◊FallEnterprise Book Call
◆ AI-Native Transformation · TBAk – TBAk engagements

Your business gets its own trained AI. Not rented. Owned.

Self-hosted open-source LLMs · fine-tuned on your business data · prompt-compressed for modest hardware · full org rollout · SLA + support. The AI transformation SMBs were promised, delivered as sovereign infrastructure you keep even if we disappear tomorrow.

Find your tier · 2 min Book discovery call
4 tiers · TBAk → TBAk 4-20 week delivery MIT licence · you own it SLA + support included at Enterprise tier

Engagement tiers

Four sizes. One shape. Compound value each tier.

Each tier includes everything from the tier below. Start where you are, upgrade when the return is proven.

Starter

Get the blueprint, ship the ops layer, prove ROI on rented AI.
TBAk fixed
4–6 weeks · pay 50/50
  • Assessment + AI-first blueprint document
  • Operational suite (enquiry bot · content · CRM · dashboard)
  • Integration with hosted LLM (Claude/OpenAI)
  • Prompt library tuned to your voice
  • Team onboarding + 30 days support
Choose Starter

Sovereign

Add your own on-box LLM. No API bills. No data leaving.
TBAk fixed
6–8 weeks · pay 40/40/20
  • Everything in Starter
  • Docker stack: Ollama + Llama 3.1 8B on your hardware
  • LiteLLM proxy (OpenAI-compatible endpoint)
  • LLMLingua-2 prompt compression layer
  • Integration adapter: swap hosted → sovereign in one config
  • GPU/CPU sizing guide + cost model
Choose Sovereign

Trained

Fine-tune the model on your business data. Your voice, your process.
TBAk fixed
10–12 weeks · pay 30/40/30
  • Everything in Sovereign
  • LoRA fine-tune on your emails, docs, transcripts
  • Business-specific eval harness + benchmark suite
  • Retraining cadence (quarterly)
  • Data governance + audit trail
  • 60 days support + 1 retraining included
Choose Trained

AI-Native Enterprise

Full org rollout. Dept-by-dept. SLA, training, custodianship.
TBAk from
16–20 weeks · milestone billing
  • Everything in Trained
  • Dept-by-dept integration plan + rollout
  • Full staff training programme + certification
  • SLA + 24/7 support · 12-month contract
  • Custom sovereign-network config for multi-site
  • Custodianship option: we run it forever, or you take it in-house
Talk to us

Interactive assessment

Six questions. One recommendation. Two minutes.

Answer honestly. The recommendation is based on your headcount, data sensitivity, current stack, and appetite for sovereignty. Answers save locally — nothing leaves your browser.

The 5-stage blueprint

Every engagement walks these five stages.

Refined across the estate's transformation work. No stage skipped, no order changed.

STAGE 1

Assessment & Mapping

Process audit across departments. Data inventory. Team capability mapping. Quick-win identification.

STAGE 2

Virtual Sandbox Setup

Isolated environment for AI experimentation. Feedback loops. No disruption to live ops.

STAGE 3

Pilot Implementation

Deploy AI in one high-impact area. Train the team. Measure productivity. Iterate rapidly.

STAGE 4

Gradual Rollout

Expand successful pilots across the org. Custom workflows per team. Integration with existing tools.

STAGE 5

AI-Native Operations

AI embedded in every process. Agents handling routine work. Ongoing continuous evolution.

Deployment options

Five ways to run AI. Sovereignty vs risk trade-off.

Most SMBs think "AI" means Corporate Cloud. It's one option of five. The right choice depends on your data sensitivity, regulatory position, and appetite for control.

Corporate Cloud OpenAI · Anthropic · Google

Use major AI providers directly. Data flows through their servers. Fastest to start, weakest control.

Sovereignty
Low
Data Risk
Medium

Private Cloud AWS · Azure · GCP

Run open-source models on your cloud infrastructure. More control over data flow and residency.

Sovereignty
Medium
Data Risk
Low-Med

On-Premises Self-hosted models

Open-source models on your own servers. Full data control. Requires GPU or capable CPU box.

Sovereignty
High
Data Risk
Low

Hybrid Mixed deployment

Corporate APIs for general tasks. Private / on-prem for sensitive data. Best of both when done right.

Sovereignty
Med-High
Data Risk
Low

Sovereign Network Distributed mesh

Distributed compute network. You own your data, contribute compute, earn from participation.

Sovereignty
Maximum
Data Risk
Very Low

Live ROI calculator

Estimate 3-year savings + productivity gain.

Adjust the sliders to your business. The output updates live. Nothing sent anywhere.

SaaS eliminated over 3 years
TBA
Productivity value / year
TBA
3-year net return
TBA
Effective ROI multiple
10.3×

Statement of Work · templates

Every tier has a signed SoW. Preview them here.

Full scope, deliverables, timeline, payment schedule, IP terms, exit clauses. Markdown and DOCX both provided.

Starter

TBAk · 4-6 weeks
MDDOCX

Sovereign

TBAk · 6-8 weeks
MDDOCX

Trained

TBAk · 10-12 weeks
MDDOCX

AI-Native

TBAk · 16-20 weeks
MDDOCX

Live sovereign demo

Chat with an on-device LLM. No API, no telemetry.

WebLLM loads a small language model straight into your browser. First load takes a minute to download the weights, then runs entirely on your machine. This is what "sovereign" feels like.

Idle · click Load Model to begin
System
Once the model loads, ask it anything. This is a demo of on-device inference — the same architecture used in the Sovereign, Trained, and AI-Native tiers, at production scale.

Book a discovery call

Tell us your setup. 24-hour written response.

No sales-team funnel. Simon reads every message and replies personally within 24 hours (usually faster). Include enough detail to skip the first back-and-forth.

How this works

You send the brief. We reply with a fit assessment and either (a) a proposed tier + start date, or (b) an honest "not us, here's who might be."

Response<24 hours, always Simon
Discovery45-min call, free, no obligation
ProposalWritten scope within 48h of call
Kickoff2-week lead time from signature
Emailsjgant80@gmail.com
Estateai-nativesolutions.com

FAQ

The questions everyone actually asks.

Isn't ChatGPT good enough?
For personal productivity — yes. For running a business — depends. Hosted AI reads every prompt you send. If your data is sensitive (client records, medical, financial, legal), that's a governance problem. Sovereign LLMs solve it by never letting data leave your box. The Sovereign tier onwards is built for exactly this.
Do we need to buy a GPU?
Not always. Llama 3.1 8B runs on modern CPUs (slower) or a modest GPU (~TBA for a 24GB card). For higher throughput or the 70B model, you'd want a proper server (~TBAk). The Sovereign tier includes hardware sizing and can price the box into the engagement.
What if you disappear tomorrow?
The whole system is MIT-licensed and delivered as source. You get every Docker image, every fine-tuned weight, every notebook, every config, every runbook. If we vanished you'd carry on unchanged — that's structural, not marketing. See the exit clause in every SoW.
Prompt compression — what and why?
LLMLingua-2 is a technique that shortens prompts by 2-5× while preserving accuracy. On self-hosted models this means faster responses and lower compute cost per query. It's a small model that shrinks big prompts before they hit the main model. Included from Sovereign tier upwards.
Can we start on Starter and upgrade later?
Yes. Every tier's price is credited toward the next if you upgrade within 90 days of delivery. Blueprint work carries over. Ops layer carries over. Only the sovereign layer needs adding on the Sovereign upgrade.
How is this different from a McKinsey AI project?
Two ways. First, we ship working software, not a deck. Second, single practitioner (Simon), no partners, no analysts, no rebadged juniors. The TBAk tier is priced at what a Big-Four Discovery Phase costs and delivers what their Phase 4 promises.
Do you sign NDAs?
Yes, before any brief exchange. Standard mutual NDA on request — we can also work under yours. All engagements include IP assignment to you for anything built specifically for your business.
Where are you based?
United Kingdom. GMT business hours for UK/EU. Evening calls available for US East Coast. Trained tier engagements typically include 2-4 on-site days per phase; other tiers fully remote.

How to use this page

1. Assessment

Six questions. Recommends a tier based on your size, data sensitivity, current stack, and appetite for sovereignty. Answers save in your browser only.

2. Tier ladder

Four tiers. Each includes everything from the tier below. Upgrade later and previous tier's fee is credited within 90 days.

3. Blueprint

Every engagement follows five stages: Assess → Sandbox → Pilot → Rollout → AI-Native. No skipping.

4. Deployment

Five ways to actually run the AI. Corporate Cloud is one option. Sovereignty vs risk trade-off shown on each card.

5. ROI

Adjust the four inputs to your numbers. Output updates live. Purely local calculation.

6. SoW downloads

Every tier has a full Statement of Work in Markdown and DOCX. Legal review before signature is expected — templates are honest starting points.

7. Live demo

Click Load Model in the Sovereign demo. Downloads ~600MB the first time (cached after) and runs a small language model entirely in your browser. This is what sovereign feels like.

8. Contact

Simon reads every brief personally. Written response within 24 hours. If we're not a fit, we say so and point at who might be.

Sovereignty check

This page: single HTML file, no analytics, no cookies, no telemetry, no third-party JS except the font CDN and (optional) WebLLM ESM. Data you type in the assessment or ROI stays in your IndexedDB. The contact form uses mailto: fallback — no server involved unless you configure one.