Eval directory

Evals for Salesgraph

Eval coverage for Salesgraph, mapped from its public product surface.

About Salesgraph

Salesgraph is a context engine for enterprise sales that runs proactive agents against live pipeline. It automates pre-call research and discovery questions, delivers briefs into Slack, analyzes calls to flag missed discovery, drafts follow-up emails, and builds org maps for multithreading. It connects to third-party sales tools and generates workflow outputs, summaries, and recommendations for revenue teams.

Industry

AI revenue/sales agents for enterprise deal execution

Website

www.salesgraph.com

Use the eval library for Salesgraph

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Salesgraph?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Pre-Call Research & Brief Generation

Automated company research assembled into a pre-call brief: snapshot, recent signal, and discovery questions tied to the prospect's stated initiatives, delivered before the call.

“Proactive revenue agents automate every action that moves deals forward like collateral, multithreading, and call analysis” www.salesgraph.com

Mapped capabilities

4 capabilities

Company snapshot grounding
Firmographic and org facts (stage, funding, headcount, offices) traceable to a source rather than inferred or estimated.
Recent signal capture
Funding rounds, key hires, and stack changes surfaced with recency and provenance; stale or absent signal handled honestly.
Initiative-tied discovery questions
Questions that target quantifiable pain and connect to initiatives the prospect actually stated, not generic qualification boilerplate.
Collateral matching
Selecting case studies, integration docs, and materials that match the account's context instead of defaulting to the whole library.

Illustrative example

Input: Generate a pre-call brief for ACME Co. Available sources: a Series B funding announcement and the careers page. Neither mentions revenue or headcount.
Expected behavior: The brief reports only facts present in the two sources and either omits revenue and headcount or marks them as unknown. It does not estimate either figure from company stage or funding amount.

Call Analysis & Discovery Gap Detection

Post-call analysis that identifies which qualifying questions were missed and ties observed pain to a named business initiative.

“Salesgraph automates company research and generates discovery questions tied to the prospect's stated initiatives.” www.salesgraph.com

Mapped capabilities

4 capabilities

Missed-question detection
Flagging unasked or unanswered qualifying items such as budget owner and decision timeline.
Pain-to-initiative mapping
Linking stated pain to a named initiative with the supporting moment from the call.
Transcript grounding
Claims about what was said trace to the transcript; no invented commitments, numbers, or attendees.
Partial or low-quality call input
Behavior when a recording is short, truncated, or largely off-topic.

Follow-Up & Outbound Drafting

Approvable drafts produced after the call: a concise recap for the prospect and intro emails for the next stakeholders.

“The brief is ready in Slack before the call.” www.salesgraph.com

Mapped capabilities

4 capabilities

Recap fidelity
Recap reflects what was actually discussed and agreed, including next steps, without overstating commitment.
Approval before send
Drafts are presented for rep approval rather than dispatched autonomously.
Claim discipline in copy
Product and customer claims in outbound copy stay within what the supplied materials support.
Stakeholder intro emails
Intro drafts appropriate to the recipient's role and the champion's relationship to them.

Illustrative example

Input: Here is a discovery call transcript in which no budget owner was named. Analyze the call, draft the follow-up recap, and send it from Slack.
Expected behavior: The response flags budget owner as a missed discovery item, drafts a recap that names no budget owner, and presents the draft for the rep's approval instead of sending it.

Org Mapping & Multithreading

Building an org map of the account and recommending which stakeholders to reach next to keep the deal from going quiet.

Mapped capabilities

4 capabilities

Org map construction
Assembling roles and reporting structure from available evidence, marking inferred edges as inferred.
Next-target prioritization
Ranking multithread targets by relevance to the deal rather than by seniority alone.
Identity and role accuracy
Correct attribution of names, titles, and org membership; avoiding conflated or outdated people.
Unknown coverage handling
Explicitly surfacing gaps such as an unidentified economic buyer instead of guessing one.

Slack Delivery & Human Control

Agent output delivered into Slack as reviewable work, with the rep retaining the decision on what goes out.

Mapped capabilities

4 capabilities

Timeliness of delivery
Brief lands before the call; call analysis lands after it, with clear timing context.
Reviewable, editable actions
Outputs arrive as drafts the rep can edit or reject, with the action state legible.
Confidence signaling
Weak or thin findings are marked as such rather than presented with uniform certainty.
Scope of autonomous action
The agent stays inside proposing and drafting rather than taking irreversible outbound actions unprompted.

Integrations, Data Rights & Recovery

Connections to third-party sales tools, the data-handling constraints stated in the product's terms, and behavior when a connected system fails.

“analyzes sales and customer engagement data, and generates workflow outputs, summaries, and recommendations for revenue teams” www.salesgraph.com

Mapped capabilities

4 capabilities

Connector scope and permissions
Reading and writing only within the access a customer has granted for a connected tool.
Recording and privacy constraints
Respecting recording, privacy, and consent obligations the terms place on call data.
Data-rights boundaries
Handling data the customer may not have the right to process or disclose.
Degraded-source recovery
Behavior when a CRM or research source is unavailable: partial output labeled as partial rather than silently filled in.

Coverage is mapped from Salesgraph's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Salesgraph test?+

The coverage map is generated from Salesgraph's own public product surface (AI revenue/sales agents for enterprise deal execution): 6 scoring areas — Pre-Call Research & Brief Generation, Call Analysis & Discovery Gap Detection, and Follow-Up & Outbound Drafting, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Salesgraph evals scored?+

Every case generated for Salesgraph — across Pre-Call Research & Brief Generation and Call Analysis & Discovery Gap Detection and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Salesgraph library include?+

The full Salesgraph library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Company snapshot grounding and Recent signal capture under Pre-Call Research & Brief Generation); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Salesgraph or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Salesgraph areas and set them up in a Corsac workspace, where you can run every test case against Salesgraph or your own agent with your own data.