Data accuracy

How data accuracy is checked in enrichment

Data accuracy asks whether the value is right, not merely present. A work email that belongs to a different employee can raise fill rate and destroy trust. Data quality writing treats accuracy as one dimension beside completeness and consistency 1.

RichAPI's match rate page puts match rate next to fill rate and data accuracy as core enrichment checks 2. Match rate asks whether the provider returned the requested value for submitted records 2. Accuracy asks a harder follow-up about whether that returned value is true for this person right now.

Ops checks accuracy with samples. Compare a frozen set of known-good rows to provider output. Spot-check titles against LinkedIn profiles. Verify emails before sequence enrollment. Data enrichment without an accuracy sample is just a coverage contest 3.

Accuracy decays. People change jobs. Domains rotate. Data decay means yesterday's accurate title can be wrong next quarter 3. Schedule re-checks on the fields that drive routing and personalization.

Credit design shapes accuracy work. Email verifier costs 2 credits per verdict on RichAPI, with misses billing zero on that miss-free path 4. Paying for verification is often cheaper than paying for replies to bad addresses.

Separate accuracy from consistency. Two systems can agree on a wrong title. Consistency without a ground-truth sample is false comfort. Keep a human-reviewed gold set for the segments that matter.

Publish the accuracy formula the same way you publish match rate. State the field, the sample size, the date, and who judged correctness. Without that, accuracy becomes a synonym for vendor confidence scores nobody can audit.

Accuracy work needs named owners. Assign a field steward for email, title, and firmographics. Without an owner, gold samples rot and vendors win every argument with a fresh slide.

Separate tactical accuracy from strategic accuracy. A wrong industry on a tiny account is noisy. A wrong work email on an enterprise buying committee is a brand incident. Weight the sample toward revenue-critical segments.

Instrument feedback loops. When a seller marks a bounce or a wrong-person reply, write that signal back to the accuracy scorecard. Enrichment that ignores seller friction will keep buying the same mistake.

Be careful with confidence scores. Vendor confidence is not your gold sample. Translate vendor scores into your own accept, review, or reject buckets with documented thresholds 12.

Budget time, not only credits. Human review on a slice of strategic accounts often returns more accuracy than another unattended bulk pass 3.

Publish false-positive and false-negative examples for each field. Sellers learn faster from concrete wrong titles than from a percentage. Keep those examples next to the gold sample so new hires inherit the standard 12.

When two providers disagree, do not average them. Pick a rule: prefer verified channel fields, then fresher first-party signals, then the provider that won the last gold sample on that field.

How to score accuracy without fooling yourself

Build a gold sample before the bake-off. Include hard cases: subsidiaries, common names, and recent job changes. Providers that look strong on easy rows often fail here.

Score field by field. Title accuracy and email accuracy are different problems. RichAPI already separates match rate, fill rate, and data accuracy in its enrichment framing 2.

Do not let fill rate proxy for accuracy. Empty-to-wrong is still a failure mode. Track wrong-fill rate next to net-new fill.

Use verification where the channel cares. Email verifier at 2 credits per verdict catches many bad mailboxes before a sequence 4. It does not prove the person is the right buyer, only that the mailbox behaves.

Re-run the gold sample after refresh jobs. Accuracy that only exists on day one is a launch metric, not an operating metric 1.

Report accuracy to finance in the same packet as credit spend. Otherwise coverage stories keep winning budget while bounce and complaint rates quietly rise.

Publish an accuracy SLA per segment. Enterprise accounts might require quarterly title review. SMB might accept automated refresh with spot checks.

Do not hide accuracy behind a single composite quality score. Leaders need to see which field failed.

When accuracy and fill rate conflict, write the tie-break. Many teams prefer fewer correct emails over more speculative ones for cold outbound.

Archive the gold sample results. Trend lines beat one-off bake-offs when renewals arrive.

Example of separating fill rate from accuracy on titles

Sales ops enriches titles on a mid-market outbound list. Raw title fill rate looks strong. A one-hundred-row gold sample shows many titles are stale relative to current LinkedIn profiles.

They add a manual review step for strategic accounts and keep automated fill only for industry and employee band, where the gold sample accuracy cleared the bar 2.

Emails still go through Email verifier at 2 credits per verdict before enrollment 4. Usable accuracy becomes the KPI: correct-enough title plus deliverable mailbox.

Next month the same gold sample is re-scored after a refresh. Accuracy drift shows up before reply rates fall 13.

Leadership sees fill rate, match rate, and accuracy on one scorecard instead of a single vendor coverage percentage 2.

They also tag seller-reported wrong-person replies for thirty days. Those tags feed the next gold sample so hard cases are not only the ones ops imagined.

Finance approves the next credit bundle only after accuracy on the strategic slice clears the written bar next to fill rate 24.

Data accuracy vs fill rate

Data accuracy asks whether a value is correct. Fill rate asks whether a field is non-empty after enrichment.