Case Study salmonrun.ai

3 million records. Just an address and a phone number.

How Salmon turned sparse CRM data into verified commercial business intelligence for a leading home services platform.

A national home services marketplace had millions of contractor and business records in their CRM. Most were barely more than an address and a phone number. No business name. No website. No way to know if the address was a residence or a commercial location. They needed the full picture — and their existing data vendors couldn't get them there.

3M+ CRM records
85%+ businesses identified
40+ fields enriched per record
4 wks onboarding to delivery

Millions of records. Almost no usable data.

The client's CRM had over three million records representing contractors and service businesses across the U.S. But the data was skeletal — the kind of records that make your ops team wince. Over 60% of records had no business name. Fewer than 15% had a website. Email addresses were almost nonexistent.

Missing Identity

Most records had no business name. Many had only a first name like "Pro" — a placeholder from self-service signups. No way to know who was actually at that address.

Address-Only Records

The most complete field was often just a street address and zip code. No website, no email, no employee count, no industry classification beyond a broad category like "HVAC."

Residential vs. Commercial

Many addresses were residential — a sole proprietor working from home. Others were commercial locations. The CRM couldn't distinguish between the two, making segmentation impossible.

They'd tried this before. ZoomInfo couldn't match — records were too sparse for their database to find a hit. D&B returned patchy results at best. Internal solutions required manual research that didn't scale and wasn't economical across millions of records. Every approach assumed you had enough data to match against. They didn't.


Research-grade enrichment. Not another database lookup.

Static data vendors couldn't solve this. The records were too sparse for a simple match — there was nothing to match against. Salmon's AI research agents went to work the way a human analyst would, but at scale.

01
Geo-locate and verify the address. Cross-reference against Google Maps, USPS data, and property records. Confirm it's a real, deliverable address. Flag residential vs. commercial.
02
Discover the business at that location. Search Google, Yelp, trade directories, and state licensing databases for businesses registered at or near that address. Match the industry category from the CRM to narrow results.
03
Verify and enrich the business entity. Confirm the business name, pull the website, phone, owner name, employee count, years in business, license status, and online reviews. Cross-validate across multiple sources.
04
Score and deliver. Every enriched field gets a confidence score and source attribution. Records are classified, deduplicated, and delivered back in a CRM-ready format.

Four weeks. From raw data to enriched CRM.

W1
Onboarding & data deep dive. Ingest the data, map the schema, understand the shape of the problem. What fields exist? What's missing? What's the industry distribution? Build the enrichment strategy.
W2
Analysis & strategy. Salmon runs initial analyses across the dataset — profiling data quality, testing match approaches, and refining the enrichment strategy before scaling to the full CRM.
W3-4
Enrichment & QA. AI agents execute across the full dataset. Discovery, verification, deduplication — all automated. A QA layer runs in parallel, flagging anomalies and low-confidence matches for review.
Delivery. Enriched records delivered in CRM-ready format with confidence scores, source attribution, and classification flags. No six-month integration project. No ongoing maintenance contract required.

01

Raw record arrives

A typical record from the CRM: a phone number, a broad industry tag like "Electrical," a partial address, and a placeholder name. That's it. No business identity, no website, no way to segment or prioritize.

02

Address verified

Salmon confirms the address is real, standardizes it to USPS format, and classifies it as commercial. The geo-coordinates are resolved. We now know exactly where this record lives.

03

Business discovered

AI agents search Google, Yelp, state licensing databases, and trade directories. A licensed electrical contractor is found at that address. The business name, phone, and reviews are pulled.

04

Entity enriched

Website, owner name, employee count, years in operation, license number, service area, and Google rating are all verified and attached. The skeleton record is now a fully qualified business profile.

05

Salmon output

A complete, confidence-scored record ready for CRM sync. Every field has a source and a score. The ops team can segment, prioritize, and act — no Googling required.

From an address to a business identity.

Raw CRM record
Name pro
Phone (336) 345-0215
Industry Electrical
Address 3806 casa vista lane
City / State winston salem, nc 27107
Business
Website
Owner
Employees
Type

This ran across millions of records. Not a sample. Not a pilot. The full CRM — enriched, classified, and scored in days, not the months a human team would need. Every record that could be matched was matched. The ones that couldn't were flagged with a reason.


From "we don't know who's there" to actionable business intelligence.

In four weeks, the client went from a CRM full of anonymous addresses to a segmented, enriched database of verified businesses. Their ops team could finally act on the data instead of working around it — and for the first time, actually quantify the pipeline sitting inside their own CRM.

85%+ Match Rate

Even starting from just an address and phone, Salmon identified the commercial business at that location with north of 85% accuracy. Records that couldn't be matched were flagged, not guessed at.

Residential / Commercial Split

Every address was classified. The client could instantly filter sole proprietors working from home vs. established commercial locations — a segmentation that was previously impossible.

Full Profiles, Deduped

Matched records came back with verified business name, website, owner, employee count, license status, and service area. Duplicates and overlapping entries were surfaced and merged. 40+ fields where there used to be 4 — and a cleaner database to work from.

The alternative: hundreds of analyst-hours. At 3M+ records, manual research would require a team of analysts working for months — conservatively 5-10 minutes per record for the kind of multi-source verification Salmon runs automatically. That's tens of thousands of hours of work, replaced by a four-week engagement. And every enriched, verified business record represents potential top-of-funnel pipeline that was previously invisible.


Address intelligence at scale.

If your CRM has addresses without context — residential vs. commercial, who's there, what kind of business — you're looking at the same problem. Different industry, same data gap.

What Salmon delivers for address-centric datasets

Whether you're qualifying service areas, segmenting install bases, or routing field teams — the starting point is always the same question: what's actually at this address?

  • Residential vs. commercial classification at the address level
  • Business entity identification — who operates there, what they do
  • Property and location attributes — building type, unit count, square footage
  • Contact enrichment for the business or property manager on-site
  • Confidence scores and source attribution on every field

The methodology is the same. Salmon's AI agents research every address the way an analyst would — Google, Maps, directories, licensing databases, property records — but across millions of records, in days.


Send us a sample. We'll show you what's there.

Share a few hundred addresses from your CRM and see what Salmon finds. No commitment, no contract — just proof.

kevin@salmonrun.ai

salmonrun.ai