Audit a Messy Enrichment Table Export

Marketing Academy · Field Work●Marketing Tools

All projects·From: Clay: The GTM Data Enrichment Platform

CoreAudit· 45 minutes

RateGain Travel Technologies

Objective: Given a 30-row enrichment export handed off by a contractor, diagnose where the waterfall is misconfigured and where AI columns are producing unmappable, multi-output text before the data gets pushed to the CRM.

You're the marketing ops analyst at RateGain Travel Technologies, the Noida-founded travel and hospitality SaaS company listed on the NSE since December 2022, reviewing a messy enrichment export from a contractor before it syncs to HubSpot.

Find where the waterfall billed more credits than it should have, and where an AI column's output is too messy to map into a CRM field.

Before you start

What you'll need

Free path (everything below is enough to finish)

Google Sheets

FreeFilter and sum the export to find billing and mapping problems

Free, and the export already arrives as a spreadsheet

Claude

FreemiumAnalyze messy AI column output and confirm how many distinct outputs it contains

Free tier is enough to review a handful of sample cells

The process

2 steps

Step 01 of 02

Diagnosing waterfall credit waste

The lesson's pricing model: a successful waterfall stop costs one credit, not one per provider attempted, since providers that miss don't charge.

The export has 30 rows and a 3-provider waterfall, but total credits billed is 118, nearly 4 credits per row. What's wrong?

Google Sheets— The provided export: columns lead_id, provider_attempted_order, credits_billed, final_email_found.

Procedure

Filter to rows where final_email_found is TRUE and sum credits_billed for just those rows.
Filter to rows where final_email_found is FALSE and sum credits_billed for those.
Flag any row billed for a provider listed AFTER a provider that already found a result in the same row.
Count how many rows show more than 1 credit billed despite a successful find.

Sample output

lead_id  provider_attempted_order         credits_billed  final_email_found
L-014    Apollo(hit), Findymail, Hunter   3               TRUE
L-019    Apollo(hit), Findymail, Hunter   3               TRUE
...
22 of 30 rows billed 3 credits despite an early hit; only 8 rows billed correctly (1 credit)

Healthy

Credits billed roughly equal the count of successful finds, one credit per stopped row.

Unhealthy

22 of 30 rows billed 3 credits despite finding an email on the first provider, meaning all three providers ran regardless of whether an earlier one already hit.

What this means

This isn't a pricing surprise, it's a broken waterfall: the providers were configured to run in parallel instead of stopping at the first hit, tripling the real cost of the same coverage.

So what do I do about it?

Symptom	Action	Effort
Credits billed are several times higher than successful finds	Reconfigure the table so each provider only fires when the prior one returned blank	30 min

YouYou can do this yourself, no engineering access required.

Step 02 of 02

Spotting multi-output AI column failures

The lesson warns that an AI column prompt asking for multiple things at once produces messy, hard-to-map output, one column should do one thing.

One AI column, meant to hold just a personalized opener, contains a paragraph mixing an opener, a lead score, and a company summary. Where did this go wrong?

Claude— The export's 'ai_opener' column, alongside the underlying prompt used to generate it.

Procedure

Paste 5 sample values from the ai_opener column into Claude and ask it to identify how many distinct pieces of information each cell contains.
Pull the original prompt used for that column and check whether it asked for more than one output.
Confirm the prompt reads like: 'Write an opener AND score this lead AND summarize the company.'
Recommend splitting it into three separate single-output columns, matching the fix pattern from the waterfall spec project.

Sample output

Cell L-014, ai_opener column:
'Hi Rahul, congrats on the recent funding round! I'd score this lead around 7/10 given the mid-size headcount, and RateGain looks like a strong enterprise travel-tech fit given their NSE listing and hotel distribution focus.'

Diagnosis: one cell contains an opener, a score, and a fit label, three outputs crammed into one column.

Healthy

Each AI column contains exactly one clean value that maps directly to one CRM field.

Unhealthy

A single cell contains multiple distinct pieces of information glued into one paragraph, unusable without manual splitting.

What this means

The output looks like a prompting quality problem but is actually a scope problem, the prompt was written to do three columns' worth of work in one shot.

So what do I do about it?

Symptom	Action	Effort
An AI column's cells mix multiple types of information	Rewrite as three single-output prompts, one per intended field	30 min

YouYou can do this yourself, no engineering access required.

Final deliverable

A short audit memo flagging every row with waterfall credit waste and every AI column cell with unmappable multi-output text, plus the specific fix for each.

See a reference example

Sample output

Go Digit General Insurance, enrichment export audit (excerpt)

CREDIT WASTE: 19 of 40 rows billed for 2+ providers despite an early hit. Estimated overspend: 21 credits.

AI COLUMN ISSUE: 'ai_summary' column mixes a company summary with a cold-outreach subject line in 14 of 40 rows. Recommend splitting into 'ai_summary' and 'ai_subject_line' as separate columns.

Success criteria

You're done when you can:

Correctly identifies rows where the waterfall billed for providers beyond the first hit
Correctly identifies AI column cells mixing more than one output
Recommends a specific fix for each problem type found