A growing number of mid-sized German companies have quietly stopped buying anything with AI in it. Not because they doubt the technology, and not because of budget. Because they have no internal AI policy yet, and until Legal writes one the rule is simple: no projects involving AI.
If you sell go-to-market systems, you meet this monthly. The need is real, the budget exists, the fit is right, and the answer is still “next year”. It looks like a lost deal. It is usually a sequencing problem.
The Policy Is Not the Obstacle. The Architecture Is.
Most outbound tooling is built so that AI is load-bearing. Remove it and nothing runs, because the personalisation, the qualification and the reply handling are all the same component. That is why “can we start without AI?” is normally answered with a polite no.
An engine assembled from separate parts behaves differently. Data enrichment, sequencing, sending, reply routing and CRM handover are distinct steps connected by an orchestration layer. AI sits at specific points in that chain, and those points have switches.
Turn them off and you have not broken the engine. You have an engine that works the way outbound worked for the twenty years before large language models existed: one template per decision-maker profile, segmented lists, disciplined follow-up. That approach built a great many pipelines.
What You Lose, Stated Honestly
Personalisation quality drops. There is no point pretending otherwise. A template addressed to a role is less specific than a message referencing a funding round, a leadership change or a hiring pattern. Reply rates will be lower than they would be with the AI layer active.
What you do not lose is everything else: the data architecture, the GDPR handling, the suppression lists, the reporting, the CRM handover, the compliance posture. Those are the expensive parts to build, and they are entirely AI-free.
So the trade is a measurable reduction in one metric, against the ability to start twelve months earlier.
The Commercial Argument for Starting Anyway
There is a second reason to run the first phase without AI, and it has nothing to do with policy.
When the AI policy does arrive, someone will have to justify the spend. That argument is far easier to win holding real numbers than a projection. “Our current engine produces this many qualified conversations per month, and the personalisation layer is expected to move it by this much” is a different conversation from “we would like to invest in an outbound system.”
Phase one becomes the business case for phase two. The policy work and the pipeline work run in parallel instead of in sequence.
What This Looks Like in Practice
The AI integration points stay in the design, documented and dormant. Activating them later is configuration, not reconstruction.
One Thing Worth Saying to Legal
An AI policy written in the abstract tends to be restrictive, because nobody wants to authorise a risk they cannot picture. An AI policy written while a system is running in front of you tends to be more precise, because the questions become concrete: which data touches which model, under what retention terms, with which human checkpoints.
Running phase one without AI does not only produce pipeline. It produces the specific questions your policy has to answer.
The Sequence
Build the engine now with the AI switched off. Prove the numbers. Let Legal write the policy against a real system rather than a hypothetical one. Switch the layer on when both are ready.
The companies that wait for the policy before starting lose a year of pipeline and arrive at the policy discussion with nothing to point at.