teradata. AI ROI · Talk Track Guide

Wave 1 · Draft v0.1 🇬🇧 EN Landing page Executor Open the base Talk Track ↗
ROI Talk Track · Field Guide

A companion, not a replacement.

The Cost of More parable ships with a rehearsed, beat-by-beat presenter script bundled in this pack. This guide is the Wave 1 sales-play layer on top of it — the CFO outreach choreography, persona-to-beat pairing, the turn to the AI Cost Model, the ROI-claim discipline, and the workshop hand-off that the base talk track doesn't cover.

↗

Open the base Talk Track first

Every reference in this guide points to a beat of the bundled storyline talk track (anchors #b0–#b7). Have it open in a second tab. The sections below map when to open which Executor persona, what to do between beats, and when to hand off to Strategy or Sovereignty. It does not repeat the base script.

🎙️ Base Talk Track EN 📊 AI Cost Model 📖 The Cost of More
👤Audience CFO · VP Infra · FinOps · Head of Data Platform ⏱Base script ~12 min (bundled) 📄Companion This guide 📊Discipline Every ROI number sourced 📰Facts Published reporting · "as reported"
Section 01

Before the meeting · Persona-tuned outreach

Goal: Use the Executor's persona essay to identify which of the six cost axes the prospect is publicly exposed on, then send a parable-anchored outreach. The Executor is not a "sales assist" — it is the discovery layer that determines where in the base talk track to dwell.

✉Sequence — Executor first, meeting second

  • 01Open the AI ROI Executor (executor.html). Paste prospect signals. Run Stage 1 to produce the persona essay.
  • 02Read the essay's "public gap" line — it tells you which beat to dwell on. Usually the margin question (Beat 1) for CFOs; the line and its layers (Beat 3) for VPs of Infrastructure; the measurement ladder (Beat 5) for FinOps leads.
  • 03Run Stage 2 to produce the outreach. Send it with the Cost of More parable link — the parable is the only asset that travels; the customer will read it and arrive primed.
  • 04The meeting then opens with the customer already carrying the distinction — a budget you can defend is not a margin you can state. That changes how hard each beat has to work.

🎯Persona → base talk-track beat pairing

CFO
Dwell on the margin. Beat 1 (a budget you can defend vs. a margin you can state) is their distinction, and Beat 4 (where it broke) is its P&L consequence — $5.8B of debt against $30M of cash, with the deciding number still unknown.
VP Infrastructure
Dwell on the line. Beat 3 (seven layers, one missing number) is their beat — and the readings on the line are written for them: every station carries its owner's green figure, and the product of those figures is the number nobody has. The bridge sentence "that is your AI estate" lands hardest with the person who owns the estate.
FinOps Lead
Dwell on the ladder. Beat 5 (four months of truth) is their beat: rungs one to three are their current practice, and rung four — the cost of a valid unit, this week — needs a denominator FinOps does not usually own. Say that respectfully; it is a missing instrument, not a failing.

▶Cue — what to test before the meeting

1Run the parable once on the machine and screen you will present from. Beat 3 is the only interactive beat — tap the stations, then work the seven-layer panel. Rehearse two moves: dragging one row to 100% (the answer barely moves), and switching the preset from 70% to 96% (four points per layer, and it is the whole story).
2Have three tabs open in order: why-the-cost-of-more.html (live), why-the-cost-of-more-talktrack.html (your script), dist/AI_Cost_Model.html (for the turn — and not before).
3Re-read the facts note at the top of the base script. The story names real companies; every figure is published reporting, and the two single-source lines — the scrap account and the 96%-in-four-months quote — are always spoken "as reported."

?Ask the Executor essay first

  1. Which of the six cost axes does the essay identify as this persona's publicly active concern? — That is where you dwell in the base talk track, and later in the model.
  2. Which axis is their public gap — the line item their published posture never mentions? — That is where the baseline workshop earns its keep.
  3. Does the essay surface a public phrase — an earnings-call line, a keynote quote — you can quote back verbatim? If yes, open the meeting with it before touching the parable.

🔀Cross-play routing signals

SIGNAL AThe essay leads with journey and roadmap language — "scaling AI", maturity, operating model. → Route to AI Strategy. Send The Ascent parable as follow-up.
SIGNAL BThe essay leads with jurisdiction and control — residency, lawful access, DPO/CISO pressure. → Route to AI Sovereignty. Send The Kill Switch parable as follow-up.
SIGNAL CThe essay says "our FinOps practice already covers this". → Do not compete with it. Ask which rung of the ladder it reaches, and position rung four — the valid-unit denominator — as the missing instrument, not a rival practice.
Transition to Section 2 Now the meeting starts. The base talk track is your storyline. This guide's Section 2 tells you what to do between the beats.
Section 02

During the meeting · Working the base talk track

Goal: Run the base talk track naturally — but use these Play-specific pauses, questions, and interaction cues to make the story land as a conversation about their invoice rather than a lecture about a battery factory.

◉Beat-by-beat pairing with the base talk track

Base TT · Beats 0–1 ~2:15
The hero · a budget you can defend vs. a margin you can state
Play addition: After the two cards, pause and ask: "If I asked your board today 'what is our marginal cost per useful AI outcome?' — what would they answer?" Do not correct whatever comes back. It is your anchor for the rest of the meeting.
Open base TT · Beat 0
Base TT · Beat 2 ~1:45
The flagship's published facts
Play addition: Walk the card line by line — $14B raised, the $55B order book, the cheapest power in Europe — and let the room feel that nothing was missing except one number. The red box only lands if the four lines above it landed first.
Open base TT · Beat 2
Base TT · Beat 3 ~2:00
Seven layers · one missing number (interactive)
Play addition: Tap Materials, then Formation, then everything-never-off. Then read the green figures under the stations aloud — every one of them says 95%, and every owner would defend it. Ask the room what the seven make together, wait through the guesses, then open the reveal bar beneath the line: 70% end to end, 1.43× per finished cell — and say "nobody in that factory reported the seventy." Take one station to 100% silently — 1.43 becomes 1.36. Then switch to the 96% preset and land the distance: four points per layer, four years against four months.
Open base TT · Beat 3
Base TT · Beats 4–5 ~3:30
Where it broke · four months of truth
Play addition: Do not rush the wall — and keep the verdict exact: "they were not choosing badly, they were choosing blind." Then the mood lifts: walk the ladder and ask the room which rung their practice stops at. Land the line flat: "The measurement saved nothing. Not one cent."
Open base TT · Beat 4
Base TT · Beats 6–7 ~3:00
Four verbs · the turn
Play addition: Name the quartet — re-time, re-place, re-choose, re-use — off the page, one layer each. The −90% pack-cost bars are compounding, drawn. Then slow down for the turn and open the model only after "stop defending the budget, start pricing the work" has landed.
Open base TT · Beat 6

▶Live discipline cues

1Never improvise a number. The only figures you speak are the ones on the page or on the approved list (Section 3). If the CFO asks "what would we save?" — the answer is the model, computed from their figures, not a guess.
2The villain is the missing instrument, never the people. Thousands of people built something remarkable; if the room hears mockery, the story dies. The base script's objection section carries the full defence.
3Do not open the AI Cost Model during the story. The parable never names it before the reveal, and neither do you.

?Ask between base TT beats

  1. After Beat 1: "When your team says the POC shows great uplift — what exactly is that measuring, and does it survive production volumes on your own invoice?"
  2. After Beat 3: "Who owns the denominator in your organization — the cost of one finished, valid unit of AI-assisted work? Is that anyone's job today?"
  3. After Beat 5: "Which rung is your FinOps practice on, honestly — spend, capacity, should-cost, or this-week's-valid-unit?"

⚠Objections the base script doesn't cover

"Our POC results are excellent. Why would production be different?"
POC results are usually real, but they measure demand-side outcomes on subsidized capacity. Production changes both variables — your invoice, and every workload sharing the same runtime. Not a criticism of the POC; a caution about extrapolating from it.
"Some of these costs are already addressed in our stack."
Maybe true. The exercise is to know which line items are still on your invoice, which are architectural, and which are inherited. Some you will accept, some you will not — the map is what lets you choose, and defend the choice to your board.
"We are growing — of course spend grows."
The flagship's demand was real too; demand was never the problem. Scale multiplies whichever margin you have. The number worth knowing before the next phase is the unit cost it will multiply — priced before it happens.
Transition to Section 3 The story ends on the reveal map. Section 3 of this guide is what to do when the room asks — and they will — "so what does our line look like?"
Section 03

The turn · From the factory line to their invoice

Goal: Hand off from the parable to the AI Cost Model — six cost axes, their own baseline, a delta computed from their figures. Then, and only then, the answer: this section holds the only moment a product is named, and every number spoken here is sourced.

◉The six cost axes — the estate's version of the seven stations

  • A1Inference. Per-call charges — the visible line. Owner: CFO · VP Infrastructure. Event: 3–5× YoY AI spend growth, with dashboards measuring everything except whether the tokens were useful.
  • A2Re-tokenization & sync. Vector store ↔ source of truth. Owner: CDO · Head of AI Platform. Event: enterprise RAG programs discovering vector-sync cost exceeded model-inference cost.
  • A3Movement. Data leaving the platform to be processed. Owner: Head of Data Platform. Event: cross-region egress that looked reasonable at POC and unaffordable at scale.
  • A4Idle capacity. Reserved capacity nobody uses. Owner: FinOps · Head of Data Platform. Event: commit terms locked before demand curves were known.
  • A5Contention & retries. Workloads fighting for the same runtime. Owner: SRE. Event: AI workloads triggering scale-out on shared clusters — BI paying AI's bill.
  • A6Version drift. The model changes quietly under the workload. Owner: Chief AI Officer. Event: silent alignment updates moving downstream metrics between API versions.

◉Working the model — tab by tab, their figures only

  • 01Cost Architecture (tab 2) — one axis per row. Walk only the two axes the customer signalled; ask which axis they think is dearest, then open that one. Coverage is demonstrated, not recited.
  • 02Your Baseline (tab 3) — their rates, their split. The model carries no prices of its own, only illustrative estates that their first real figure replaces. Say so — it is the trust move.
  • 03Self-Assessment (tab 4) — rate the two axes they care about on Hidden · Transparent · Optimized. The recoverable-spend figure moves as you rate, an unrated line is a finding, and an optimised estate scores zero — the model can return good news.
  • 04Business Case (tab 6) — compute the delta from their spend, volume, growth and phase. Change one assumption in front of them and let them watch the answer move: "this is your number, not mine."

◉The answer — the only place a product is named

  • 01One platform, six answers. Teradata Autonomous Knowledge Platform presents a coherent architectural response at every one of the six axes — not a stitched collection of point solutions. Read only the answers for the two axes the customer named.
  • 02In-Database processing answers A2 and A3 structurally: one document, one embedding, one storage location — no sync round-trips, no boundary crossings.
  • 03Workload isolation by design. Always-on baseline plus on-demand peaks — the AI workload pays the AI bill, the BI workload pays the BI bill. That is A5 answered at the architecture, not the invoice.
  • 04Unit-based consumption pricing correlates spend with useful outcome — Nucleus measured 14% year-3 cost reduction on this basis, and 427% ROI with an 11-month payback across Teradata customers.
  • 05Bring-your-own-LLM and embedding, pinned versions. The model you priced last quarter is the model you pay for today — A6 answered structurally.

!ROI-claim discipline — do not slip

  • ✓Every ROI number spoken comes from the approved list, with its source in the same sentence: 427% ROI over three years, 11-month payback (Nucleus Research, July 2025) · 14% year-3 cost reduction (Nucleus) · $0.0009 vs $0.0686 per query · 62× query volume vs Snowflake.
  • ✗Never project a savings number for this customer's business. The model computes recoverable spend from figures they enter — and returns zero for an estate that has already earned its margin.
  • !The model shows its published basis and the Teradata field uplift apart — the uplift figures are labelled and still awaiting field-data sign-off. Lead with the published basis; treat the uplift as illustrative until it is signed off.

?Ask at the turn

  1. Of the six axes, which two are still line items no one owns? Which one would your FinOps lead name first?
  2. Of the two answers we just walked, which changes how you would architect that axis differently than you had planned?
  3. If you could prove the address on one axis — small, contained — which would it be, and what does "proof" look like to your CFO office?

⚠Anticipated objections

"Isn't the 427% cherry-picked?"
Fair to test. It is a Nucleus cross-customer aggregate — published methodology, heterogeneous sample — not a Teradata projection. And it lands on a marginal-cost basis, which is why CFO offices accept it where they reject POC-uplift numbers.
"Consumption pricing sounds like variable cost. Isn't that the problem?"
The problem is variable pricing that scales with waste — tokens, sync, egress. Unit-based consumption plus architectural answers gives you variable pricing that scales with outcome. Different animal — and it is what the 14% year-3 finding measures.
"We can't rip out everything to consolidate on one platform."
Nobody is proposing that. The pattern most customers walk is to consolidate the workloads where the axes bleed hardest — usually A2 and A4 — and let the rest sit until the next natural refresh. Coexistence is the norm.
"What about the cloud commits we already signed?"
They stand. The answer runs alongside them, per workload; where the current architecture is fine, nothing changes. The compounding starts on the workloads you move — and shows up on next quarter's invoice, not next year's.
Transition to Section 4 You leave the room with two dates for the workshop — or you don't. Section 4 of this guide covers both.
Section 04

After the meeting · Workshop or follow-up loop

Goal: Convert the meeting into a scheduled 2-hour AI ROI Baseline Workshop with named participants. If it doesn't close on a date, run the follow-up loop back through the Executor. Either way: the parable link is the only asset you send — the model is used live, never distributed.

◉Path A — the meeting closed on a workshop date

  • 01Same day. Send a calendar hold with the parable link. Say: "Ask your FinOps team to walk the story before we meet — it makes the workshop twice as productive."
  • 02Day 3. Confirm participants. Insist on: CFO or VP Infrastructure + FinOps lead + one architecture lead — plus one skeptic. The skeptic makes the baseline credible internally.
  • 03Day of workshop. 30-minute baseline capture (tab 3, their rates and split) · 60-minute six-axis mapping across their top-3 AI workloads (tabs 2 + 4) · 30-minute compounding scenario (tab 6). Facilitated by a Teradata ROI architect.
  • 04Output. A written AI ROI Baseline, scored by axis and by compounding impact, every number sourced or computed from their inputs. Theirs to keep. No sales follow-up committed unless they request it.

◉Path B — the meeting did NOT close on a date

  • 01Same day. Do NOT send a "let me know when to reschedule" email. Send the parable link plus one of the six axes — the one their own words flagged in the meeting.
  • 02Day 3. Return to the AI ROI Executor. Regenerate Stage 2 with the follow-up variant. Include a memorable quote from the meeting itself — the Executor will weave it in.
  • 03Day 7. Send one paragraph from The Cost of More as a quote — the ladder, or "the unmeasured months are not delayed — gone." No link. Force a reply.
  • 04Day 14. Route through the FinOps lead — 30 minutes on the six axes is an easier yes than another CFO meeting, and it is where the real conversation starts anyway.

▶Follow-up guardrails

!Every number in every follow-up comes from the approved list, sourced in the same sentence. The parable link is the only asset you attach; the base talk track and this guide are internal, and the AI Cost Model is used live with the customer, never sent.

?Debrief with yourself

  1. Which of the six axes did the customer's language actually concentrate on? Was it the one the Executor essay predicted? If not — update the essay in your notes.
  2. Did anyone volunteer who owns the denominator — or that nobody does? Note the words — that sentence is the workshop's opening slide.
  3. Which routing signal (journey = Strategy, jurisdiction = Sovereignty, FinOps practice = calibration) did the customer trip, even weakly?
  4. What was the memorable phrase from this meeting that becomes the hook for the Day-3 follow-up? Write it down before you forget.

🔀Post-meeting hand-off matrix

CLOSEDPath A · 2h AI ROI Baseline Workshop scheduled. Collect their top-3 AI workloads and current phase in advance — the baseline capture starts faster with them named.
STALLEDPath B · Day-3 / Day-7 / Day-14 sequence. Cycle back through the Executor.
PIVOTCustomer signalled Strategy or Sovereignty concerns. Route to The Ascent or The Kill Switch. Loop back to ROI when the pivot conversation seasons.
End of guide The AI ROI play converts through the workshop, and the workshop's product is the baseline. Everything before it is earning the right to price the work. Everything after is selling from a baseline the customer computed themselves. Use the base talk track. Use this guide. Do not improvise a number.
The guide in one paragraph

The base talk track is the storyline. This guide is the sales choreography. Outreach → story → the turn → workshop — every number sourced, every axis priced from their figures.

🎯 What "landed" looks like

The CFO volunteers which two axes concern them most. They ask who reads which invoice line internally. Someone admits nobody owns the denominator. They accept two dates for the 2-hour AI ROI Baseline Workshop before you leave the room.

🛑 What "did not land" looks like

The CFO nods politely and names no axis. They ask for the deck. They say "let us think about it." → Do not push. Send the parable link, book a 30-minute follow-up with the FinOps lead, loop back through the Executor.

📄 Companion assets in order

1. executor.html — persona-tuned outreach.
2. Base Talk Track EN — the storyline script.
3. AI Cost Model — opened at the turn, used live.
4. Landing page — the sales-play strawman.

📊 Discipline you must never break

Approved list only, source in the same sentence. Never project a number for their business — the model computes it from their inputs. Single-source story lines are spoken "as reported." Naming per Guide D044422.

🎙️Open base Talk Track EN 📊Open the AI Cost Model ✉Open the Executor 🏠Sales-play landing page