Meet the Sales Director — and the Pipeline Skills That Work for Any Funnel
Every working creative eventually discovers they're also running a sales operation. The wedding inquiries you meant to answer. The brand that said "circle back next quarter" — which quarter? The retainer conversation that stalled somewhere between "sounds great" and a signed agreement. For most of us this pipeline lives in three places at once: a spreadsheet you stopped updating in March, a folder of starred emails, and your memory.
Today we're shipping something for exactly that. The Sales Director is the newest default AI character on your account — sitting next to the Director of Photography and the Video Director — and it's built differently from a chat assistant that merely talks about sales. It keeps the books:
You: Had a great call with Meridian Coffee today. They want a brand film plus monthly social cuts, probably $4–5k/month. Their marketing lead Dana is the decision maker.
Sales Director: That's a strong signal — recurring scope with a named decision maker. I've added Meridian Coffee to your pipeline at qualified, logged today's call, and recorded Dana as the primary contact. Want me to draft an intro proposal email for your review, and check back Thursday if they haven't replied?
Every part of that answer is real. The deal now exists in a pipeline you can open. The call is in an append-only history. The follow-up will actually happen Thursday. And the email — this matters — is a draft waiting for your approval, not something an AI sent on your behalf.
A coach that remembers it's an operator
The Sales Director's personality is a distillation of how disciplined sales teams actually work: qualification frameworks (budget, authority, need, timing), objection handling that treats "we don't have budget" as the start of a conversation rather than the end, and cadence advice that doesn't burn goodwill. Ask it to pressure-test your pricing, rehearse a negotiation, or turn a rambling discovery call into next steps — that's the coaching half, and it works from the first message.
The operating half is five skills it ships with:
| Skill | What it does |
|---|---|
| Update Pipeline | Creates and updates tracked items and their contacts, and appends activity-log entries — from conversation |
| Read Pipeline | Answers from the store: summaries, stalled deals, one item's full history, due follow-ups |
| Schedule Follow-up | Schedules its own one-time or recurring tasks ("check on Meridian Thursday") |
| Notify Owner | Sends you a push notification when something genuinely needs you — capped and throttled |
| Stage Email Draft | Writes outreach into your approval queue — it can never send anything |
Visit your AI Characters page and the Sales Director will be there — defaults are provisioned automatically for new and existing accounts. (Characters are a Premium feature, and like everything AI at 5AM, they run on your own Gemini API key.)
The three design constraints that make it trustworthy
An AI that writes into a CRM and drafts emails to real people needs harder guarantees than an AI that writes captions. We built three in as architecture, not policy.
1. The platform never sends outreach. Ever. Stage Email Draft does exactly what its name says — stages. Drafts land in the Outreach tab on the character's page, where you read them, approve or reject them, and send the approved ones yourself — one click copies the email, another opens it in your own mail app, prefilled. Then you hit "Mark sent" and the deal's history records it. There is no send button on our side, no sending infrastructure behind the feature, and the character's own instructions tell it to never claim an email went out — because until you sent it, it didn't. Cold outreach from your identity stays under your fingers.
2. Research carries receipts. When the character records something it found — "Meridian just opened a second location" — the entry must declare its confidence: confirmed, likely, or inferred. And confirmed is only accepted with the source URL it was read at. The Pipeline tab shows these as badges on every research entry, with the link one click away. An agent that can't file unsourced claims as facts is an agent whose pipeline you can act on.
3. The history is append-only, and everything has a ceiling. Activities can't be edited or deleted — a touch that happened stays happened, and every stage move is logged automatically with what changed. Hard caps bound everything the character can create: 200 pipeline items, 10 contacts per item, 25 drafts awaiting your review, 20 active self-scheduled tasks, 10 notifications a day. A confused model run can't flood your phone or your pipeline.
One more thing worth knowing: the pipeline is yours, not the character's. It's stored against your account — delete the character, rebuild it, or equip a different character with the same skills, and your deals, contacts, and history are all still there.
How to actually work with it
After a few weeks of living with the Sales Director ourselves, the pattern that works is simple: talk normally, and make it write things down.
End conversations with "log that." The store is only as good as what reaches it. After any real interaction — a call, a reply, a meeting — one sentence does it: "Log that as a call on Meridian: they asked for references, I'm sending two tomorrow." An unlogged touch didn't happen; the character's own instructions say so, and it will usually offer before you ask.
Open sessions with a question, not a scroll. "What's my pipeline looking like?" gets you counts by stage, total value, what's stalled, and drafts waiting on you. "What's gone quiet?" surfaces open deals with no activity in a week — the exact deals your spreadsheet was silently losing. Because the character reads from the store rather than the chat scrollback, this works in a brand-new conversation, weeks later.
Let it schedule the memory work. "If Meridian hasn't replied by Thursday, remind me" becomes a real one-time task — it runs Thursday, once, and then completes itself. One honest limitation: scheduling is by day, not time of day. "Thursday at 3pm" degrades to "Thursday," and the character will tell you so rather than pretend otherwise.
Use rejection as feedback. When you reject a draft in the Outreach tab, you can attach a note — "too formal, and never open with an apology." The character can read its rejection notes. The next draft is better. This is the tightest improvement loop in the whole feature: you're not configuring anything, you're just saying why.
Ask for a dashboard when words get slow. The Sales Director also carries the render_html skill — "show me my pipeline as a dashboard" gets you stage columns, values, and stalled flags as an inline visual, built from the real store.
And when you'd rather look than ask: the character's page has a Pipeline tab (every deal grouped by stage, expandable into contacts, history with confidence badges, and drafts) and the Outreach tab (the approval queue). The model writes; you read the same board.
The part we didn't expect: it's not actually a sales feature
Here's what happened during the build. Halfway in, we noticed that nothing about the machinery was really about sales. A pipeline is: named things, moving through named stages, with contacts, a history, follow-ups, and a done-state. That's also hiring. And fundraising. And wedding-season booking management. And choosing a print lab.
So we made the stage machine configuration. The skills are officially called Update Pipeline and Read Pipeline — not Update Sales Pipeline — and each character carries its own stage vocabulary in the skill's config:
{
"name": "pipeline_upsert",
"enabled": true,
"config": {
"stages": ["inquiry", "toured", "proposal_sent", "booked", "declined"],
"terminal_stages": ["booked", "declined"]
}
}
That's a venue-booking funnel. A recruiting character would carry ["sourced", "screening", "interviewing", "offer", "hired", "rejected"]. An investor pipeline: ["researched", "intro_made", "meeting", "diligence", "committed", "passed"]. The rules are minimal and universal: stages are yours in any order, moving into a terminal stage requires a stated reason (why you won, why you lost — the data that makes retrospectives possible), and reopening a closed item requires a note. Everything else — the append-only history, provenance rules, follow-ups, the drafts-only approval queue (a job application email and a sales email deserve the same "you send it" guarantee) — carries over unchanged.
Two details make this genuinely usable rather than technically-possible. First, the character always knows its own vocabulary: the configured stages are injected into its tools live, so it files things correctly on the first try instead of guessing at stage names. Second, the web UI adapts: the Pipeline tab groups by whatever stages actually exist in your data, and "stalled" is computed from open-vs-closed — not from hardcoded sales stage names — so a quiet candidate flags exactly like a quiet deal.
The Sales Director simply ships with the sales vocabulary as its default. Today, pointing a character at a different vocabulary means setting that config block via the character update API (a PUT with your personal access token — include your full skills list, since the payload replaces it); a visual stage editor in the Skills tab is on the roadmap. One current constraint to plan around: there's one pipeline per account, shared by every character you equip — continuity is the feature, but it means picking one domain for the board (or embracing the mix; open/closed logic stays correct regardless of whose vocabulary wrote each item).
Start here
- Open AI Characters — the Sales Director is already on your list.
- Tell it about one real, live deal — whatever's most at risk of slipping through the cracks — and let it set up the pipeline.
- Ask it to draft the follow-up you've been putting off. Review it in the Outreach tab. Send it from your own mailbox.
- Say "check back with me Thursday."
- Thursday, when the nudge arrives, ask: "what's stalled?"
That's the whole loop: research → record → follow up → draft → you send → repeat. The judgment stays yours. The bookkeeping finally isn't.
The Sales Director is available now for Premium accounts, alongside the pipeline skills — which any of your characters can be equipped with. Full technical details on the store, caps, and approval flow live in our docs; and as always, the platform never sends email on your behalf.
