Interview Findings — Lisa (Sales) & Debi (Continuity), 2026-07-06

Distilled from Lisa Interview (1) Transcript and Debi Interview (1) Transcript. This is the ground truth that drives Specification. Corrections welcome — if anything here misstates what was said, fix it here first.

The headline corrections to our June analysis

1. Podtrac scraping is dead. Download numbers come from Uniblab, an internal analytics tool Patrick built. Numbers are pulled manually every Friday into per-client Google Sheets ("reporting documents"). Nothing numeric happens in TwitAds at all — no downloads, no delivery stats. The entire stats subsystem of the legacy app is unused.
2. The rundown PDF is dead; the rundown is not. After a 2-week experiment without rundowns produced chaos, Debi generates the rundown daily and copy-pastes it (not PDF) to the editors, plus per-ad notes in Slack. Editors and Micah rely on it. This is the most load-bearing daily artifact.
3. Host fees are still 100% in the system. Monthly cycle: report run on the 1st, per-episode host checkmarks, report goes to Lisa + Jennifer for payment.
4. "One user at a time" is specifically an order-builder bug. Three continuity people work concurrently fine; but two people building orders simultaneously can get lines cross-assigned to the wrong order.
5. Billboards, newsletters, social posts are sold today but live OUTSIDE the system (Excel orders). Lisa's #1 structural ask: everything sellable in one system.
6. Specials, banner ads (web display), blue pages / lower thirds, MP3 upload, surveys, live stats: all dead. Paul (current maintainer) has hidden most of it because deleting anything breaks the system.

The order lifecycle (canonical, from Lisa)

• The proposal IS the order. Conversations happen first; the system is only touched when it's time to send ideas. Often 2–3 proposals per client (recommended / minimum / option A/B) with a title field to tell them apart.
• Proposals hold nothing. The same inventory can sit in 20 pending proposals. "Nothing means shit until it's signed." First to sign gets the inventory.
• Finalize (on hard signature, via DocuSign today — print PDF → attach → DocuSign with 24h reminders): moves pending → finalized, secures inventory, enters fair rotation. Old proposals for that client are then deleted.
• Sea of red: when inventory sells out elsewhere, any pending proposal containing those dates shows red and cannot be finalized until amended. Must survive into the new system, and proposals needing attention should be highlighted.
• After finalize: signed hard copy → Google Drive archive; order → Jennifer for QBO booking (invoice spread, monthly billing); reporting document created; continuity runs the new/existing-client checklist; ad copy + placement propagate automatically in-system ("I typed nothing, the system does it automatically" — Debi loves this).
• Makegoods are rare (~2/year) because value-add is layered in up front. Handled as a separate $0 order noted "make good." Keep exactly this simple.
• Manual backup habit: every closed order is also recorded in a Google Drive doc in case the system crashes. Will continue during parallel run.

Lisa's requirements (sales)

Order building

• Everything currently on the order form is used — keep it all. Remove only: Add Miscellaneous, Add News (broken), and fix/replace the broken banner line.
• Editable/deletable advertiser names. Today a typo'd advertiser is permanent (she keeps an "availability" dummy account as a workaround). Delete OK to block if the advertiser has finalized history.
• Fields: title, advertiser, agency (or "TWiT" if direct), salesperson, quarter selection (Q1–4 / half / annual / check-all) + year, "start dates from today" behavior retained.
• Date picker per show works well today: white = available, red = sold, grey = full/unclickable. Keep it. Overselling is structurally impossible today and must remain so.
• Any salesperson building a proposal can override rates/downloads/discounts freely — no approval bottleneck; continuity reviews every order before it goes out anyway. Only the master rate card is restricted (Lisa decides, Debi enters — just those two).
• Multiple windows open at once (inventory beside order editor) is a daily pattern.

Sellable catalog — finite, not arbitrary: podcast ads (embedded/interstitial), opening billboard sold in weekly chunks, newsletter, social media posts, banners — everything on the rate card — plus exactly one custom/one-off line type (e.g., live event, free text + flat amount).

Softness / value-add math (the feature she most wants automated because "it's always wrong" done by hand):

• Free spots are layered into most orders as insurance against soft delivery. Value-add lines are zero-rated: contribute $0 and 0 guaranteed impressions.
• Wanted: mark line items as "layered in for softness"; below the line show subtotal → less value-add (impressions AND dollars deducted) → revised total. Some clients prefer it lumped (zeroed lines inline, current style) — need both presentations.
• Then a separate order-level discount (dollars only, never impressions).
• Agency orders show gross → net-to-TWiT (gross rate card is current and likely permanent; the net-card experiment is over).

Terms

• Net 15/30/45/60/75/90 + due-upon-receipt. Selecting one auto-populates the boilerplate below (prepay variant: pay 15 days before each calendar month). Advertiser name auto-fills into the terms text (manual today).
• The word "broadcast" disappears everywhere: it's an insertion order (IO), traditional calendar, calendar months. (10% of buyers still think in broadcast calendar — invoicing tolerates it, the UI default is traditional.)

Views & reports

• Inventory page is her most valued screen: % sold by show over any date range, drill-down to see which advertisers are on which episodes (competitive separation — TWiT offers episodic separation, rare in podcasting). Switch it to traditional calendar.
• Orders list: pending (blue) vs finalized (green) at a glance; advertiser history she consults constantly — wants collapsible per-advertiser rollups with total spend and active years, in date order.
• Sales-by-year report (the one Jeff built) reconciled against QBO quarterly — keep; extend to show gross / commissions / net to TWiT.
• Export everything to Excel and Google Sheets (prefers Sheets). Print order → PDF for DocuSign.

Explicitly deferred by Lisa (phase 2+): invoicing (QBO is the billing source of truth; Jennifer spreads monthly invoices), internal sales commissions (paid on client payment — needs QBO data), client reporting document automation, CRM (nice-to-have only; her Rolodex is LinkedIn), DocuSign API automation (print-PDF is fine at 30–40 signed orders/year). "Number one priority is a real sales system that does orders."

Debi's requirements (continuity)

Order intake (agency side, e.g. True Native Media): compare inbound IO against rate card + acceptable terms (e.g., refuse 15-day cancellation, require 30), negotiate via Slack, build it in the system to verify the math ("the system is not gonna lie"), leave pending until Lisa approves, sign, finalize.

Fair rotation — the daily craft: system auto-places on finalize; Debi then rearranges via up/down arrows so every advertiser rotates through positions fairly over the life of the order (small buyers get middle positions, never all-last). She re-audits monthly across the whole year. True Native tried to automate this and made "a huge mess" — manual control is required; automation at most suggests.

• Alias wrinkle: contract name ≠ brand name (Abine signs, DeleteMe airs). Placeholders sort under the contract name. Advertiser alias support would help.

Ad copy

• Authored in Google Docs — keep this workflow; the system stores the Doc identifier plus per-show promo code / URL / other, and substitutes {SHOW_TITLE} / {PROMO_CODE} / {PROMO_URL} / {PROMO_OTHER}. Enter once, propagates everywhere — their #1 error-elimination win, do not regress it.
• Copy arrives as agency scripts (6-page instruction docs), client paragraphs, or Veva writes it from the website (already using AI drafts; client approval loop applies).
• Assignment rules vary: same copy for months, per-time-window (Helix Sleep), sometimes per-episode.
• URL/code checking is manual (all three of them), and copy sometimes lands 2 days late — a morning attention list (missing copy, missing/broken URLs, unassigned slots) was welcomed.

Rundown — daily web page would be fine (better than today); PDF explicitly not needed. Editors need billboard notes/reminders per show (e.g., current Black Hat billboards). Micah gets the same rundown as editors; Leo gets copy links in Slack.

Episode schedule: Debi + Paul maintain. Day-swaps (e.g., Threat Locker) are done temporarily and immediately reverted "otherwise it breaks the entire system" — new system needs real, safe episode date moves. Show-day ad swaps effectively never happen; day-before rarely.

Frustrations: speed ("it's just very slow"), the order-builder concurrency bug. Otherwise: "for continuity, it works great." The prime directive is error elimination — errors cost money and force makegoods.

Deliverables owed: Debi is sending her new-advertiser and existing-advertiser checklists (with links to their docs); Shopify copy sample already sent.

People / roles clarified

What this changes in the plan

• Stats/reporting integration drops out of core scope entirely (Uniblab/Spotify integration is a clean phase-2 candidate; Leo already has code that logs into Uniblab).
• The product catalog (billboard weeks, newsletter, social, custom line) moves UP in priority — it's Lisa's #1.
• Softness/value-add/discount math is a first-class domain feature, not a notes field.
• Rundown output = auto-updating web page + copy-paste text block, not PDF assembly. No Google Apps Script, no personal-account infrastructure.
• Fair rotation UI (manual reorder) is core continuity workflow.
• Invoicing, commissions, client reporting docs, CRM, DocuSign API: explicitly phase 2+.