Autopilot
Get in touch
// Method

How we work.

Output in weeks, not in quarters. Every step human-reviewed, every insight documented in the Mandant-Memory.

// Measurable, not felt

The state of the Mittelstand 2026.

Six numbers from current studies. Each one is known. Together they make the case for a retainer instead of a third consultancy.

Mittelstand · AI maturity · Q1 2026 n = six indicators · last updated 2026-05-08
01 / Awareness
86%
see the relevance of AI
02 / Delivery
23%
have shipped a project successfully
03 / Resources
70%
“we lack the people”
04 / Capacity
66%
“we lack the time”
05 / Data
76%
“our data isn’t ready”
06 / Strategy
43%
have no AI strategy at all

“Companies don’t fail at AI — they fail because their own knowledge is invisible, scattered and tied to individual heads.”

— Materna, 2025

What you’re missing isn’t another AI tool. What’s missing is a team that makes sure AI delivers real impact in your business every day. One team, one retainer, every question.

// HITL mechanic

How every output runs through the crew.

Four steps. Who reviews, what gets reviewed, how long it takes. A promise you can hold us to — no consulting-speak claim.

Step Who What’s reviewed Turnaround
01 AI draft
Tooling — Anthropic / OpenAI / local LLM
Tool is chosen per use case + hosting class. No output leaves this step directly.
instant
02 Account-lead review
Your fixed account lead (default: Martin)
Fact-check, context, tonality, EU-AI-Act risk classification, marked caveats.
< 1 business day
03 Specialist review
Domain reviewer from the pool (high-risk / regulatory / industry-specific)
Domain correctness, source robustness, industry-legal implications.
+ 1 business day
04 Sign-off & delivery
Account lead
Final stamp with reviewer names, models, risk class. The output leaves the house.
instant after step 2 / 3

Step 3 is only triggered when the output falls under EU-AI-Act Annex III or needs industry-legal depth. Otherwise the output goes straight from step 2 to sign-off.

// Visible proof

Every AC output ends with this.

A uniform sign-off block — who reviewed, which models were used, which risk class. You show this in audits, to the supervisory board, to the data protection team. Brand asset and trust anchor in one.

Sample output · Make-or-Buy memo

…our recommendation is to use Tool X instead of building in-house. Three reasons:

  1. GDPR-compliant hosted in Frankfurt, EU-AI-Act-compliant classification
  2. Setup costs 4× cheaper than in-house over six months
  3. Vendor since 2019, three Mittelstand references verified

Risks & assumptions in the appendix. Sources 2 and 5 unconfirmed by reviewer, marked in the body with [?].

— — — — — — — — — — — — — — — — — — — — — — — —
AC · Review Sign-Off 2026-05-27 · 14:32
Reviewed by
M. Zielinski (account lead)
Specialist reviewer
A. Schmidt (tax law, 18 years practice)
AI models
Anthropic Claude Opus 4.7, OpenAI GPT-5o
Risk class
limited · EU-AI-Act
Notes
Sources 2 + 5 unconfirmed by reviewer, marked in body.

If this output got something wrong, write to martin@noorder.partners. We correct within 1 business day.

// Reviewer statistics

What the crew caught.

As of 2026-05-27 Next update 2026-07-01
— Outputs reviewed
— AI hallucinations caught
— Source / fact corrections

The statistic starts as soon as the first mandates have received outputs. We don’t fill in polished numbers — empty tiles with an honest date beat an impressive marketing snapshot with no data behind it.

// 01 — Mandant-Memory

The company profile that grows with you.

On the first mandate, consultants start at zero. On the tenth mandate, consultants start at zero. With us there is one file per mandate that grows — strategy, taboos, decisions, quarterly results. The account lead has it open when you call. You have it open whenever you want.

// Excerpt

Mandate profile

Current · Q3 2026
Industry
Machinery · 240 employees
Strategy
Cut service throughput from 14 to 8 days
Hosting
EU cloud, no US vendor
Taboos
2 documented · HR decisions, one tool
Phases
Q1Q2Q3

You see everything we know about your company. Corrections any time.

What lives in the memory

  • Business model & strategy — target picture, key segments, what’s growing and stagnating
  • Departments & contacts — who owns which topic
  • Hosting classes per use case — on-prem / EU cloud / hybrid, plus rationale
  • Taboos — data, tools, vendors that are out
  • Past decisions — what was evaluated, what was bought, what was rejected
  • Phase log — quarter by quarter: topics, outputs, impact
  • Reviewer stamp — per output: the risk class and the reviewer

Lives in the backend, visible in the mandate dashboard. You can correct any entry. When you leave, the memory leaves with you — export is GDPR-mandated and the default.

Onboarding — 10 questions, no workshop

In the first conversation and the first two weeks, unobtrusive. Answers land in the memory, correctable later.

  1. 01 Strategy What is your most important business goal for the next 12 months?
  2. 02 Business model Who are your three highest-revenue customer segments — and which is growing, which is stagnating?
  3. 03 Departments Which divisions do you have — and who is the contact for AI topics there?
  4. 04 Past attempts Which AI attempts have you made — what worked, what disappointed, why?
  5. 05 Data Where does your relevant data live — and which data must not leave the company?
  6. 06 Hosting Are there sovereignty requirements (on-prem only, EU cloud, hybrid)?
  7. 07 Taboos Which topics, tools or vendors are out for you — and why?
  8. 08 Compliance Which regulatory frameworks must you comply with (EU AI Act, industry law, KRITIS)?
  9. 09 Decision paths At which order size does who decide — managing director, department head, procurement?
  10. 10 Success measure How would you tell in 6 months that Autopilot was worth it?
// 03 — Intelligence Match

Autopilot Ventures researches. We match for you.

Autopilot Ventures is our sister brand — podcast, newsletter, daily scans across 165+ industry sources. What gets researched there, you as a Mittelstand leader would never have time to read. You don’t have to — we match the findings against your Mandant-Memory and only send what fits you.

AV source
VDMA study

“AI adoption in plant engineering Q2 2026" — 47 mid-sized plant builders, 5 categories of use cases, 3 tools with proven ROI > 25%.

Match against Mandant-Memory
M-014 · Plant engineering · 240 employees High relevance
Industry fits · AI strategy open in memory · two of three tools not in taboo list
Push to account lead

Study read, 3-point relevance note written, suggestion when to talk to the mandate about it. Not into your inbox — into the account-lead workflow.

// Impact

How we measure whether this works.

Others count logins and engagement scores. We count what is different at the end of the quarter. Four commitments — and a quarterly review you can show your advisory board.

01

Output, not activity

We don’t count requests or hours, we count what reaches you. What got shipped, what runs in production, what it concretely delivered.

02

Three hard metrics per quarter

Time saved (e.g. “service handling from 8 to 5 minutes”), error reduction (e.g. “reviewer corrections from 12 to 4 per month”), or revenue effect (e.g. “quote throughput from 14 to 6 days”). Individual per mandate, agreed before the first quarter.

03

Visible in the dashboard

Status, running applications, open tickets, quarterly KPIs. You can look in anytime — we don’t report behind closed doors.

04

Written quarterly review

One-pager per quarter: what was shipped, what it delivered, what remains open. Goes to you plus all stakeholders you name.

Before the first quarter we agree together on three metrics that make sense for your business. We don’t adjust them later to make them look better. If a quarter didn’t deliver, that is what the review says.

// Support, not control

Support, not control: AI for the team, not over the team.

Bringing AI into your company without poisoning the team — that is our line. Four promises we want to be measured against.

01

Anonymised team results

We measure impact at team level, not behaviour at person level. Who uses how much AI is not our question.

02

Impact, not usage

Others count logins. We show which business outcomes came out of AI use. Like a fuel bill — it makes things visible, it does not judge.

03

Local where it has to be

Sensitive data never leaves your servers. Where cloud is allowed, we use EU cloud. Where nothing may leave, it runs locally.

04

Bring people along

Not a layoff project. No pace against the team. Those who get AI become more valuable — not replaced.

// Stance

Where we stand.

Three principles. They are not up for negotiation.

01

Plain talk

We say what works and what doesn’t — in your words, no buzzword bingo. Concrete outputs count, not consulting-speak.

02

Delivered, not debated

From hundreds of projects we know what works. Concept and execution come straight through — days instead of months.

03

30% impact

We only go after topics where something measurable happens. Substance before shine.

// Ready?

Get in touch or book a discovery call.

30 minutes, no pitch, no pressure. We listen, we say honestly whether we’re the right vendor.

Get in touch Or try 7 days first