Finding integration and partner targets with co-occurrence data

Integration roadmaps are usually decided by whoever asked loudest. A big customer wants a connector to something, an exec meets a counterpart at a conference, a competitor ships an integration and someone forwards the changelog. Each of these produces a plausible candidate. None produces a ranking.

Co-occurrence gives you the ranking. The question it answers is simple: which vendors do the companies that already buy from us also run? That population — your customer base — is the best available proxy for the market you serve, and the tools it clusters around are your highest-value integrations and your warmest partner conversations.

Step 1 — enrich your customer base

Export your customer domains. One lookup each:

while read -r domain; do
  curl -sS "https://api.vendorstacks.com/v1/company/$domain" \
    -H "Authorization: Bearer $VENDORSTACKS_KEY" \
    >> customers.jsonl
  sleep 1.1   # 60 rpm default limit
done < customer_domains.txt

/v1/company is the right endpoint: index-only, never scans, instant, 1 credit. For a 300-customer base that's under 300 credits — misses cost nothing — so roughly $2.70 at the $10/1,100 pack.

Step 2 — count the co-occurrences

Flatten every vendor_stack into (category, vendor) pairs and tally across the base:

jq -r 'select(.found) | .vendor_stack
        | to_entries[]
        | .key as $cat
        | .value[]
        | "\($cat)\t\(.)"' customers.jsonl \
  | sort | uniq -c | sort -rn | head -30

What comes out is a ranked list of the vendors your customers run. Read it with one correction in mind, covered next.

Step 3 — correct for ubiquity

The top of that raw list will be the infrastructure everybody uses. AWS. A payments processor. A major email sender. Building an integration with a vendor that 90% of all companies run tells you nothing about your market — you'd have gotten the same answer from any customer base.

What you want is the same lift calculation an ICP analysis uses: how much more common is this vendor among your customers than in the broader population? The public vendor pages are useful here — each one shows what commonly co-occurs with that vendor across the corpus, so you can eyeball whether a pairing is distinctive or just background radiation. If a vendor sits high in your list and is not ubiquitous generally, that gap is your signal.

A rough scoring frame that works:

• Penetration — what share of your customers run it. Below ~15% and an integration is hard to justify on demand alone.
• Lift — penetration in your base divided by penetration in the broader corpus. Above ~2x means the pairing is characteristic of your market, not of the market.
• Adjacency — does data actually need to move between the two products? High penetration with no data flow is a co-marketing partner, not an integration.
• Category coherence — a vendor in a category you complement beats one in a category you compete with, at equal numbers.

Step 4 — size the opportunity beyond your base

Your customer base tells you what to build. The reverse lookup tells you how much market sits on the other side of it:

curl "https://api.vendorstacks.com/v1/prospect?vendor=Snowflake&category=analytics_data&page=1" \
  -H "Authorization: Bearer $VENDORSTACKS_KEY"

Ten entities per page, 1 credit per result returned, and a page that returns nothing costs nothing. Walk enough pages to understand the scale of the population that runs the candidate vendor but is not yet a customer of yours. That number — not the integration's technical difficulty — is what should decide whether it gets built this quarter.

It's also the number that makes the partner conversation work. "We'd like to integrate" is a request. "A meaningful share of our customer base already runs your product, and here's the overlap population we'd both be serving" is a business case. Partner teams field the first constantly and the second rarely.

Step 5 — turn the ranking into conversations

The output is three tiers, and each has a different motion.

Build. High penetration, high lift, genuine data adjacency. These go on the roadmap with the overlap count attached, so the decision survives the next reprioritization meeting.

Partner. High penetration, high lift, no obvious data flow. Co-marketing, joint content, referral motion. The overlap number is the whole pitch.

Watch. Rising penetration, low absolute numbers. Re-run the analysis quarterly and see which of these are trending. A vendor going from 6% to 14% of your base in two quarters is a better bet than one flat at 20%.

That last tier is the argument for re-running this. The full analysis costs a few hundred credits; running it quarterly and diffing the penetration numbers turns a one-time ranking into a trend line, which is what actually catches the integration you should have built before your competitor did.

Two honest caveats

Your base is a biased sample of your market. It reflects who you have sold to, which reflects where you have sold. Co-occurrence tells you about the customers you have; combine it with the reverse-lookup population before you conclude anything about the customers you want.

Coverage is uneven. This is built on public web evidence, and companies vary in how much they publish. A customer with found: false hasn't told you they run nothing — it means no public evidence was located. Exclude misses from your denominator rather than counting them as zeros, or you will systematically understate every penetration number.

The API reference covers the endpoints, and the GTM plays post has the other eleven things the same enriched base is good for.

Get a key — 25 free credits, no card — and enrich 25 of your customers. Even at that sample size the top of the co-occurrence list usually contains one name nobody on the BD team had on their list.