AI Agents · The suite

Six AI agents. One category engine.

SOMIN's agents each run a marketing job end to end -- strategy, paid media, influencer & GMV, social, CRM, and reporting. Every one reasons from what's actually winning in your category, and every one keeps a human at the gate.

Book a call See the platform

P&GAudiFerrariMothercareMcDonald's

AI agents, one engine

13min

Brief to pre-scored concepts

Human

Sign-off at every gate

Runs on SOMONITOR

Every agent runs on SOMONITOR, SOMIN's real-time category engine -- so each one reasons from what's landing in your category right now, refreshed live, not a generic model or last quarter's survey. One engine underneath; six jobs on top.

The suite

Six agents. One end-to-end loop.

From deciding the angle to scoring the ad, briefing the creator, running the inbox test, posting daily, and reporting it back -- each agent owns a stage of the loop, and hands off to the next.

Decide the angleLive

Strategy

Reads the creative winning in your category, maps the whitespace, pre-scores the concepts, and hands you a client-ready brief.

Explore the agent ↓

Score before spendLive

Paid Media

Predicts how an ad will perform and builds a fresh, on-brand funnel -- every concept scored for click-through before a penny of media moves.

Explore the agent ↓

Grow GMVLive

Influencer & GMV

Gives every creator an ownable territory and a brief in their own voice, tuned to lift affiliate sales -- not likes.

Explore the agent ↓

Post daily, on brandLive

Social Media Manager

Finds what's landing, writes it in your voice, generates the image, and publishes across every channel -- autonomous or gated.

Explore the agent ↓

Ship the winning emailPreview

CRM & Email

Generates thousands of category-derived email strategies, angles each to the individual, and runs the test-to-winner loop.

Explore the agent ↓

Always-current reportingComing

Performance Reporting

Turns live category signal into the client's own self-serve dashboard -- always current between reviews, in the client's language.

Explore the agent ↓

Agent 01 · Strategy · Live

Decide the angle from what's winning.

The Strategy agent reads the creative actually winning in your category -- competitor ads decoded for proposition and tension -- maps the whitespace, pre-scores the concepts, and hands you a client-ready deck in the brand's own look. Median 13 minutes from brief to a stack of pre-scored concepts.

See the Strategy solution → Compare vs ChatGPT, Perplexity, Meltwater & GWI →

Speed

13min

Median brief to a stack of pre-scored concepts.

Coverage

4k+

Competitor ads decoded per category.

Output

Deck

Client-ready, in the client's own brand.

Grounding

Livecat.

Your category, not the open web.

Agent 02 · Paid Media · Live

Score the ad before you spend.

The Paid Media agent reads which live ads are winning for your goal, decodes the winning concept and the category's paid whitespace, generates a fresh on-brand funnel, and scores every concept for click-through before a penny of media moves. Top-scoring ads are pushed into your campaign built paused -- nothing activates without your sign-off.

See the Paid Media solution →

Accuracy

78%

CTR-prediction accuracy across benchmarks.

Baseline

3×

Roughly three times the human baseline.

Training

10k+

Labelled ads behind the SoCraft model.

Method

Peer-rev.

Published at ACM WSDM 2023.

Agent 03 · Influencer & GMV · Live

Brief every creator on what lifts sales.

The Influencer agent ranks your roster by real GMV, gives each creator an ownable territory with zero overlap, decodes why the top sellers sell, and auto-builds a bespoke brief in each creator's own voice -- tuned to grow GMV, not likes. Proven on a top-3 global beauty roster.

See the Influencer & GMV solution →

Cohort ranked

1,047→6

Real sellers identified across the roster.

Territories

Ownable lanes mapped, with zero overlap.

Whitespace

31%

Of the top seller's audience in an unclaimed lane.

Per brief

Days→min

Built automatically, in the creator's voice.

Agent 05 · CRM & Email · Preview

Know the winning email before you hit send.

The CRM agent generates thousands of category-derived email strategies per setup, angles each message to the individual, builds the sequences in your sending platform, and runs the whole test-to-winner loop -- with two human approval gates and a sub-60-second kill switch. In preview today.

See the CRM & Email solution →

Scale

7k+

Distinct strategies generated per setup.

Control

Human approval gates before anything sends.

Safety

<60s

Kill switch pauses the whole loop.

Method

Peer-rev.

Personalisation research at ACM WSDM 2021.

Agent 06 · Performance Reporting · Coming

Your client's report, always current.

The Reporting agent turns the same category signal into the client's own reporting view -- a live, self-serve dashboard, in the client's language, with period-on-period comparison. The engine that collapses reporting time runs today; the productised dashboard is what's next.

See the Reporting solution →

Reporting

9×

Faster reporting on SOMONITOR.

Analysis

18→3hrs

Per cycle, cut from a day-plus to an afternoon.

Freshness

Live

Always current between reviews.

Access

Self-serve

A view the client opens themselves.

Control

Agentic -- but you hold the gate.

These agents propose; a person signs off. Autonomy is a setting you choose per agent, never a default you're stuck with -- and predictions are always shown as predictions.

Nothing irreversible happens without a human. The Paid Media agent pushes ads built paused until you approve the spend. The CRM agent needs two approvals and stops the whole loop in under a minute. The Social agent runs on its own only if you say so -- otherwise it waits at a checkpoint. Agentic doesn't mean unattended.

Paid Media

Paused

Ads pushed built paused until you approve the spend.

CRM & Email

2gates

Two human approvals before anything sends.

Kill switch

<60s

Pauses the whole loop in under a minute.

Social

Yourcall

Fully autonomous, or a sign-off checkpoint.

Under the hood

Named models. Published research. Verifiable claims.

Every agent runs on Google Cloud, where Gemini 2.5 Pro powers large-scale ad understanding and idea generation, and SOMIN's in-house SoCraft models predict creative performance. The architecture is peer-reviewed at ACM -- the same claims you read here are documented in the ACM Digital Library and Google Cloud's customer story.

LLM core

Gemini 2.5 Pro

on Google Cloud

Processes competitor ad and video content, clusters audiences, and generates insight -- the engine behind the 13-minute brief.

Google Cloud customer story →

Prediction

SoCraft CTR models

In-house deep learning

78% CTR-prediction accuracy, roughly 3× the human baseline, trained on tens of thousands of labelled ads.

Retrieves category content more accurately than general-purpose LLMs -- nDCG@10 0.83 vs .75 / .77 for ChatGPT and Gemini.

ACM MM '25 platform paper →

Explainability

XAI + LLM layer

Peer-reviewed at ACM

Every recommendation traces back to the ads and engagement signals that produced it -- the "Against Opacity" architecture.

ACM MM '23 paper →

Read the research Google Scholar · Prof. Aleks Farseev ACM DL · SOMIN Platform Paper Google Cloud · Customer Story

Two playbooks

One agent suite. Two ways to win.

Brand teams put the agents on their own categories and campaigns. Agencies deploy them across clients, white-labelled. Same engine, two motions.

For brands

Run every function on live category signal.

Marketing, brand, performance, CRM, and social leads put each agent on their own category -- deciding angles, scoring ads, briefing creators, and posting daily, all before the spend commits.

Strategy in 13 minutes, not three weeks
Paid concepts scored before media commits
Creator briefs tuned to GMV, in each creator's voice
A daily on-brand social presence across every channel

See the brand playbook →

For agencies

Deploy the agents across every client.

Strategy, creative, performance, and account teams at Dentsu, OMG, GroupM, Havas, and Publicis run the agents client-by-client, white-labelled, to pitch and deliver with unfair speed.

Reusable across clients, white-label deployment
Analyst hours per pitch cut by 75%+
Always-current, self-serve client reporting
Higher win rate on pitches that lead with live data

See the agency playbook →

Questions

What teams ask about the agents.

What is an "AI agent" in SOMIN?

Each agent owns a marketing job end to end -- deciding the angle, scoring the ad, briefing the creator, running the email test, posting daily, or reporting it back. They're not chat assistants: they read live category signal via SOMONITOR, do the work, and hand you an output ready to act on -- a deck, a funnel, a brief, a sequence, a dashboard.

Which agents are live, and which are coming?

Strategy, Paid Media, Influencer & GMV, and the Social Media Manager are live today. The CRM & Email agent is in preview -- the engine runs, the guided approval flow is rolling out. Social Performance Reporting is coming, built on reporting-time gains customers already see. Talk to us about early access to the preview and upcoming agents.

Do the agents act without my approval?

No. Agentic doesn't mean unattended. Paid Media pushes ads built paused until you approve the spend; the CRM agent needs two approvals and stops in under a minute; the Social agent runs autonomously only if you choose -- otherwise it waits at a sign-off checkpoint. Predictions are always shown as predictions.

What do all the agents have in common?

SOMONITOR, SOMIN's real-time category engine. Every agent reasons from what's actually winning in your category right now, refreshed live -- not a generic model, a mention feed, or a quarterly survey. One engine underneath; six jobs on top.

How is this different from ChatGPT or a general AI assistant?

General assistants answer from the open web. SOMIN's agents answer from the creative actually winning in your category, score concepts before you spend, and output work in your brand. See the full breakdown on the Strategy comparison vs ChatGPT, Perplexity, Meltwater, and GWI.

What models and infrastructure power the agents?

Google Cloud, with Gemini 2.5 Pro for large-scale ad and video understanding and idea generation, alongside SOMIN's in-house SoCraft CTR models -- published at ACM WSDM 2023, 78% CTR-prediction accuracy, about 3× the human baseline. An explainable-AI layer keeps every recommendation traceable to the underlying data.

Is the research peer-reviewed?

Yes -- the platform and its models are published at top-tier ACM venues, including ACM Multimedia 2025, ACM Multimedia 2023, and ACM WSDM 2023, with the research line going back to ACM WSDM 2021 and ACM Multimedia 2018.

How fast can I see an agent working?

Book a 30-minute walkthrough on your own category -- median 13 minutes from a brief to a stack of pre-scored concepts, independently documented in Google Cloud's customer story on SOMIN.

Put the agents on your category.

A 30-minute walkthrough -- your category, your competitors, and the agents that fit the brief you're on this week.

Book a call See the platform

Six agents. One end-to-end loop.

Strategy

Paid Media

Influencer & GMV

Social Media Manager

CRM & Email

Performance Reporting

Decide the angle from what's winning.

Score the ad before you spend.

Brief every creator on what lifts sales.

A daily, on-brand presence that runs itself.

Know the winning email before you hit send.

Your client's report, always current.

Agentic -- but you hold the gate.

Named models. Published research. Verifiable claims.

One agent suite. Two ways to win.

Run every function on live category signal.

Deploy the agents across every client.

What teams ask about the agents.

Put the agents on your category.