The best signals show that AI has reached production and created a problem someone must fix. Watch for job posts for inference, ML platform or AI infrastructure engineers, an AI feature shipped to paying customers, use of several models at once, and new AI security or governance roles. A company that only mentions AI in its hiring tells you almost nothing.
Why does "uses AI" fail as a signal?
Almost every company now qualifies. Stanford's AI Index reported that 78% of organisations used AI in 2024, up from 55% the year before (Stanford HAI, April 2025). Indeed Hiring Lab found that AI-related terms appeared in 5.9% of US job postings by late June 2026, past the 2022 peak of 3.3% (Indeed Hiring Lab, July 2026).
Some functions are saturated. In December 2025, nearly 45% of data and analytics postings mentioned AI, along with 14.9% of marketing postings and 8.8% of HR postings (Indeed Hiring Lab, January 2026). A data analyst post that asks for "experience with AI tools" says the company uses a chatbot. It says nothing about a purchase of serving, retrieval or evaluation software.
An infrastructure purchase needs three conditions: a system in production, a team that owns it, and a cost or failure that the team cannot fix with the tools it already has. Look for signals that show at least two of the three.
Which signals point to an infrastructure purchase?
These are the signals we would rank highest for an AI infrastructure or developer tools seller, with the product category each one fits:
| Signal | What it shows | Products it fits | Where to find it |
|---|---|---|---|
| Job post for inference, model serving or ML platform engineer | A team owns production models and needs more capacity | Serving, GPU scheduling, observability | Company careers page, job boards |
| AI feature shipped to paying customers | Production traffic, with uptime and cost now visible to customers | Serving, caching, evaluation, guardrails | Changelogs, release notes, product pages |
| Engineering blog post or talk about GPU cost, latency or model quality | The team has measured a problem and published it | Cost optimisation, evaluation, routing | Engineering blog, conference agendas |
| Several models in use | Routing, fallback and evaluation across providers | Gateways, routing, evaluation | Docs, job posts naming several model providers |
| AI governance, AI security or model risk hire | Security and compliance review of AI systems has started | Access control, audit logs, data protection | Job boards, leadership announcements |
| Data platform rebuild for AI | Retrieval quality is limiting the product | Vector search, data pipelines, document processing | Job posts, migration posts on the engineering blog |
| Funding round that names an AI product | New budget aimed at that product | Any layer the product depends on | Press releases, funding news |
Multi-model use deserves attention. In its survey of 100 CIOs, a16z found that 37% of respondents used five or more models, up from 29% a year earlier (a16z, June 2025). The same report said agentic workflows were making model switches cost a lot of engineering time. A company in that position has a routing and evaluation problem, whether or not it has named it yet.
How do you read a job post for buying intent?
Score the words that show ownership of a system, and ignore the words that show tool use. Use this procedure on every post:
- Find the verb. "Build", "own", "scale" and "operate" point to a team that runs infrastructure. "Use" and "familiarity with" point to a team that consumes it.
- Find the system. Score posts that name inference, model serving, GPU clusters, evaluation pipelines, retrieval or fine-tuning. Skip posts that only name a chat assistant or a coding assistant.
- Find the pain. Phrases such as "reduce inference cost", "improve latency", "reliability at scale" or "move from prototype to production" describe the problem your product solves. Quote them back in the first message.
- Count the seats. Two or more open roles on the same AI platform team mean the team is growing faster than its tooling.
- Check the reporting line. A role that reports to a CTO or VP of Engineering sits close to the budget. A role inside a business unit tells you which unit pays.
A post that passes steps 1, 2 and 3 is worth a message this week. A post that passes only step 2 belongs on a watch list.
How do you match the signal to the right company?
Match the exact company name first, then confirm the domain. Namesakes are the main data risk in signal-based outbound. In our own monitoring, 91 signals named a company that our contact database could not match to exactly one firm of 11 to 1,000 staff (our data, October 2026). We also caught two signals, one of them a hiring signal, that matched a different company with the same name. Our fix is to take the top-ranked exact-name match across all company sizes, then apply the size filter.
Many AI company names reuse short common words, so check that the domain on the job post or press release matches the domain of the record before you contact anyone. Our wider filter rules for hiring, funding and TVL signals are in which buying signals predict a B2B purchase.
How should you weight signals that arrive together?
Weight combinations above single events. One job post can be a backfill. A job post for an inference engineer in the same month as a launch of an AI feature shows a production system and a team under load. Use a simple score:
| Combination | Action |
|---|---|
| Ownership job post plus a shipped AI feature | Contact the platform lead this week |
| Engineering post about cost or latency, alone | Contact the author with a specific comparison |
| Governance or AI security hire plus a shipped feature | Lead with deployment options and data handling |
| Funding round only | Add to the watch list until a hiring or product signal follows |
| Generic "AI skills" in non-technical posts | Ignore |
A decision rule for AI infrastructure signals
Contact an account when you can show two things from public evidence: AI runs in a product today, and a named team owns a problem your product fixes. Write the first line from that evidence, in the company's own words. Keep every other account on a watch list and check it monthly.
To build signal rules and a scored account list for your product, start with The Diagnostic at $2,500.