Who we sell to, and how every account earns its score. The model is deterministic and explainable: each point traces to a rule, a signal, and its evidence.
B2B software and technology companies with a meaningful AI/ML engineering org that are deploying LLM applications or agents to customers and need them to be reliable.
Informational; timing points come from funding signals. Min. headcount growth 5%.
Points available per category · always sums to 100
Industry, company size and geography against the ICP.
Buying signals: hiring surges, launches, relevant postings, tech adoption, pricing visits.
Funding, executive hires, expansion, and headcount growth.
AI/ML team size, LLM stack and platform stack compatibility.
Product usage and sales engagement with us.
Max points each signal type can contribute inside its category, and how fast it decays
| Signal | Category | Max points | Typical strength | Half-life | What it means |
|---|---|---|---|---|---|
| intent · signal points total 37, capped at the category budget of 25 | |||||
| AI/ML hiring surge | Intent | 10 | 0.85 | 45 d | Material increase in open AI/ML roles; the AI org is scaling. |
| Relevant job posting | Intent | 6 | 0.65 | 30 d | A posting that names the problem we solve (LLM evaluation, AI reliability, agent ops). |
| AI product launch | Intent | ||||
Same inputs always produce the same score. A 90-day half-life means a funding round is worth half its points after three months and a quarter after six. Band boundaries come from the score distribution and how much a team can work, not from round numbers — the panel below shows whether they order conversion correctly.
Backtest against realised outcomes · 1,212 contacted accounts · outcome: booked a meeting
The probability that a converting account outranks a non-converting one. 0.5 is a coin flip.
No leakage: every input is a firmographic or technographic fact known before first contact.
Low leakage: signals are external events. A signal GTMOS itself caused (a webinar we ran) is the exception and is not separated out in V1.
Leaks: engagement points are awarded for replies and meetings, which is part of the outcome being predicted. Its advantage over the other variants is the size of that circularity.
Edit a draft of v1, preview the impact on every account, then save it as a new version.
| 9 |
| 0.80 |
| 60 d |
| Shipped an LLM feature or agent to customers; production reliability now matters. |
| Technology adoption | Intent | 5 | 0.60 | 90 d | Adopted an LLM framework, model provider or vector database. |
| Pricing page activity | Intent | 7 | 0.70 | 14 d | Repeated visits to our pricing page from the account's network or known users. |
| Competitor adopted | Intent | ignored | 0.90 | 180 d | Public evidence the account deployed a competing platform. Not permanent — renewals come round — but the near-term window is closed. |
| AI initiative cancelled | Intent | ignored | 0.80 | 150 d | The programme our product serves was publicly shelved. The use case, not the budget, is gone. |
| timing · signal points total 14, capped at the category budget of 15 (shared with non-signal rules) |
| Funding round | Timing | 7 | 0.90 | 90 d | New priced equity round. Budget is being allocated and new initiatives start. |
| Executive hire | Timing | 5 | 0.75 | 120 d | New technical executive (CTO, VP Eng, Head of AI). New leaders re-evaluate tooling. |
| Expansion | Timing | 2 | 0.50 | 120 d | New office, region or business line; operating complexity rises. |
| Layoffs announced | Timing | ignored | 0.80 | 120 d | Workforce reduction announced. Budgets freeze, champions leave, and headcount-based fit figures are now stale. |
| Budget freeze | Timing | ignored | 0.85 | 90 d | Hiring or spending freeze reported. New spend needs an exception, which lengthens every cycle. |
| engagement · signal points total 10, capped at the category budget of 10 (shared with non-signal rules) |
| Website engagement | Engagement | 1 | 0.30 | 14 d | Visits to docs, blog or product pages. |
| Product signup | Engagement | 2 | 0.60 | 30 d | Someone at the account created a free workspace. |
| Invited teammates | Engagement | 2 | 0.70 | 30 d | Free workspace invited additional teammates: usage is spreading. |
| Integration activated | Engagement | 2 | 0.80 | 30 d | Connected a production model provider or tracing SDK. |
| Usage threshold reached | Engagement | 3 | 0.90 | 21 d | Crossed the free-tier trace volume threshold: a product-qualified account. |
| Champion departed | Engagement | ignored | 0.90 | 60 d | The contact driving the evaluation left the company. The deal's memory left with them. |
| Unsubscribed or asked to stop | Engagement | ignored | 1.00 | 365 d | Someone at the account asked not to be contacted. Compliance, not preference. |
The total score's advantage over the structural score is contaminated: engagement points are downstream of the outcome. Read the structural AUC as the score's real prospecting value.
Range restriction (contacted only) biases downward; targeting feedback (all accounts) biases upward. The only unbiased measurement is a holdout: contact a random sample of accounts regardless of score and compare conversion across score bands. GTMOS does not run one, because the demo sends no email.
| Grade | Accounts | Converted | Rate | 95% CI | Lift |
|---|---|---|---|---|---|
| Asmall n | 6 | 2 | 33.3% | 10%–70% | 2.91× |
| B | 160 | 28 | 17.5% | 12%–24% | 1.53× |
| C | 448 | 54 | 12.0% | 9%–15% | 1.05× |
| D | 598 | 55 | 9.2% | 7%–12% | 0.80× |