A client’s paid channel showed flat numbers for three days. Traffic was down, conversions were down, and the temptation was to adjust the campaign.

The campaign was fine. A data connector had failed on Monday morning, silently — no error message, no alert, nothing. The data had stopped updating and the dashboard kept showing whatever it last recorded. Three days of decisions, all made against numbers that had stopped being true on Monday.

Most marketing data problems don’t announce themselves. A broken connector delivers stale data; a misconfigured tag drops events silently; an API rate limit quietly skips a day’s worth of records. The dashboard looks normal. The numbers look plausible. The problem only surfaces when a decision goes wrong and someone traces it back.

The fix is a systematic connector audit — five steps that verify your data pipeline is live before you read a single metric. It takes about ten minutes and saves hours of confused analysis. This post covers the full method, including how to brief Claude for business productivity to run the check for you.

Why Marketing Data Goes Wrong Without Warning

Marketing data pipelines are built from layers. Your ad platform connects to an analytics dashboard. Your ecommerce platform connects to a data warehouse. A third-party tool connects to a reporting layer. Each connection is a point of failure, and most of them fail without telling you.

The three most common silent failures:

Connector expiry. API credentials expire and need re-authentication. When they do, the connector stops pulling fresh data but doesn’t always surface an error in the dashboard. You see the last-cached numbers until someone notices they haven’t changed.

Timestamp drift. Some connectors update on a schedule, not in real time. If the scheduled pull fails or slows, data goes stale. A dashboard set to refresh every 24 hours might be showing yesterday’s numbers when you’re making today’s decisions.

Rate limit throttling. High-volume data sources sometimes throttle API calls. When that happens, records get skipped and gaps appear silently. The dashboard shows a partial picture; you see a trend that isn’t actually there.

All three failures look like normal data until you check the timestamps. The dashboard’s status indicator tells you the connector is configured — it doesn’t tell you whether the data flowing through it is current.

The 5-Step Connector Audit

This is the check that now runs at the start of every weekly dashboard review, before reading any individual metric.

Step 1: Map every data source to its connector

List every source feeding your dashboard — ad platforms, analytics, ecommerce, email, CRM — and for each one record the connector name, the platform it connects to, and the expected refresh frequency (hourly, daily, near-real-time). Most setups have five to ten connections. The mapping takes about fifteen minutes once and keeps every future audit fast.

Step 2: Check last-data timestamps before anything else

Every data source has a “last updated” timestamp somewhere — in the platform settings, in the connector’s status log, or in the raw data itself. Pull that timestamp for each source before you read any metric. The status indicator tells you the connector is configured. The timestamp tells you whether it is actually working.

Step 3: Flag any source more than 24 hours stale

If a connector’s last data is more than 24 hours old, treat every downstream metric for that source as unverified. Don’t make decisions against those numbers until the feed is restored. Set your threshold based on expected refresh frequency — for a near-real-time connector, flag anything more than two hours stale; for a daily batch connector, flag anything more than 36 hours old.

Step 4: Fix before reading

Rebuild or restart any connector that fails the timestamp check before pulling your weekly numbers. One stale connector can make an entire channel’s performance invisible — and if you’re measuring cross-channel efficiency, a gap in one source corrupts the comparisons. Most fixes are simpler than they sound: re-authenticate the credential, restart the scheduled job, or trigger a manual sync. The “resync” option is usually buried in the connector settings.

Step 5: Gate your analysis until all sources are verified

If you use Claude to help with weekly marketing reviews — and it saves about an hour each week when set up properly — build the connector check into the brief as the first step. A simple brief: “Before reviewing any metrics, confirm the last-data timestamp for each source in the list below. Flag any source where the last data is more than [X] hours old. Only proceed to analysis once all sources clear the check.”

This is the give Claude a job, not a question principle applied directly to your data stack. Define the outcome — verified data before any analysis — rather than jumping straight to “summarise this week’s metrics.”

How to Brief Claude for the Connector Audit

The five-step audit is straightforward to run with Claude, particularly if you’ve already set up a CLAUDE.md and actions.md structure for your business. The brief structure has five parts:

Connector inventory. Paste your connector map — source name, platform, refresh frequency — as context at the top of the brief. This becomes the standing list Claude verifies against each week.

Current timestamps. Either paste the last-update data or ask Claude to walk you through where to retrieve it per platform. Most platforms surface this in the integrations or connected accounts section.

Staleness threshold. Define the rule explicitly — more than 24 hours for daily connectors, more than two hours for near-real-time. A clear rule means Claude flags the right things and doesn’t flag the wrong ones.

Action on failure. Specify what you want: flag and pause, or flag and suggest the specific fix. Both work; the important thing is not to let the analysis proceed until the flagged connector is resolved.

Gate to analysis. Make this explicit in the brief: only proceed to the weekly metrics review once all sources are verified green. This is the same dependency-mapping principle that makes email automation reliable — document what needs to be true before the system runs, not after something breaks.

When you’ve built this into your workflow using the AI workflow stack audit method, the connector check becomes two minutes at the top of every session. The alternative is periodic fire-fighting — realising three days later that a campaign decision was based on bad data.

The Signal That Usually Points to a Stale Connector

The pattern to watch for: a metric that looks plausible but hasn’t moved much, in a week when you made significant changes. Revenue flat, traffic flat, ad performance flat — all of it stable while you were actively testing new creative, adjusting budgets, or changing targeting. That’s the scenario that often traces back to a stale connector, not a quiet market.

The other pattern: a sudden spike or drop that disappears when you look at it the next day. That’s usually a delayed data flush — records arriving out of sequence because the connector was behind and caught up all at once.

Both patterns are invisible if you’re not checking timestamps. Both become obvious once you are. The scope-before-rank method for prioritising growth actions depends on accurate data at the top of the stack — a stale connector corrupts the entire prioritisation exercise before it starts.

Data Quality as a Weekly Habit

The connector audit works best as a standing opening step in your weekly review — not a one-off check you run when something looks wrong. Build it into your weekly brief for Claude as the first gate, before any analysis begins.

The deeper principle is one that comes up across every part of using Claude for business: the quality of the output depends on the quality of the input. Claude can analyse your marketing data accurately and surface the right patterns — but only if the data it’s working from is actually current. Verifying the pipeline before analysis begins is the same habit that makes any automated task brief reliable: define what needs to be true before the process starts.

Your dashboard is only as reliable as the weakest connection in its pipeline. Ten minutes at the start of the week is cheaper than four weeks of confused results.

If you want a structured starting point for building repeatable weekly reviews into your workflow — including how to brief Claude to run the verification steps — the free NP System course covers how to set up the full process, from connector audit through to growth decisions.

Frequently Asked Questions

How do I know if my marketing data connector is broken? The most reliable indicator is the last-data timestamp. If the connector shows data more than 24 hours old for a daily-refresh connector, it has likely failed. Status indicators can show green even when the connection isn’t delivering fresh data — always check the timestamp, not the status light.

What are the most common reasons marketing data goes stale? API credential expiry, scheduled pull failures, and rate limit throttling are the three most common causes. Each one stops new data from reaching your dashboard without necessarily showing an error.

How do I check the last-update timestamp for my connectors? Most ad platforms and analytics tools surface this in the connector settings or integrations section. In Google Analytics 4, check the data freshness indicator in reports. In most ad platforms, look in the connected accounts section for the last sync time.

Can Claude help me audit my marketing data automatically? Yes. Build a brief that asks Claude to verify timestamps per source and flag any stale connectors before running any analysis. This works particularly well inside a recurring weekly review — the check becomes the first step of every session, not an afterthought.

How often should I run a connector audit? At a minimum, at the start of every weekly dashboard review. For businesses spending significantly on paid media, a daily check is worth building in — a 24-hour data gap on an active campaign can lead to meaningful misspend before the error is caught.