Hawky × Uni Cards · Solution & pilot proposal · September 2026
Prepared for Hitesh, Akansha and Sai
01 · The scope we aligned on
These are the four you named on the call. They are not four tools. Each one hands its output to the next, and the fourth returns evidence to the first.
Use case 01
Read why a creative works at element level, then say which live creative to replace and with what.
Use case 02
Recommendations with the action attached — pause, reallocate, scale — inside limits Uni sets.
Use case 03
What the category is scaling, decoded the same way as yours, produced and queued against a weak slot.
Use case 04
Script, production, approval and launch as one tracked lifecycle instead of four inboxes.
The intel names the weak creative, the agent decides what to do about it, the competitor read supplies the angle, and production ships the replacement. Run any one of these alone and it stalls at the handover. Run them together and the cycle closes without a meeting.
02 · Where the cycle breaks
Nothing in the current stack is broken on its own. What costs Uni money is the distance between knowing a creative is tired and having its replacement live.
The read
Reporting says which ad won. It does not say which hook, which offer framing or which opening three seconds carried it, so the next creative repeats the ad rather than the reason.
The decision
Fatigue arrives on its own schedule. A creative that started sliding on a Tuesday keeps spending until someone opens the deck on Friday.
The angle
A rival scaling a new angle is visible in the ad libraries days before it shows up in your own cost per application. Nobody has time to look every morning.
The approval
In a regulated category the script is quick and the sign-off is slow. Brand, legal and compliance run over mail and chat, and the version that ships is rarely the one that was approved.
Every day between the signal and the replacement is paid for at your current cost per approved application. The lever is not a better report. It is a shorter distance from signal to live creative.
03 · Use case 01
Every creative is decomposed into 100+ elements and joined to your KPI, so the finding is a conditional pattern rather than a score on a whole ad.
Vision models read the frame — card visual, presenter, colour, on-screen text, pacing, cut structure. Language models read the copy — hook type, claim type, offer framing, CTA, language.
Elements are correlated to your KPI within placement, audience, objective and time window, never pooled across them. A pattern carries the conditions it was true under.
Fatigue is flagged per element before the cost line moves. The output names the live creative to retire and the pattern its replacement should carry.
Creatives and their results pulled into one place, joined to the KPI you actually buy on — cost per approved application, cost per activation — not to impressions.
Each video carries the second its hook, reveal and CTA land on, so drop-off is read against what was on screen at that moment.
Winners and losers are grouped by what they have in common. This is how thirteen weak creatives turn out to be one creative shape made thirteen times.
Scale this pattern, test that one with a single variable changed, retire the shape absorbing budget. Each line carries the evidence it came from.
We condition, we separate account-wide events from creative signal, and we show competitor activity alongside. We do not claim to quantify how much of a shift was seasonality versus creative versus auction. Structured in-market tests settle that, one variable at a time.
04 · Use case 02
The performance agent does not send a report and wait. It holds a number, drafts the way to it, and acts inside the conditions Uni signs off.
Goal agent
Give it a target — cost per approved application, CTR, ROAS — and it works the account toward it daily, with the hypothesis behind each move written down.
Proactive agent
Runs on a schedule and surfaces what it found: fatigue starting, a CTR spike worth spending into, audience saturation. Delivered to Slack or mail, not buried in a dashboard.
Rule agent
Graduation from test to scale, pausing below a threshold, reallocating between ad sets — on conditions you configure, with every action logged against the rule it applied.
Control
Each agent runs in recommend-only or execute mode, per campaign. During the pilot every agent starts in recommend-only and moves up when you are ready.
The agent closes the gap between Tuesday's signal and Friday's meeting. The team keeps the decisions worth a human; the agent takes the ones that only needed to be timely.
05 · Use case 03
Competitor creatives are indexed from the public ad libraries and creator collaborations, then taken apart with the same element taxonomy as yours — so you compare element to element, not screenshot to screenshot.
Meta, Google and YouTube ad libraries plus creator and influencer collaborations across the card and credit category. Refreshed on a fixed cycle, not on request.
Ranked on impression velocity, longevity and launch pattern. A rival scaling a new angle or winding one down is flagged before it shows in your numbers.
The gap is returned as elements you are not running: an offer framing, a proof device, a hook structure. That gap is the brief.
Not a copy of their asset. The winning structure rebuilt on your brand kit, your claims and your compliance rules, and queued against the weakest live creative it is meant to replace.
Public ad library and creator data, plus modelled spend and share of voice. No paid third-party targeting data, and no claim to see a competitor's actual spend.
06 · Use case 04
The whole lifecycle sits in one place: a script written from evidence, produced on the brand kit, routed for sign-off, and launched — with every version and every approval held against the asset.
A winner of yours, a competitor angle or a brief becomes hook, beats, on-screen text and CTA — per persona, language and placement, each ranked by the evidence behind it.
Statics and every aspect ratio are generated on your brand kit inside the platform. Video is produced by Hawky's in-house production team working in the same workspace; automated first-cut video is on the roadmap.
Your rulebook — brand kit, approved claims, never-say list, mandatory regulatory phrasing — is checked at generation, and anything that fails is held. Brand, legal and compliance approve in the platform or in Slack, and the trail stays with the asset.
The approved asset is launched to Meta or Google from inside Hawky, against the slot the intel named — and goes straight back into the read as the next data point.
Not the approval — the chasing. Every generated output explains which rule it obeyed, so the reviewer is checking a decision rather than reconstructing one. Audit becomes a report instead of a search through mail.
07 · The system
Each use case hands its output to the next. The fourth returns evidence to the first, and the loop tightens every month it runs.
Element-level fatigue on a live creative, or a rival angle scaling in the category. Both arrive as a pattern with its conditions attached.
The agent sizes it against the goal it holds and either recommends or acts — pause the weak slot, hold the budget, brief the replacement.
Script, production, compliance check, approval, launch. The new creative goes live against the slot the signal named.
Every creative, result, rule and decision is kept per brand in FeatherDB. The learning stays when people move, and the read in month three starts from everything months one and two proved — not from zero.
08 · The questions you will ask
A card brand cannot run a creative engine that treats compliance as a review step at the end. Here is how each of these works, before you ask.
Compliance
Approved claims, banned comparisons, mandatory disclosures and partner-bank phrasing are configured as a rulebook. Output that fails is held before a person ever sees it.
Data
We work with creative assets and campaign performance. No customer-level data, no application data, no PII is required for any of the four use cases.
Control
Every agent starts without execution rights. Autonomy is granted per campaign, on your thresholds, and every action is logged against the rule that triggered it.
Effort
Account connection, historic decode and rulebook configuration are done by our team in week one. The load on your side is the brand and claims review, once.
The first reliable hypotheses take about three to three and a half weeks on a new account. That is why the pilot runs two months and not one — month one builds the read, month two acts on it.
09 · Proof
Every capability in this deck is live in client accounts. The numbers below are ours to stand behind; the case studies are published and linked.
Case study · Fintech
Case study · Marketplace at scale
“It's not just the quantity I love, it's the quality. I uploaded my brand book once and every ad speaks the language we want to speak.”
Pooja Jena · Brand Marketer · Cars24 — 1,000+ creatives / month · 15–20% lower CPA10 · Pricing
One subscription covers all four use cases. Both numbers are on this page, so the pilot decision is not a decision about an unknown price later.
Months 1–2
Per month, per brand · ₹1,00,000 for the two months
Each month includes
Who works with you
Customer Success and Creative Strategy work alongside your team on implementation and the day-to-day. This is not onboarding and then a handover to a dashboard.Included in the subscription
| What runs each month | Included in the pilot | Covers |
|---|---|---|
| Creative analyses | 500 | Your own creatives, static and video, element level |
| Competitive creative analyses | 650 | Category and competitor creatives, static and video |
| Goal agents | 20 | Running simultaneously, each holding a target |
| Proactive agents | 20 | Running simultaneously, on a schedule you set |
| Copilot queries | 145 / month | Ask across your creatives, results and the competitor set, on the top models |
| Customer Success & Creative Strategy | Included | Alongside your team for the full two months |
| Subscription · pilot | ₹50,000 / month | Per brand · ₹1,00,000 across the two months |
| Subscription · month 3 onward | ₹90,000 / month | Per brand · allowances and term set at the review |
Month 3 onward
Per month, per brand
What changes
Creative production · monthly
Per month · ₹365 per creative
Video production · per batch
Per batch of 50 · ₹2,500 per video
The price covers the platform, the agents, the analysis that trains the engine on Uni's brand, and two Hawky teams working next to yours. The setup analysis is not metered, because it is what makes everything after it accurate.
11 · The pilot
The pilot is scoped so that success or failure is visible without argument. Here is what runs, week by week, and what we agree to measure it on.
Meta and Google connected, your live and recent creatives decoded, competitor set agreed, and the rulebook configured with your claims and mandatory phrasing. Our team does this.
Element-level patterns on your own account, the competitor gap list, and the first scripts through the approval cycle. Agents run in recommend-only.
Produced creatives replace the slots the intel named, one variable at a time at equal spend. Agents move to execute on the campaigns you nominate.
We present against the criteria below. If they are not met, you do not continue — that is the point of pricing the pilot the way we have.
Criterion 01
A count of creatives produced in the pilot that beat the incumbent on your primary KPI.
Criterion 02
Measured against the same audience and placement, not against the account average.
Criterion 03
The number you actually buy on, over the last four weeks of the pilot.
Criterion 04
Days between a fatigue flag and its approved replacement going live. This is the one the loop is built to move.
The four criteria are agreed. The thresholds are not, and we would rather set them with you than assert them here. We will fix the numbers together in the kickoff, before week one starts.
12 · Next steps
Confirm the pilot and nominate one brand and its ad accounts. We will have the historic decode and the first competitor read in front of you inside seven days of access.
From Uni
Meta and Google ad account access, the brand kit and claims list, and one person who can approve creative.
From Hawky
Connection, historic decode, competitor set and rulebook configured by our team in week one.
Together
Agree the competitor set, the primary KPI and the four pilot thresholds.
Then
Measured against the criteria, with the decision to continue made on numbers.
Confirm the two-month pilot and give us ad account access this month, so the first read lands before the next campaign cycle rather than after it.