of marketers call AI the biggest disruption to their work in 20 years
AI marketing operations for teams that want infrastructure, not tools.
Most marketing teams have tried ChatGPT for first drafts, bought a Clay seat, and called it an AI strategy. We build the semantic knowledge base underneath: the persistent, queryable infrastructure that makes every AI workflow actually compound. We run our own. We build them for clients who want one.
01ANSWER
What is AI marketing operations, and why does it matter?
ANSWER
AI marketing operations is the infrastructure layer that makes AI genuinely useful inside a marketing team. Not the tools themselves (prompts, ChatGPT, Clay and Claude Projects are table stakes by 2026) but the persistent semantic knowledge base underneath them, wired into your stack through MCP, that turns every AI workflow from a one-off into compounding memory. 61% of marketers call AI the biggest disruption to their work in 20 years. Only 6% have fully embedded it. The gap is infrastructure.
have fully embedded it
THE GAP BETWEEN THE TWO
is infrastructure. 02METHOD
What we actually build.
- 01
Semantic knowledge base
A vector-backed semantic knowledge base wired into your stack via MCP. Six canonical tables cover voice samples, content queries, decision history, client patterns, source evidence, and content operations. Our reference build runs on Supabase; yours can live on AWS, Google Cloud, or wherever your data already sits, and it works with Claude, Gemini, or any MCP-compatible tool.
- 02
MCP connector wiring
The knowledge base only matters if it can reach the tools your team already uses. We wire connectors for platforms such as HubSpot, Gmail, Google Drive, Slack, Ahrefs, and Notion, with scoped permissions, a documented access model, and full auditability.
- 03
Voice sample curation
We help curate 50 to 75 voice samples per writer, tagged by tone and context, then refreshed monthly. Every AI draft retrieves the most relevant samples as examples. This is usually the difference between drafts that sound like your team and drafts that sound generic.
- 04
Scheduled routines
Claude Code routines can run daily, weekly, or monthly against the Open Brain: content opportunity digests, cross-client pattern reports, draft momentum checks, and content refresh audits. The team stops doing manual reporting work. The infrastructure does it.
- 05
Governance and failure-mode design
Permission models, prompt-injection defence, data-loss prevention, voice-drift detection, and a governance document that tells the team what the system will and will not do. We ship the safety layer alongside the capability layer.
03TOOLS VS INFRASTRUCTURE
AI tools vs AI infrastructure.
The tools are table stakes. What makes AI genuinely useful is the owned structure underneath them.
AI TOOLS
AI INFRASTRUCTURE
PURCHASE UNIT
AI toolsPer-seat tools such as Clay, Notion AI, Jasper, and Copilot
AI infrastructureAn owned, MCP-wired semantic knowledge base on infrastructure you choose
MEMORY
AI toolsStarts from zero on every workflow
AI infrastructurePersistent and queryable, compounds across every engagement
VOICE FIDELITY
AI toolsSounds like ChatGPT with some brand tuning
AI infrastructureSounds like your team, retrieved from curated voice samples
STACK ACCESS
AI toolsEach tool sees whatever its account integrates
AI infrastructureMCP connectors across HubSpot, Gmail, Slack, Drive, Ahrefs, and Notion
OWNERSHIP
AI toolsVendor controls schema, permissions, and roadmap
AI infrastructureYou own the schema, the data, and the permission model
PORTABILITY
AI toolsLocked to the vendor's platform, pricing, and roadmap
AI infrastructureYours to move: the schema and data port to any cloud or model vendor
The difference is whether every AI workflow compounds, or whether every workflow starts from zero.
04OUTCOMES
What you actually get.
OUTCOME 01
A working Open Brain you own
Six canonical tables configured and populated, with vector-backed semantic search and hybrid keyword-plus-semantic queries, on infrastructure you own and can move.
OUTCOME 02
Connector layer and voice sample library
MCP wiring for the tools that matter, plus voice sample libraries curated per writer, tagged and refreshed monthly.
OUTCOME 03
Workflows, routines, and governance
Core workflows set up in the AI tool your team runs (Claude, in our reference build), scheduled routines against the knowledge base, and a governance layer the team can operate safely.
05DETAIL
Is AI operations work right for your business?
GOOD FIT
yes- +B2B companies with £5M to £50M in annual revenue
- +Marketing teams of 3 to 20 producing content, outbound, or multiple campaigns
- +Teams where AI use is already significant but outputs feel generic
- +Leadership who see AI infrastructure as an investment, not a productivity hack
- +Companies with at least one person who can own the system after build
NOT A FIT
no- xTeams of 1 to 2 with simple needs
- xCompanies that have not tried AI tools seriously yet
- xOrganisations looking for a Copilot or ChatGPT replacement
- xTeams unwilling to keep voice samples and source evidence fresh
If you see yourself in the not-a-fit list, start with off-the-shelf AI tools, build the habit, and come back when you hit the infrastructure wall.
WHO ACTUALLY DOES THE WORK
Jon
Leads AI operations engagements
A decade leading B2B content teams, fintech and agency side. Now based in Barcelona, deep in both content infrastructure and hands-on AI engineering.
Robin
Strategic architecture and client integration
Shapes strategic architecture and client-side integration, especially where the Open Brain connects to RevOps and client communications.
Katie
Content-specific systems
Contributes on voice sampling, editorial workflow design, and integration with content production.
06CASE STUDY / WE ARE ALL CONNECTED
We run the infrastructure we build.
Our Open Brain holds PEI Group decision history, Zaptec patterns, ForGood findings, cross-client technical SEO patterns, voice samples, and every piece of client-facing content we have published. It is wired into Gmail, Slack, Google Drive, HubSpot, and Ahrefs via MCP.
Client briefings take minutes, not half-days. Draft-to-publish cycles have compressed from weeks to days. The institutional memory that used to live in Robin's head now lives in a queryable store the whole team works from.
Read the Open Brain blueprint07HOW TO START
How does an engagement actually start?
SCOPING
Scoping Audit
From £8,500
3 TO 4 WEEKS, ONE-OFF
A diagnostic of your current AI usage, infrastructure gaps, and tool sprawl, plus a proposed Open Brain schema specific to your business.
BUILD
Build Project
From £22,000
4 TO 6 WEEKS END-TO-END
The full Open Brain setup: knowledge base configuration, schema build, MCP connector wiring, voice sample curation, workflow setup, prompt library, governance, and training.
RETAINER
Monthly Retainer
From £5,000/month
MINIMUM 6-MONTH COMMITMENT
Ongoing embedded AI operations work: new connectors, new routines, prompt iteration, voice library refresh, schema evolution, and strategic support as AI capabilities shift.
Not sure which fits? Book a call. We will tell you honestly which one is right, or that you do not need any of them yet.
Book a call08FAQ
Frequently asked questions.
Is this just a prompt library?
No. A prompt library is a small part of what we build. The core asset is the semantic knowledge base: the persistent, queryable store of your voice, decisions, client patterns, and evidence.
Does this replace our existing marketing tools?
No. It sits alongside them and connects them. Your HubSpot stays. Your Ahrefs stays. Your Notion stays. The Open Brain makes the useful context queryable through AI workflows.
How is this different from Glean, Notion AI, or Dust?
Those products index documents and provide search. We build an owned, auditable semantic knowledge base where you control the schema, the data, the permissions, and the future evolution.
What is the ongoing cost of running an Open Brain?
Infrastructure is usually a few hundred pounds per month at typical scale, depending on hosting, model usage, and connector volume. Retainer fees buy our ongoing involvement in operating and evolving it.
Can our existing developers build this instead?
Probably, given enough time and strong judgement. The technical components are documented. The hard part is schema design, operational discipline, voice sampling, and deciding what should become durable memory.
09RELATED
This work does not happen in isolation.
If your team wants AI infrastructure, not more AI tools, we should talk.
Tell us what you've tried, where it stalled, and what you wish was working. We'll tell you honestly whether a full Open Brain build makes sense, or whether you should start somewhere smaller.
Book a call