Social Beat is excited to launch its updated version of the AI-first media & programmatic solution, PRISM. It now combines Media Brain & SB Studio. One plan. One frequency cap. One set of reconciled numbers — and a creative intelligence that rebuilds the ad from the same performance data.
The problem we built for
A brand buys YouTube on one platform, JioHotstar on a second, a smart-TV OEM on a third, and the open exchange on a fourth. Every auction treats the same household as new. Nobody can say how often one viewer saw the film.
We can now put a number on that. Before we moved accounts onto a single plan, the average user in an Indian video buy saw the same ad about 15 times. Six is enough. Everything else was paid for twice.
Creative runs in its own silo alongside it. The media team knows which ad is failing by Wednesday. The creative team hears about it the following Monday, rebriefs, and ships a variant ten days later — by which point the flight has moved on. Both teams are competent. The workflow is the problem.
Google's and Meta's AI have absorbed most of the targeting control an agency used to add on top. What they have not absorbed is the join across platforms
PRISM — Programmatic Reach and Impact Solution for Marketers — is the buying, intelligence, and creative layer that sits above the DSPs, the ad server, and the publisher deals we trade through.
It is not a DSP. We do not want to be one. Premium DSPs, the ad server, and the publisher relationships stay where they are; PRISM is the layer that makes them behave as one plan, with one audience taxonomy, one frequency cap, and one measurement frame.
It has three working parts.
The three share one store of brand rules, past creative learnings, and performance history — the Brand Brain. That shared store is the reason a creative decision and a media decision can be made on the same evidence in the same hour.
Inventory in the plan spans YouTube, Samsung Ads, Google Display, JioHotstar, Amazon Prime, premium OTT apps, smart TV operating systems, the open exchange, retail media and programmatic out-of-home, traded through Display & Video 360 and ad-served through Campaign Manager 360. Since Cheil South West Asia acquired Social Beat in November 2025, Samsung connected-TV inventory comes to our brands with no resale chain in between.
The mechanism is unglamorous. One ad server across DSP, YouTube, publisher direct and retail media. One audience taxonomy. One cap that holds across all of it at the same time. A user counted once is counted once — without that, deduplicated reach is an estimate and a frequency cap is a claim.
What it has returned across the accounts we have migrated:
| The buy | What changed | Result |
|---|---|---|
| Large FMCG player: three video platforms bought separately | Consolidated into one plan, one cap, budget held flat to keep the comparison honest | 372,000+ incremental unique users. 17% of media cost saved on an unchanged budget. |
| Private life insurer, retirement product | YouTube, premium publishers, display and CTV as one plan; brand lift measured across every exchange in the buy | +14% incremental reach — 6.6 million users the TV and platform buys missed. ₹43.4 lakh media saved. +7.7% ad recall, +5.3% awareness. Google published it as an external case study. |
| D2C health brand, launch | CTV-led across smart TV OEMs, OTT and YouTube, deduplicated in one plan; quick-commerce cohorts closing the funnel | 82M+ incremental reach. 29.1% cross-exchange brand lift on CTV. 106% lift on branded search. 13% media saved. |
| Jewellery retailer, Diwali | 50+ line items across 6+ platforms under one cap | 61.7M reach against 38.1M planned. CPM ₹79 against ₹117 planned. ₹36.9 lakh saved. |
Average media saving across migrated accounts is 14%. Average frequency fell from 15 exposures to six. 72% of retained media brands now trade on unified plans.
Most Indian programmatic targeting still runs on broad platform interest categories, or on online behaviour read in isolation. Interest is not intent.
PRISM's audience layer combines aggregated third-party audience intelligence from multiple partners, including Vserv AudiencePro, PayU, Razorpay, Pine Labs, Zepto, and Swiggy based on observed commerce behaviour.
All of it maps to one taxonomy, as privacy-safe, activation-ready cohorts, usable across CTV, video, display, the open web and retail media. The practical difference: we can separate someone who reads about a category from a cohort that is transacting in it this fortnight, and we can connect online media planning to offline buying behaviour without touching raw personal data.
Programmatic was the first problem. In performance media, the ad platforms now make most of the targeting calls, and each one marks its own homework. Spend sits in the ad accounts. Leads sit in the CRM. Neither can say which campaign produced a lead the sales team would call.
Media Brain is the planning and daily tuning engine inside PRISM. Every night it turns the brand's own numbers into a short list of things worth doing today, each with its evidence attached.
Spend comes straight from each ad platform's own API. Leads come from the brand's CRM through a direct connection, or from a recurring lead report where the CRM has none. The two are joined every night, campaign by campaign and down to the city, project, or product line the brand plans by.
What counts as a qualified lead differs by brand. For one, it is a consult booked; for another, a site visit. That definition is written once into the brand's Brand Brain and applied the same way every day, so cost per qualified lead means the same thing this Monday as it did last quarter. Platform-reported CPL still shows. It stops deciding where the money goes.
Much of the work is in small rules that stop a number from misleading anyone. Leads are counted whole, and cost per lead is divided by the whole count, so the screen never shows a fraction of a lead beside a seven-figure CPL. Branding campaigns stay in the spend total but get no performance recommendations; judging an awareness film on cost per lead gets the answer wrong. Where a campaign was built to buy website conversions, those conversions count as its leads. A missing CRM feed shows as “no data”. It never shows as zero.
Before any number reaches a screen, spend and conversions are reconciled against the platforms. When a window does not tie out, the dashboard says the data is being verified and holds the figure back. There is no sample or demo data anywhere in the system. If a number is on screen, it came from the brand's own accounts.
The refresh runs at 02:00 IST. The dashboard opens on ranked recommendations; charts come after. Each recommendation carries its cost in rupees on one weekly scale, so a pacing slip and a lead-quality gap compete directly, and the biggest leak sits on top, whatever caused it. The brief only changes when something has actually moved since yesterday.
The reads it runs today:
| The read | What it tells the brand |
|---|---|
| Plan against achieved | Where each line of the media plan stands against target, pro-rated to the day. Verdicts built on too few leads are marked thin instead of called. |
| Change impact | The week before and the week after every bid, budget or targeting change the team made. Overlapping changes are flagged; changes too recent to judge are marked as maturing. |
| Spend pacing | Whether each campaign is running ahead of or behind its budget for the month. |
| Auction share | Campaigns losing impressions because the budget runs out, Performance Max included. |
| Search terms | Queries eating spend without converting, by project, with an export for the search team. |
| Theme shift | When the message mix in live ads moves. The shift is tested for significance before it is reported. |
| Video drop-off | The point in each film where viewers leave. |
| Asset laggards | Google ad assets pulling their campaign down. |
| Stagnation | Campaigns that have stopped improving and need a fresh test. |
The media plan goes in the way the account team already writes it. The first upload teaches the system the layout; after that the team corrects it in a preview, or uses one standard template. Every insight is then read against that plan. A cost-per-lead rise on a line still beating plan is demoted. A lead-quality gap is costed in the qualified leads it is missing. A line with no plan says so, and the system does not guess a target for it.
A language model reads the day's data and writes the narrative. Code then checks it. Any insight the model adds has to cite a field and a value that exist in that day's data, and a narration that introduces a number of its own is thrown out. If the model fails, the rule-based read publishes without it. Five correct calls beat eight plausible ones.
Every recommendation also needs a person. Approve, snooze or reject, with a written reason each time. Rejected ideas are demoted, never deleted, and when the same pattern returns the card shows what was decided last time and by whom. Approval records the decision; the account team still makes the change in the ad account. Nothing edits a live campaign unattended.
Ask PRISM is a copilot on the same reconciled data. Ask it why qualified-lead cost rose in one city last week, and the answer comes from the dashboard's own numbers. The verification gate applies here too. If a window has not reconciled, Ask PRISM will not answer from it, and every question is logged.
Examples of how we have brought this to life. On a healthcare brand, joining media spend to admissions data through the brand's CRM exposed spend under-reporting in two cities that had been invisible in platform reporting. On a lead business, we push CRM-qualified outcomes back into bidding through server-side conversion APIs rather than raw form fills. In insurance, that is the difference between optimising to a lead and optimising to a policy.
Media Brain scores creative where the budget is decided: bottom-of-funnel ads first, judged against the account's own cost per lead. An ad without enough data reads as unknown. It is never scored as a zero. Every ad is tagged by format and by message theme, and the themes are read from the ad copy each night instead of being typed in by hand.
The tab opens on the playbook: which themes and formats are carrying the account right now, rebuilt from live ads every night instead of written up once a quarter. Beside it sits a coverage grid of every project against every format. Green means a creative is working there. Amber means one is live but tiring. Red means nothing exists, a project with no video, say. The scorecard then names what to scale and what to retire, and for video it shows where viewers leave.
Competitor scans run from the same screen. A planner names a rival, and the scan reads that brand's live ads, video openings, and audio hooks included, then returns the gaps and hooks worth acting on with a workbook to download. It takes under half an hour and needs nobody outside the account team.
With no history in a format, suggestions fall back to aggregated category patterns (never another client's creative or numbers), then to shared creative principles. Each says which tier it came from.
The metric this is built to move is creative hit rate — the share of assets that clear the account's own CPA benchmark. Not volume. Volume is easy now with our AI-first platform, SB Studio.
Nothing publishes for a brand until its Brand Brain is written: the qualified-lead definition, the brand's rules and tone, the category it plays in. Teams without direct access to the AI tools fill out a downloadable form instead. Category knowledge for healthcare, real estate, and lead-generation funnels sits one layer above and is shared as principles. It never carries another client's data.
A new brand goes live through a setup wizard. The team finds the ad accounts by name or ID, connects the CRM and adds the Brand Brain, and the wizard returns a readiness verdict before the first run. Once an ad account is saved, the brand joins the nightly run on its own and its history backfills on the first night. A publish guard stops an older build from overwriting a newer one.
Access is by sign-in, with three permission tiers. Every client-facing screen passes a naming check before it deploys, so no third-party tool name reaches a brand. Pages run under an enforcing content security policy, and every release has to clear the full automated test suite.
Several pieces are built and in testing. Hourly WhatsApp alerts will flag a campaign breaking during the working day. A new read will show how upper-funnel spend moves overall growth. Change impact will be measured on qualified leads, not only platform leads. After that, one-click apply for an approved recommendation, brand logins, the brief delivered by email, and read-only machine access so a brand's own martech stack or AI agent can query its numbers.
Once one plan can test twenty variants a week, the creative line becomes the constraint. SB Studio removes it.
Brief in, governed creative out: statics and video, generated through a fully AI workflow that meshes creative best practice with each brand's own Brand Brain — its rules, its tone, its past winners. With over 5,000 creatives out of the system until now, adaptation time has fallen by roughly 75%.
Two examples of what the volume buys.
A two-wheeler brand's riding gear. We built AI video down two routes and ran both: model-led lifestyle films, and product-centric films where the shoe, glove or jacket is the hero with no model at all. The gear-led films returned about 3% click-through against roughly 1% at best for lifestyle. We paused the lifestyle concepts mid-flight and moved the budget. The read is specific to shopping placements, where the buyer is already looking at gear — the product is the hook, not the person. Lifestyle AI on those placements looks good and gives the viewer nothing to act on.
An FMCG major's programmatic creative work won Bronze for Best Programmatic Creative Innovation at the Programmatic Asia Awards 2026 — creative testing inside the media plan, judged as media innovation.
This is the part that only works because the three pieces share a brain.
Media Brain reads a creative as underperforming and names the attribute — the hook is too slow, the brand lands too late, the promise is wrong for that cohort. SB Studio takes that reading against the same Brand Brain and regenerates the variant. The new asset runs inside the same plan, under the same frequency cap, measured in the same frame. No rebrief email. No ten-day gap. No argument about whether the media or the creative was at fault, because both are reading one set of numbers.
Creative and media stop being two workflows. That is the whole design intent, and it is why we resisted buying a creative-intelligence product and a media product and a reporting product separately.
There is a broader reason. Every specialist AI tool in this category has a commercial incentive to keep the user inside its own interface — that is how it gets renewed. An agency has the opposite incentive. PRISM's defensibility is not that it analyses better than any single specialist tool. It is that PRISM is the only layer in the chain with a reason to stitch everything together, with specialist tools plugged in behind it as read-only inputs. The moment creative intelligence lands in a spreadsheet, it dies.
At the Programmatic Asia Awards 2026 the work took 18 awards — seven Gold, four Silver, six Bronze — the highest count of any agency at the show, plus Best Programmatic Strategy Team (Mid-Size Agency) and Gold for Programmatic Agency Innovation of the Year. Five more followed at the Connected TV Asia Awards 2026 across three campaigns.Google has published two of these campaigns as external case studies too.
This is just the start of the AI Level Up Journey at Social Beat