By Zach Wright, Cofounder at Syft AI
Teams trying to make Claude better at outbound usually upgrade the prompt and leave the inputs alone. That's backwards. Claude drafts and reasons well when the brief is specific. It invents personalization theater when the brief is a list.
Syft's answer to "how do I feed Claude who and why now" is a Value Match: a specific company, an active problem you solve, why the timing is now, and dated evidence with source links Claude can trust before it writes a word.
This isn't a Claude tutorial. It's the context layer that sits upstream of Claude in an outbound motion.
A row with company name, persona title, and website tells Claude what the account is. It doesn't tell Claude what is happening inside the account this week. Without that, Claude fills the gap with industry clichés and funding congratulations.
Who and why now means five resolved fields:
That package is also what an agentic GTM loop needs. Claude is one reasoning surface. The brief doesn't change if you later hand the same object to a sequencer or an AI SDR.
Common setup: an enrichment tool dumps 200 accounts into a sheet, someone pastes ten rows into Claude, Claude returns ten "personalized" emails. Reply rates look busy. Second meetings don't follow. The model did its job on bad context.
A cleaner test: for each account Claude will touch, write one sentence stating the reason to engage and paste the source URL and date. If you can't do that before Claude drafts, Claude can't invent a trustworthy reason either.
Syft ingests tribal knowledge from what you sell (Ingest), scans public evidence for those problems (Scan), and returns Value Matches (Match). Each match is the who/why object: account, problem, timing, rationale, evidence, sources.
Claude consumes that object through MCP or the Value Match API instead of through a paste of scraped prose. The model requests structured context, decides the outbound job, and your sequencer still sends.
Syft doesn't need to own the mailbox. It needs to stop Claude from guessing the reason.
Measure second meetings and pipeline, not only opens. If second meetings are flat, the context is still thin.
Enrichment and Clay-style workflows are strong at assembling attributes and running multi-step research recipes. Claude is strong at reasoning over a brief. Neither replaces continuous Scan/Match against your tribal knowledge unless you staff that layer yourself.
Many teams use both: Syft for Value Matches, Claude for the jobs on top of those matches.
Syft is the who/why layer for human and agentic outbound. Value Matches give Claude (and other assistants) checkable account context. MCP is the delivery path. Sequencing stays in the tools you already use.
How do I feed Claude who and why now for outbound?
Give Claude a resolved account brief: identity, active problem, why now, dated evidence with sources. At Syft that object is a Value Match delivered over MCP or API.
Is a better system prompt enough?
No. Prompt quality can't invent checkable evidence about a specific company this week.
Can I replace Clay with Claude?
Claude can reason and draft. Continuous detection of live problems against your win patterns is a different job. Teams often keep enrichment for assembly and Syft for Value Matches.
What context do AI sales agents need?
The same who/why package. Claude is one consumer of that package.
Do I need MCP?
MCP is the cleanest way for Claude to pull live Value Matches. Manual paste of structured fields works for a pilot. It doesn't scale as a weekly motion.
