period change rule on the saved metric (a weekly grain, firing when the change crosses, say, −15%) and dry-run it to see how often it would have fired in the last 30 days. If signups are a goal, the weekly scheduled report under Configuration › Settings › Notifications already includes conversions for enabled goals, on plans that include scheduled reports.

The boundaries: the assistant answers and explains, but it does not create monitors or send notifications, and a monitor only notifies; nobody's campaign gets paused for you. That is the point of the thesis: the query took seconds, and the saved insight, metric and monitor are what make the next ad hoc question cheaper.

Sources

Was this explanation clear?

We use your answers to fix and expand this page.

Su He

Data Analyst

Explores traffic sources, conversion funnels, and retention, using clear metric definitions and concrete examples to explain what the numbers mean.

FAQ

Questions people ask

Short answers to the follow-up questions that come up most.

Still have a question?

Chat with the team behind TapCub — we usually reply within a few hours on business days.

What is the difference between ad hoc analysis and a scheduled report?

Ad hoc analysis answers one specific question now, usually because no existing report covers it; the output is an answer, not a recurring dashboard. A scheduled report runs the same saved definitions on a calendar — daily, weekly or monthly — for questions that repeat. When an ad hoc question starts repeating on a schedule, it should become a scheduled report.

What is an example of an ad hoc analysis?

Signups fell from 1,200 to 900 in a week. An ad hoc analysis decomposes the −300 by channel: paid search lost 3,000 sessions (−150 signups at the old rate) and its signup rate fell 0.6 points (−30), together 60% of the drop. The answer — a paid-traffic cut, not a site-wide conversion problem — comes from one query, not a permanent dashboard.

Do you need SQL to do ad hoc analysis?

No. A BI console with filters and group-by covers most one-off questions, and an AI assistant can answer why a metric changed in plain language. SQL helps when you need a custom decomposition — like splitting a change into traffic and rate effects — or when the console cannot express the metric. The skill that matters is writing the question and the definition, not the syntax.

When should an ad hoc analysis become a scheduled report?

When the same question is asked three or more times, the same definition is reused across questions, or a stakeholder waits for it weekly. Save the metric and segment definitions first, then schedule the report. Otherwise keep it ad hoc and review the pile periodically. The goal is to stop rebuilding the definition from zero every time.

See this number on your own site

Add one script tag for web and apps, and the first reports fill in within minutes.

See your data clearly. Find your growth.

Every click, backed by data. Install one line of code and see your first numbers in a minute.

No credit card · Free plans for analytics and chat · Cookieless analytics