The Mediaocean file

01The company

Who they are, who owns them, who runs them.

Mediaocean is the back office of the advertising industry. Formed in 2012 by the merger of Donovan Data Systems and MediaBank (announced 2011), it is the system of record through which agencies plan, buy, traffic, reconcile, and pay for media: over $200 billion in annualized ad spend, more than 100,000 Prisma users, and a finance line (Prisma Finance) that alone touches $150 billion a year. The company itself is smaller than the flow it carries: about 890 people on the last published count (2020, before Flashtalking and Innovid folded in, so today's number is materially higher and still small against the flow; revenue is private and unreported, so treat any dollar figure you see quoted as modeled). That is the profile of a clearing rail: a small company carrying enormous flow. It owns no media inventory and positions itself as the neutral, independent platform for the buy side. Bill Wise, co‑founder and CEO: "Our industry doesn't need more point solutions; it needs cohesion."

Ownership explains behavior. Vista Equity took majority ownership in 2015 and held it until CVC and TA Associates acquired its stake in August 2021. Then the striking move: in December 2024, IPG, Omnicom, and WPP each took a small minority equity stake, tied to a Certified Service Partner program under which they resell Flashtalking and Protected (the stakes cover the ad‑tech resale, not Prisma itself). The largest customers are now shareholders and resale channels at once; Publicis has run on a seven‑year international engagement since February 2023. Neutrality, at this point, is capitalized by the client base itself.

The leadership map

Sources: CVC/TA · holdco stakes, AdAge · Markman appointment

02The machine

Prisma plan‑to‑pay, stage by stage, and where we land.

Prisma organizes a media buy as six stages. Planners and buyers own the front of the pipe: campaign setup, RFPs and seller negotiation, the media plan, sending insertion orders, pushing placements to ad servers and DSPs over API. After launch the media desks (planner, buyer, trafficker; separate chairs at holdco scale) own optimization, actualization, and reconciliation from inside the campaign. Finance runs a separate track (invoice entry, credit memos, cash receipts, adjustments), and Mediaocean's own training curriculum tells finance users, in Prisma, to "review the planner's workflow after a financial adjustment." That planner‑to‑finance round trip runs by hand today, which makes it the place a priced record earns its keep first.

The money side matters as much as the workflow side. Prisma Finance carries estimates and job costing, invoice‑to‑order matching, sequential liability enforcement, accruals, and ERP integration. The mechanics are concrete: DV360, configured at budget‑grouping level, emits one invoice per Prisma order number, and Prisma shows booked budget against invoice amount and flags the discrepancy so finance can pay or dispute without manual matching. Electronic invoicing already connects Meta, Google, FreeWheel, Magnite, IAS and others. Realized outcome and committed spend already sit side by side inside their system. That is precisely the surface a calibration track needs, and it is why the reconciliation wedge is our first, not our last, proposal.

The integration surface

Third parties integrate through the Prisma Integration API, gated on a Mediaocean developer account and an issued key. The developer portal is closed to the public. Getting a key is a partnership decision, not a signup form, which is why the relationship is the product here.

Sources: prisma‑media · Prisma learning plan · prisma‑finance · integrations · DV360 invoicing

03The estate, 2024–2026

Three brands, the deals that reshaped them, and where NIVO actually sits.

One piece of deal archaeology worth knowing cold, because it shows how they buy: the Innovid merger proxy reveals Mediaocean opened at $1.70 a share in September 2023 (rejected; the stock sat at $1.21), then walked the price up through $1.80, a $3.00–3.50 range, $3.11, $3.20, and $3.25, settling at $3.15 after diligence, still a 94% premium, with rival bidders in the process and Mediaocean's own COO (Flashtalking's founder) having initiated contact. When they decide a capability is strategic, they pursue it for eighteen months and pay nearly double the market for it. That is the buyer psychology walking into the August room.

After a decade of acquisitions the estate has settled into three brands. Prisma is the workflow and finance system of record, now in five lines: Media, Finance, Production, Direct, and for Sellers. Innovid is the unified ad‑serving brand (the $500M Innovid deal closed February 2025 and absorbed Flashtalking, which was retired as a brand a month later; TVSquared came with it). Protected is MRC‑accredited verification. Planning is the notable exit: Lumina was divested in 2023 to Guideline's parent, and the Cannes 2026 news made Guideline the exclusive planning partner with a bidirectional API into Prisma, pitched openly as killing spreadsheet reconciliation. That exclusivity matters for us: Guideline owns plan‑vs‑actual, so our calibration story must run at the decision level (predicted vs realized per decision), never as plan‑vs‑actual reconciliation.

NIVO AI launched June 11, 2026: "NIVO thinks, Orchestrator connects, Agents act," twelve agents in four groups (Create, Deliver, Measure, Optimize), with pilots at Canvas Worldwide, FanDuel, Optimum, Paddy Power, and Tailwind claiming up to 90% faster campaign launch (a figure from five pilot accounts). Two facts about NIVO shape our whole approach. First, it launched on Innovid's serving foundation but is headed for Prisma: Mediaocean says NIVO "is expected to surface across all Mediaocean surfaces and solutions," and Innovid's president confirms expansion to Prisma and Protected. Second, every NIVO announcement to date omits governance of autonomous decisions, audit trails, incrementality, and human‑oversight rules. One adjacent fact: Innovid's Outcomes suite (extended with Affinity Solutions in June 2025) sells purchase attribution at scale, so expect it as the answer to any measurement framing; attribution says what happened after the ad ran, and still nobody prices the decision or calibrates the prediction. So the wedge: the governance their roadmap needs is what the substrate provides as built, landing Prisma‑side, with NIVO inheriting the answer without being asked to change.

The moves, dated

The pattern across every row: Mediaocean buys or partners for capability; it does not build analytics itself. Guideline got planning as an exclusive bilateral API. Nielsen got audiences. Basis got execution. That is the partner slot Precise walks into, and the precedent for how it gets structured.

Sources: Innovid close · brand unification · Prisma Direct · NIVO launch · Guideline API · Basis interconnect · Nielsen

04Who pays, and for what

Value by customer type: what they buy, at which stage it lands, and where it runs thin.

Two through‑lines for the team. First, the moat is the ledger and the liability chain, not the software: switching cost lives in finance, which is why the platform holds strong through every product cycle. Second, every segment pays Mediaocean to keep the record of what happened: what was ordered, what ran, what it cost, who owes whom. Nobody in the building, on any row of that table, is paid to say what it was worth. Hold that sentence; section 08 is built on it.

Sources: CSP stakes, Marketing Dive · independent‑agency view, Bionic · Prisma Direct, ppc.land · Wren agent‑to‑agent pilots, Digiday · Prisma Finance

05The seam

What nobody in the building owns, in Mediaocean's own numbers.

All four figures from Mediaocean's own 2026 Advertising Outlook. Release.

Set NIVO's launch sheet against that survey and the seam shows. The launch covered speed: 90% faster campaign setup, twelve agents, pilots named. The governance layer is still open: autonomous‑decision governance, exportable audit trails, incrementality, and human‑oversight rules all remain to be written, and their own roadmap points straight at them. Meanwhile the IAB reports two‑thirds of buyers testing or planning agentic AI, IAB Tech Lab is standardizing exactly that requirement through the AAMP protocols and the January 2026 AI Transparency and Disclosure Framework. That standard is Mediaocean's to meet first among its peers, and our signed decision records are the compliance story ready to meet it.

The screen anatomy sharpens the same point. Prisma's invoice module already has a History icon: a field‑level audit trail of every edit. It logs edits, not decisions. It tells you who changed a cell; the decision record adds what was known, what was chosen, what was predicted, and whether the prediction held. One icon over, the same grid that auto‑flags booked‑vs‑invoiced discrepancies is one column away from saying what a flagged variance was worth. And the settlement rail our records should ride already exists: IPS processes over $50B a year in agency payables into Prisma for eight of the nine largest holdcos.

Omnicom's Wren unit is piloting agent‑to‑agent direct buys explicitly framed as cutting the intermediary toll, and Prisma Direct's own publisher pitch is "fewer intermediaries." When agents transact with agents, a workflow rail becomes most valuable when it carries the one thing agents cannot supply themselves: trust in the record. That is the biggest version of the seam, and Mediaocean is its natural owner.

So: Mediaocean owns workflow, finance, serving, and verification. Its partners own audiences and panels. Its customers' in‑house stacks want the decisioning. Nobody owns the record a governed agent needs: what the agent knew, what it chose, what it predicted, what happened, and what that taught the system, written down completely and signed so anyone can re‑check it. That record is our product. Everything else in this file is the route to selling it.

06Our arc

Cannes to now. Times local, New York.

• 2026‑06‑24The arc starts early. Konstant's first Mediaocean deck ("The score under the spend": system of record to system of judgment, the Boombox thesis, crawl‑walk‑run, the partnership shape, the honesty beat) is built a month before the Jul 24 call. konstant-explainers/mediaocean.html
• June 2026 · CannesBill Wise opens the door. Bill (a personal investor in Precise) asks for help with "true orchestration, beyond any single capability plugging in." We convert that into two parallel Phase 1s rather than one grand integration.
• 2026‑07‑22Plan‑to‑pay demo built. One forward‑safe screen: a converged linear × digital CPG flight, planned share against credited share by publisher, a priced rebalance, a sealed decision ledger, and a governed human release gate. Live at precise-plantopay-a17c697b.surge.sh.
• 2026‑07‑24The call: Adam pitches Bill, Yvonne, and the room. Bill frames two threads: "accelerate what exists" and "run the business differently" (his Gmail‑vs‑Office line). The room asks for things we did not have to pitch: predict Prisma retention (NRR was 104–106%, dipped to the mid‑90s on one client), cross‑sell the graph, hear every call, read outside signals. Bill: "Because our legacy businesses are so big… it could be an accelerant like we've never seen." Agreed on the spot: a skunkworks with us inside and at most one other design partner; multi‑thread the Zvika conversation with NIVO as the likely home; feed the call recording into the system.
• 2026‑07‑24 · 13:07The Bill Wise explainer. A dedicated deck for Bill in the Konstant register: the calibrated price of the decision, independence is the moat, two surfaces one score, NIVO's agents optimize the how, own the contribution benchmark, calibration is not causation, you built the neutral spine. konstant-explainers/mediaocean-wise.html
• 2026‑07‑24 · 16:15Same‑day deck. Twelve slides built from the call transcript before end of day, live at precise-infrastructure-1cf027a2ba.surge.sh.
• 2026‑07‑24The 13‑board on‑site deck recovered and extended. Extended with the compounding‑network board. Live at prisma-deck-d2e07bb49e.surge.sh.
• 2026‑07‑28 · 10:00The NIVO discovery call: 61 minutes, and the political map in plain view. Adam, Spencer, Matt, Jordan (day two as CTO) with Bill, Zvika, Dan Zeevi (product lead, platform and the NIVO agents), and Cody Wymore (VP, NIVO Growth). Bill plays offense the whole hour: "can what you're doing and what we're doing on the NIVO orchestration side come together, where we can accelerate both our businesses"; with decisioning plus interoperability "DSPs and SSPs can go away… and the principals end up winning." Zvika plays defense, and says so honestly: handing orchestration to somebody else is "not gonna happen"; it is "the essence of… our future existence"; we read as "super early"; his offer is two contained entry points, a creative test with Dan Zeevi (benchmark: Alison.ai) and a low‑stakes tech session. Adam floats the governed containers (register capabilities, onboard legacy software, run workflows downstream); Zvika's door stays open: "if there's a specific technology or tool that will enhance what we're building, absolutely." Bill closes on Omnicom: "we're talking to the Blueprint team, you guys are talking to the Omni team… that's a way for us to attack and own Omnicom together." Transcript on file: ~/Downloads/NIVO + Precise AI Introduction & Discovery Call.txt
• 2026‑07‑30The deck that came out of it. "Building Enterprise Value Together" ships two days later: three workstreams with named first proofs, the Alison.ai answer, the substrate scope sentence, the Blueprint × Omni working session. mediaocean-artifact-v2-b83f41d7c9.surge.sh Same day: the iHeart deck ships (15:05) and customer‑questions publishes (20:28).
• 2026‑07‑31Substrate groundwork. Prepared for Jordan ahead of the technical session.
• July 2026The hub and the briefs. Internal hub for the Precise team (the tuned send‑to‑Bill email lives there), the shareable "Price · Govern · Settle" brief with PDF, and the internal network‑play brief with the below‑the‑waterline section.
• 2026‑08‑03This file. The research program folded together for the team ahead of the on‑site: three keyword lanes, then the Exa‑first correction pass and the screen‑anatomy pass, with the full asset inventory.

Asset inventory

07What we assume, and how each assumption gets tested

The research, operationalized. Every row: something we believe, the documented fact it stands on, the question that tests it, and the board that shows it. This table is the internal conversation.

Where a row survives Wednesday, it graduates from assumption to plan. Where it takes damage, the question column already says what to ask next. Either way the table gets updated, which is the point: this file works when it changes.

ROOMSWalking to the network

The network play as a habitable forecast, run as meetings. Each room: what is true inside it, the meetings that move us through, the pins that must fall (with the knock‑down move, the owner, and the early warning), and whose yes opens the next door.

The re‑rooming commitment, which is the whole discipline: this file updates after every meeting above. Outcomes become walls, projections re‑derive, and a fired pin becomes history with a date. We log what we project each room to take and grade ourselves as the actuals land. That grading is the product, practiced on ourselves first.

08Where we stand

The on‑site, the room, and what must be answered before we walk in.

The room, Wednesday

Ninety minutes with Prisma, this Wednesday. Confirmed attendees, with the part of the program each one owns:

The ninety minutes, the run sheet

Built from the room composition, offered to be torn apart before Wednesday. Each beat: who carries it, what is on screen, and which pin is live with the counter ready.

The workflow questions to bring

Grounded in the screen anatomy: each question opens a door the boards need. The first answer to most of them is a person's name and a date.

Held for the technical session with Jordan: Q04 (the placement‑to‑tag map) and Q11 (BYOA registration). Held for the Dan Zeevi thread: creative‑test scoping (assets, holdout, the Alison.ai benchmark).

Before the room: deck v2, /prisma, and the plan‑to‑pay demo open in tabs; the question list printed with owners marked; one of us on notes for same‑day re‑rooming, outcomes into this file before the day ends. After the room: the ledger: which pins fell, which questions got answers, which forks we took.

Success at ninety days stays as agreed: a recorded calibration track on real data, plus one deliberate holdout. A track record, not a slide. And the meeting succeeds if it ends with three calendar invitations: the retrospective campaign named, the technical session Zvika offered, booked for Jordan, the Omnicom working session booked with Bill and Spencer.

Verify before the room

• verifyIs the ad server open to a third‑party value tag, or locked to MRC‑accredited vendors?Decides whether the serving wedge is a product conversation or a policy conversation.
• verifyDefault‑on or opt‑in for anything we ship inside their workflow?Changes adoption math by an order of magnitude.
• priorityCalibration and holdout data rights across the $200B ledger.The biggest governance question in the whole deal. Who may learn from whose data, and what may the learning touch next?
• verifyInnovid vs Flashtalking naming inside the building.Small tell, real signal: it reveals which org we are actually talking to.
• verifyWhat do they want Precise to be: a NIVO feature, a co‑marketed currency, or an independent third party?Each answer names a different economic buyer. Kassan and Wise can shortcut this one; ask them directly.

On the last question, remember the capital map: Wise and Kassan both hold personal stakes in Precise. That is two shareholder‑advocates, one of whom occupies the seat built for making partner deals, reporting into an AI agenda Markman was hired to run. The sponsorship problem is smaller than it looks. The scoping problem (what, exactly, gets piloted on whose data) is bigger than it looks.

09The program, from first principles

Three workstreams, named on the Jul 28 call and fixed in the Jul 30 deck, planned together and run by one orchestrator. Not a menu: each creates value on its own, and they feed each other from week one.

Start from what Mediaocean actually is: the company is roughly 900 people on the last published count, carrying $200 billion of other companies' economic activity. Its true asset is the relationship graph: every agency, advertiser, seller, and ad‑tech vendor that transacts with every other one, through Prisma, on trust that Mediaocean is neutral. Today that graph records what happened. Every lane below is the same move at a different scale: make the graph record what things were worth, and the graph starts learning.

The undercurrent for every room: Bill wants to take the bet that lets Mediaocean control the next phase of media. Their internal path is good, and it should proceed exactly as planned. The flyer, the part only we can offer, is the thinking that turns their existing estate into that control: decisions priced, the legacy products deployed as capabilities, the network owned. Everything below is built to make that bet feel takeable. The politics are part of the terrain and we work them honestly: the growth conversation belongs to Bill and Markman; the Prisma work belongs to Kane and Carafello's lane; Zvika's lane belongs to Zvika, entered only through the doors he opens. And the clock is shared: Adam gave it about a year before the big platforms make these workflows point‑and‑click, and Zvika answered with his race‑car line: "I'm trying to drive a race car, there's a finish line… we have like five seconds to get this thing." Both sides feel the window. The program is how it gets used.

The Jul 28 call named the program's shape; the Jul 30 deck fixed it. Three workstreams, each with a named first proof, each with named people. What follows is those three as agreed, and then the part the deck could not say out loud: why they are one machine.

Workstream 1 · creative optimization (Zvika's literal ask, honored at the size he offered it)

Requested in so many words on Jul 28: creative is "at this point, where I have a need… something that is in my control, and I have a lot of data on." The thread runs with Dan Zeevi ("how a test would look like"), benchmarked where Zvika pointed us: Alison.ai. Our answer holds the layer line: they score creative elements and brief what comes next; we sit a layer under, because every impression the ad server places is a choice among eligible versions, and choices have prices. Scoring tells you what won. Pricing tells you what it was worth. Zvika also told us what makes it valuable to him: "contribution value calculated as a report is not interesting; it's more as a feed into optimization." The political read for our side of the table: this is a contained test offered by a protective owner, and that is fine. We execute it excellently, at face value, inside his lane, without overselling. First proof: one contained retrospective with Dan's team against an agreed holdout.

The loop Zvika described, mapped, with everyone placed on it:

The Alison read, for our side of the table

Is the pointer a setup to shut us down? The honest answer from the transcript: it is a sizing move with a cheap exit attached, and it is convertible. Three things are true at once. He raised creative himself as one of his two entry points and assigned Dan Zeevi by name, which is real. He also said "we have a team building something somewhat similar" and "maybe we'll say it's not worth it," which means the comparable doubles as cover for a no: if our test lands in Alison's category, his team can close it with "we build that." And the pointer is genuinely useful prep: he told us the bar his own people will judge against.

The conversion is the success measure. Alison scores creative and briefs the next one; so does his internal team; that slot is taken and we never contest it. The serve decision is unpriced by anyone, including Alison, including his team, and Zvika himself set the value bar that only our shape clears: "contribution value calculated as a report is not interesting; it's more as a feed into optimization." So the test is written so that an insights deck cannot satisfy it: priced serve decisions against a holdout, delivered as a feed his optimizer can act on. If that is the measure, the Alison comparable has no move against it, and Dan's own build gets stronger either way, which is what makes the yes easy.

Workstream 2 · the capability substrate under the estate (the actual goal)

This is the actual goal of the program. The position we are building toward is the governed capability substrate under Mediaocean's estate: legacy products and workflows wrapped by Boombox as agents and capabilities, registered once, callable everywhere, every action carrying the five‑field record. Everything else in this file is either a foothold that earns it or a consequence that follows from it.

The route runs through Bill and Markman, because it is a growth story and it is theirs. Bill said the ambition out loud on Jul 28: "recreate all the connections we have in Prisma and Innovid… in new architecture where agents can talk to agents, and we want to own that orchestration layer"; and further, with decisioning plus interoperability, "DSPs and SSPs can go away… and the principals end up winning." He is feeling for what is missing to get there at speed; the substrate is the answer, and it lands at the corporate and Prisma level, where Wednesday's room actually sits. Zvika's boundary is respected to the letter: NIVO's orchestration is "the essence of our future existence" and stays entirely his; nothing in this workstream requires his sponsorship, and everything in it makes his agents stronger (a richer capability roster, governance by construction, the record his compliance story will need). The door he opened stays open at the stakes he set: the low‑key tech session with Jordan, "show us, let's talk," where the governed containers Adam described ("register our capabilities, the NIVO capabilities… onboard legacy software into these little containers and have them run") get shown, engineer to engineer. First proof: wrap one Prisma‑side workflow on Boombox and run it governed, in the lane where our access is strongest. Bill previewed it on Jul 24.

Workstream 3 · Omnicom: Blueprint × Omni, through OMG

Bill named it on the call: Omnicom has "two competing things right now, the Blueprint and Omni… we're talking to the Blueprint team, you guys are talking to the Omni team… that's a way for us to attack and own Omnicom together." And the commercial line that moves the whole building came from Zvika himself: "if you say, look, this could get you to Omnicom, a five‑year renewal deal on media… you don't need to meet with me, just tell me what's needed, then we'll get it done." Business won is the one argument with no opposition anywhere in Mediaocean. First proof: one commercial working session, Bill and Spencer and the Omnicom relationship owners. The measure of success lands in the contract, and read against section 05, this is also the network's first economic edge.

Where the boards fit: decision pricing inside Prisma

The boards in part two are the priced‑decision capability, registered on the workstream‑2 substrate and pointed at the buy: decision value at authorization, calibration at actualization, signed records at settlement, governance under agents. This is what changes Prisma from the system of record into the system of judgment, and it is precedented (Guideline, Nielsen, and Basis prove the partner shape). Because the priced‑decision workflow is a registered capability on the workstream‑2 substrate rather than a Prisma feature, it moves: build it once against reconciled spend, drop it into the creative loop, point it at retention, extend it to Direct.

The network (the real size)

First principles again: the buy is not really an internal workflow. It is an economic relationship between two companies: an ask, an offer, an authorization, accepted work, an outcome, a payment. Prisma already carries all of those artifacts; it just carries them as paperwork. The federated version makes each relationship programmable while every company keeps its own context, authority, objectives, and economics: the agency's advocate and the seller's advocate exchange bounded, mandated, signed work; capabilities (Precise's valuation among them, NIVO's twelve agents among them) are the supply any edge can call; and every completed edge leaves the full record: who asked, who offered, who approved, what was delivered, what it was worth, who got paid.

In that world Mediaocean is the custodian of the network: the neutral party every edge already trusts, now holding the trust layer agentic commerce is missing. This is their own brand promise made technical: federation is what neutrality looks like when it is enforced by construction instead of by reputation. The orchestration layer at the top runs on Konstant's capability network, which Mediaocean gets without building; it sits above the Innovid level and does not disturb it. NIVO's agents are not displaced; they become network‑callable capabilities with the largest possible market: every counterparty on the graph.

Why the workstreams are one machine, not three projects

This is the part to plan deliberately, because it only works if it is planned together. The priced‑decision workflow, built once as a capability on the workstream‑2 substrate, drops into workstream 1 and captures the whole creative loop: creative decisions become priced decisions with calibration tracks, which is far more than calibration‑as‑a‑service. Workstream 1's records seed the network's first edges, and workstream 3 is the first edge: Blueprint and Omni converged through the graph. The substrate makes every workstream cheaper to run and impossible to silo. Run as three separate integrations, this is three vendor projects that never touch. Run on Boombox correctly, it is one orchestrator and one capability system: a workflow registered anywhere runs everywhere, and value created anywhere lands everywhere. All three start at once; each first proof is one conversation; the proof sequences itself.

The other side: what would be there

Concretely, for the Prisma reader imagining the after‑state. A buy is an edge between two advocates; the IO carries its mandate and its record. A discrepancy settles on value, with a number both sides can re‑run, instead of on an email thread. The retention question answers itself, because the graph knows which relationships are deepening and which are ready for attention. When holdco agents start transacting agent‑to‑agent (Omnicom's Wren is already piloting this to route around intermediary tolls), they transact through the Mediaocean network, because the mandate, the governance, and the record they need live there; the rail survives by becoming the trust rail. And Mediaocean's economics grow a third leg beyond seats and serving: network custody and capability certification, the Aquila fee rail generalized to every kind of value that moves.

Why Prisma carries the center of gravity: it is where every counterparty already meets, and it is where our access is best: Bill and Kassan hold conviction in this direction with their own capital, Markman owns the AI transformation the network play belongs to, and the room asked us for the retention workload unprompted. Creative optimization runs at the Innovid level from day one. The network grows from Prisma on the substrate and touches the Innovid estate only at the capability rail, which is why all three workstreams can run at once.

The deck's one‑line version for the room: Innovid measures what happened; Precise prices what it is worth; the network Mediaocean already owns socially becomes a network it owns technically. NIVO's agents then optimize against calibrated objectives, and the governance questions NIVO's launch left open get answered in the build rather than by policy memo.

What this week's research changed

The precedent is set, and it favors us. Guideline (planning), Nielsen (audiences), and Basis (execution) prove the pattern: Mediaocean signs deep bilateral APIs for capability it will not build. Partner certification is even structured by API category (ad serving, verification, social, ACP, DSP, ERP); "decision valuation" as a new certified category is a concrete, precedented ask.

Their own boundary is our opening. Kane, on the record about Prisma Direct: "Publisher decisioning remains within existing publisher and partner stacks." Digiday's gloss of the launch: workflow efficiency, accelerating rather than replacing execution and measurement. We open with that boundary and stand on their side of it.

The reconciliation wedge got stronger, with one hard constraint. Their own curriculum documents the planner‑to‑finance swivel chair (and a June 2026 IPG Mediabrands job posting still staffs it by hand), Prisma Finance already holds realized delivery beside committed spend at $150B a year, and IPS pipes $50B+ of payables underneath. But Guideline's exclusive API owns plan‑vs‑actual: our calibration is priced per decision, predicted against realized, which Guideline does not touch and cannot.

The governance gap is now public and being standardized. NIVO launched without audit trails, incrementality, or oversight rules; NIVO‑in‑Prisma is publicly on their roadmap; IAB Tech Lab is codifying agent audit trails through AAMP. The governed version of their own roadmap is the joint build.

Prisma Direct is a fresh surface until Q3. Signed decision records embedded at build time beat a retrofit; the Disney lane is the moment.

And retention is the door the room opened itself: they asked us to predict Prisma NRR on Jul 24. A calibration track on their own churn signal is a legitimate first workload.

Four ways in, one ask

The ask stays singular: point Precise at one slice of reconciled spend. Ninety days later there is either a recorded calibration track with a deliberate holdout, or there is not. Either way, everyone learns something real, which is the pitch behaving like the product.

Candor rules for the room

Say "tamper‑evident, signed, independently re‑runnable"; on‑chain anchoring is roadmap and we say so. The move‑level calibration write path is designed, not built; expect their architects to find that in the first hour, so we say it in the first five minutes. The data boundary today is declared‑and‑audited, not runtime‑enforced; that is their most likely first objection and it is also the first thing the skunkworks should build. Live proof is Blockdaemon in production plus the incremental‑reach optimizer running in the Omnicom engagement: one platform client, one media deployment, and we say exactly that. Every dollar figure in the demo is illustrative and labeled once. We charge a percentage of media, never a percentage of savings; the referee does not bet on the game. And we never compete with NIVO: the moat is worth building together, the sentence every readout to the Innovid side should carry.

Build order, forced by the deal, in case engineering asks what happens Monday: the calibration ledger write path first, the signed‑locator egress gate second, MCP registration into the Orchestrator third, the generalized find‑the‑move engine (beyond the shipped incremental‑reach case) fourth.

PART TWOWhere Precise lands in Prisma, board by board

The screen‑level walkthrough, inline. Reading the boards: everything in teal is Precise; everything else is Prisma exactly as it works today. Campaign and account data on these boards is illustrative. Standalone copy at /prisma.

SURFACE 01The buy: price the decision before the IO goes out

Prisma Media · Buy tab · the placement grid inside a campaign

• 1
  The decision-value column. One number per placement line: predicted contribution to the campaign objective per dollar, with its uncertainty band, computed before authorization. Reads like a rate, sits beside the rate, respects the line's Cost Method and Unit Type.route 02 · building
• 2
  The pre‑order panel. Prisma already gates the way out twice: financials before the order send ("Financial update pending. Update financials before sending this order") and client authorization before the ad‑server push ("Campaign must be authorized by client to send to ad server"). We add the third gate question: what is this order worth, and what one move improves it? The rebalance is priced, not suggested.shipped as the incremental‑reach optimizer · running in the Omnicom engagement
• 3
  The release gate. Nothing auto‑executes. The buyer releases the move; the Order‑view send (which already stamps order number, version, and Send date) seals the first three fields of the decision record: what was known, what was chosen, what we predicted. Surfaces 02 and 03 write the last two.governed release · shipped pattern

Native vocabulary on this board: Buyer workflow · Campaign ID (CP…) · C/P/E · placement / package (cube) / roadblock (road icon, created by starring a primary child placement) · Placement ID (P10…) · Cost Method (CPM/CPA/CPMV/Flat/Free) · Unit Type · Planned Unit Amount · Served by · order versions R0.01 / Rev 0.01 · buyer order states Not sent / Sent / Agency Accepted / Accepted / Revision requested · changed cells highlight purple, mandatory cells blue

SURFACE 02Actualization: the calibration track writes itself

Buyer workflow · Actualize Performance view inside Buy · planned and actual side by side, payable delivery by source, then "Update financials"

• 1
  Predicted / Realized / Band, per line. Actualization is the one place in adtech where committed spend and realized delivery already sit side by side, monthly, under variance rules, with the payable source chosen line by line. We add the third pair: what we said the decision was worth, and what it proved to be worth. No new plumbing; the daily delivery feed and the source picker already exist.route 01 · ready as read‑only value report
• 2
  The calibration track. Every month of actualization appends to a recorded track: predictions against outcomes, bands honored or broken. Ninety days in, this is a track record nobody can argue with; it is the agreed success test from the July 24 call.write path designed · the skunkworks build
• 3
  Variance priced in contribution, not face dollars. Prisma flags a −7.2% delivery miss and lays the sources side by side; the teal pair adds what the miss meant, priced. This is deliberately not plan‑vs‑actual reconciliation, which Guideline's exclusive API owns; it is decision‑level calibration, which nobody owns.boundary honored

Native vocabulary: Actualize Performance (view in Buy) · billable and payable lines · Actual > Payable · payable source: Planned / Invoice (weighted) / Invoice / Supplier / 3rd party · Redistribute all variances for <month> · Update financials · variance rules (agency / user level) · MOS · delivery data pulled daily

SURFACE 03Reconcile mode: History logs edits; this logs decisions

Buyer workflow · Invoices module · Reconcile mode, where OK to pay lives, with statuses and the tolerance gate verbatim from the Invoice Management quick card

• 1
  A record chip per placement line. Prisma already auto‑matches booked to invoiced and gates payment on tolerance. The chip answers the next question: did the flagged variance matter, and what did it teach the system?route 01 · ready
• 2
  The dispute gets a number. Today the tolerance gate says that sources disagree, never why or which deserves to win, and the investigation moves to email ("Review the Planner's workflow after a financial adjustment" is literally a training module). With the record, finance disputes on value.the reconciliation wedge · first workload
• 3
  The five fields, on their settlement rail. Every decision written down completely: what was known, chosen, predicted, what happened, what it taught. Signed so anyone can re‑check it, appended beside the History audit trail (Event / On / Modified by) that today records edits, riding the same close into Ignitia payables and the IPS rails.the product

Native vocabulary: Reconcile mode · OK to pay · Incomplete / Pending / Reconciled / Pending credit / Pending adjustment / Revision requested / Reversed · Actualized? · Approved source: Planned / Invoice (weighted) / Invoice / Supplier / 3rd party · allowable variance · Itemize invoice · credit memos and cash receipts always negative, adjustments always positive · Reverse reconciled · History (Event / On / Modified by) · Recommended for payment (Ignitia)

SURFACE 04The substrate: where the whole program is headed

This board is the goal. The estate running as governed capabilities on Boombox, every action carrying the record. Shown here through a NIVO run view because that is the world it serves; the position itself sits underneath, and NIVO's orchestration stays NIVO's.

• 1
  Agents optimize a calibrated objective. Today an optimization agent chases a dashboard. Under the record, it chases the same calibrated contribution track Surfaces 01–03 write, so "90% faster" means "90% faster at the right thing."lands Prisma‑side; shown to NIVO engineering in the session Zvika offered
• 2
  Numbers reach reports through the registry only. The measurement agent can only speak numbers that exist in the record; it structurally cannot write a number into prose. We run this pattern in production today.trace‑talk pattern · runnable now
• 3
  The POV shift. The likely house view runs new agents on the new stack, with legacy as the install base. The bet runs the other way: Boombox wraps the legacy products as agents and capabilities, NIVO registers them, and the twelve agents get the deepest toolset in media on day one. Governance (autonomy policies, release gates, the exportable trail) arrives with it, by construction. Their roadmap stays good; this is the flyer on top of it.the session Zvika offered · lands with Bill

Native vocabulary: NIVO thinks, Orchestrator connects, Agents act · twelve agents · Create / Deliver / Measure / Optimize · Taxonomy agent, Campaign Trafficking agent, Campaign QA agent, Decisioning agent · "notify a human" workflow steps · Bring your own agents (BYOA) · Hypermode / Standard Mode

SURFACE 05The workload they asked for: predict Prisma retention

The skunkworks console · run on Mediaocean's data about Mediaocean

• 1
  Renewals priced as decisions. Same math as Surface 01, different asset: predicted outcome, band, the single driver, the single move. Calibration on churn signals is the fastest possible proof loop because renewals actualize on a known calendar.skunkworks candidate · zero client‑data exposure
• 2
  The trust ladder. Prove calibration on their own book first; earn the reconciled‑spend slice second; govern NIVO third. Each rung writes records the next rung stands on.sequencing per the Jul 24 agreements

SURFACE 06The connection map: how we physically get in

Six ways in, from zero‑permission to joint build. What makes sense, where, and why.

The sequence that makes sense: the Boombox skunkworks trial runs from day one on the portfolio seat we already hold, the fastest full‑fidelity proof of the whole stack. Beside it: the value report starts the calibration track this quarter, the invited overlay makes it visible in their workspace during the pilot, the certified API is the deal, BYOA is the NIVO lane entered through its own door, and the settlement rail is the compounding end state.

And the template travels. The method ships whole (backstory file, screen‑level walkthrough, connection map) and is the repeatable play for every platform partner. iHeart is next: same record, same wedge logic, pointed at their ad platform, shows, and attribution surfaces instead of Prisma's grids.

SURFACE 07The network: what the whole thing becomes

Boards 01–06 are the seed. This is the plant: Prisma's relationship graph made programmable, Mediaocean as custodian, every company keeping its own authority.

• Why everyone in the room wins. Agencies keep their authority and their client relationships. Sellers keep decisioning on their side (their own stated boundary). NIVO keeps its agents and gains distribution. Mediaocean keeps neutrality; the network is neutrality, made enforceable.the room‑safe version of the biggest idea
• Why it starts at Prisma, not Innovid. The ask already on the table (calibrate creative) is well‑formed and small; say yes and treat it as the demo. The network needs the place where every counterparty already meets and where our access is strongest. That is Prisma.sequencing per the File's Four ways in table
• What Boards 01–06 have to do with it. Every amber element earlier in this document is a network primitive in disguise: the priced decision is the edge's offer, the calibration track is the edge's trust, the signed record is the edge's settlement, the release gate is the mandate. Build the boards, and the network is what they compound into.compounding