Skip to content
The world's first marketing-native LLM

The AI thatknows what goodlooks like.

Generic AI gives generic answers. Meridian is trained on 24 million real campaign outcomes, live benchmark data from 15 channels, and the collective knowledge of agencies managing billions in ad spend. It doesn't just describe your numbers. It tells you if they're any good.

Request access How it works
24M+ training examples
real campaigns, real outcomes
15 channel integrations
from Google to TikTok to Klaviyo
12 sectors covered
with budget-tier benchmarks
meridian.ai/analyse
POST /v1/analyse
sector: "ecommerce"
channel: "meta_ads"
roas: 2.8 // your value
budget: 15000 // monthly
Benchmark verdictBelow median
1.8x
P25
2.9x
P50
2.8x
You
4.6x
P75
7.2x
Top
Meridian insight

Your 2.8x ROAS sits at the 47th percentile for e-commerce Meta Ads at £10-20k/month. Accounts at your spend level typically hit 3.5-4.2x once audience segmentation is dialled in. Primary lever: creative frequency at 3.2, above the 2.4 fatigue threshold for your sector.

Pause 3 creatives above 3.0 frequencyHigh
Shift budget to ROAS > 4x ad setsHigh
Test UGC format, +34% CTR in your sectorMedium
Training data quality94%
0 channels
trained across
every major paid and organic platform
0M+
labelled training examples
real agency campaigns with verified outcomes
0%
benchmark accuracy
independently verified against held-out dataset
0x
more precise on marketing
vs GPT-4o on sector-specific queries
Why generic AI fails marketers

ChatGPT doesn't know
your sector. Meridian does.

Ask any general LLM about your 2.8x ROAS and it will say something about strong performance depending on your margins. Ask Meridian and it tells you you're at the 47th percentile for your sector, what the top 10% are doing differently, and which lever to pull first.

Generic LLM
Response to: "My Meta ROAS is 2.8x"

"A ROAS of 2.8x means you're generating £2.80 for every £1 spent. This could indicate strong performance depending on your industry and margins. You may want to consider whether your campaigns are meeting your business objectives and review your audience targeting."

No sector context
No budget-range awareness
No actionable next step
Could apply to any business
Meridian
Same question

"At 2.8x, you're at the 47th percentile for e-commerce Meta accounts at £10-20k/month. Median for your tier is 3.1x and top quartile accounts average 4.6x. Your creative frequency is 3.2, above the 2.4 threshold where e-commerce audiences fatigue. Pause ad sets above 3.0 frequency and rotate creative. Historically this recovers 0.4-0.8x ROAS within 14 days for accounts in your range."

Sector-specific percentile rank
Budget-range benchmarks
Specific next action
Evidence from comparable accounts
Architecture

Three versions.
One mission.

Meridian isn't bolt-on AI. It's a ground-up model stack built on a foundation LLM, fine-tuned on 24 million labelled marketing outcomes, and augmented with a live benchmark database that updates every 24 hours.

v1

Foundation model

Llama 3.1 70B base

Built on Llama 3.1 70B, one of the strongest open-source foundation models available. Deep language understanding, long-context reasoning, and a proven base for domain-specific fine-tuning.

v2

Marketing fine-tune

24M+ labelled examples

Fine-tuned on 24 million labelled agency report outputs, channel-specific analyses, and expert commentary, all labelled by actual campaign outcomes. This is what makes Meridian different from every other model.

v3

Live benchmark injection

Updated every 24 hours

At inference time, every query is enriched with live benchmark data: your sector, your channel, your budget tier. Meridian doesn't guess what good looks like. It looks it up.

The data flywheel

Every agency that joins Meridian makes it smarter for everyone else. More data means better benchmarks, which means better recommendations, which means better results, which means more agencies. That compounding advantage is the moat no competitor can buy.

Agencies run Meridian
Real outcome data flows in
Benchmarks get sharper
Recommendations improve
Clients get better results
Proprietary benchmark data

Percentile rankings,
not vague verdicts.

Meridian's benchmark database covers 15 channels, 12 sectors, and 6 budget tiers. Submit any metric and it immediately places you in the distribution, compares you to adjacent accounts, and tells you what separates the top quartile from everyone else.

15 channels, 12 sectors, 6 budget tiers
Updated every 24 hours from live campaign data
Percentile positions, not just averages
Budget-normalised so the comparison is always fair
800,000+ ad accounts contributing signal
Live benchmark database
Sector / ChannelMetricP25MedianP75Top 10%
E-commerce
Meta Ads
ROAS1.8x2.9x4.6x7.2x
Education
Google Ads
CPA£94£61£38£22
Hospitality
Google Ads
CTR2.1%3.8%5.9%9.2%
SaaS / B2B
LinkedIn Ads
CPL£88£54£31£19
Charity
Meta Ads
CPC£1.42£0.89£0.52£0.28
Retail
Google Shopping
ROAS2.2x3.6x5.8x9.4x
Healthcare
Google Ads
CPA£142£88£51£29
Travel
Meta Ads
CPC£0.92£0.58£0.34£0.18
Finance / Fintech
Google Ads
CPL£74£46£28£14
Automotive
YouTube Ads
CPV£0.04£0.025£0.014£0.008
Fashion / Beauty
TikTok Ads
ROAS1.4x2.6x4.1x6.8x
Property
Meta Ads
CPL£38£22£13£7
Updated daily · 15 channels · 12 sectors · 6 budget tiers · 800,000+ accounts
What it can do

Nine capabilities.
None of them generic.

Core feature

Benchmark-aware analysis

Every metric you submit gets benchmarked against real accounts in your sector, your channel, and your budget range. 47th percentile. Not just good or bad.

Forecasting

Outcome-calibrated forecasts

Meridian forecasts aren't statistical projections. They're built from what actually happened when agencies in similar positions took similar actions.

Paid media

Budget reallocation intelligence

Budget move recommendations come with real confidence intervals drawn from 24 million labelled data points across your exact spend tier.

Creative

Creative fatigue and format intelligence

Meridian knows which creative formats burn out fastest in your sector, which angles drive the highest CTR, and when to rotate before you lose ROAS.

Diagnostics

Anomaly root-cause diagnosis

Something dropped 30% overnight. Meridian cross-references 12 signal types across your account history and surfaces the most probable cause in seconds.

Reporting

Cross-channel performance narrative

15 channels. One story. Meridian reads across Google, Meta, TikTok, LinkedIn, Klaviyo and more to tell you what is actually driving growth.

Strategy

Real-time strategy generation

Brief Meridian on a client objective and it outputs a prioritised 90-day strategy grounded in what has worked for comparable accounts in that sector.

Audience

Audience and segment insight

Meridian identifies audience fatigue, overlap, and expansion opportunities by drawing on behavioural signal patterns from 800,000+ ad accounts.

Attribution

Attribution and revenue modelling

Go beyond last-click. Meridian builds data-driven attribution models trained on first-party outcomes across your exact channel mix and funnel length.

Meridian vs every other model

GPT-4o, Claude 3.5, Gemini 1.5 Pro, and Mistral Large are all excellent general-purpose models. None of them know what a good ROAS looks like for a fashion brand running TikTok at £20k a month.

CapabilityMeridianGPT-4oClaude 3.5Gemini 1.5Mistral
ROAS and CPA benchmarks by sector and budget tier
Trained on real campaign outcomes (not simulations)
Platform-specific anomaly detection thresholds
Budget-tier-aware recommendations
Creative fatigue and frequency intelligence
Cross-channel attribution modelling
partial
partial
partial
Learns from recommendation outcomes over time
Understands Google Ads quality score dynamics
partial
Sector-specific creative performance patterns
Live benchmark database injection at inference
White-label API with custom model IDs
Fluent marketing and copywriting output
API access

Drop in. No migration.
OpenAI-compatible.

Meridian's API matches the OpenAI spec exactly. Swap the base URL and model ID and you're done. If it works with the OpenAI SDK today, it works with Meridian tomorrow.

OpenAI SDK compatible
One-line swap from openai to meridian. No refactor.
Your data is never trained on
Client data is isolated and never used to update the model.
Streaming supported
Server-sent events for real-time UI streaming out of the box.
Usage analytics built in
Token spend, latency, cost per insight, and model version history.
Sub-800ms median latency
Optimised inference stack with global edge nodes.
15 native integrations
Google Ads, Meta, TikTok, LinkedIn, GA4, Klaviyo, Search Console, and more.
import OpenAI from 'openai'

// Just swap the baseURL — nothing else changes
const client = new OpenAI({
  baseURL: 'https://api.meridian.ai/v1',
  apiKey: process.env.MERIDIAN_API_KEY,
});

const response = await client.chat.completions.create({
  model: 'meridian-marketing-v1',
  messages: [
    { role: 'system', content: 'sector:ecommerce budget:15000' },
    { role: 'user', content: 'Analyse my Meta ROAS of 2.8x' },
  ],
});
Compatible with OpenAI SDK v4+ · No code refactor needed

Access Meridian

Meridian is in private alpha. Pricing is based on token volume and team size. Every tier includes benchmark database access.

Most popular

Agency API

Starting from £299/mo

Direct API access for your own tools and integrations. Replaces your OpenAI subscription in minutes.

meridian-marketing-v3 model access
2M tokens/month included
Live benchmark database queries
Sector, budget tier, and channel context layer
Usage and cost analytics dashboard
Email + chat support
Request access
Recommended
Best value

White-label Platform

From £899/mo

The full StratOS platform rebranded as your own product, with Meridian powering every AI feature inside it.

Everything in Agency API
Full StratOS dashboard (your brand)
15-channel data integrations
Client portal with magic-link auth
Unlimited client accounts
Priority support and onboarding call
Talk to us
Custom

Enterprise

Custom

Private deployment, fine-tuning on your own first-party data, dedicated inference, and full SLAs.

Private model deployment
Fine-tuning on your own data
Custom benchmark database
Dedicated GPU inference cluster
Custom rate limits and SLAs
Dedicated account and engineering team
Contact us

Join the waitlist.

Meridian is in private alpha with a select group of agencies. Tell us what you're building and we'll prioritise access based on fit.

No commitment · Private alpha · We'll reach out within 48 hours