The Cosa Foundations are the architecture of an AI-native GTM system: six levels, each building on the ones below it. Start at the lowest precondition you cannot prove, because an agent on an unproven level works on assumptions.
Why is “How do we implement AI in sales?” the wrong question?
Because it starts with the solution. The right question is: where are you currently blocked in your system, and why? Otherwise you will proceed to fail, with a higher token bill.
The answer is the lowest precondition you cannot prove. While it is missing, you cannot tell whether a bad result comes from there, and a good one can be an accident. A metric says something about a level only when everything it rests on is proven.
What does an AI caller do on an unproven level?
An AI caller on an unproven level scales the uncertainty, not the result. Take an agent for Level 4 that calls on your lead engine. Suppose proof is missing for the business case or the engine.
The agent calls more often. If the Booking Rate drops, you cannot tell whether the agent, the engine or the business case is the cause. If it rises, you cannot tell whether it will repeat. Your agent’s ceiling is your current performance. It cannot tell the proven from the assumed and will likely burn more of your contacts and markets than it converts.
Which precondition is missing first?
Five steps you can take yourself today:
Read the six questions from Level 1 upwards.
Cross out every felt yes: only what you can show with customers, numbers or concrete commitments counts.
Mark the lowest level you cannot prove: that is where you work first.
Prepare the levels above it, but do not optimize, scale or automate there until the marked level is proven.
Put an agent on a level only once a human has proven it in your own system.
- Read from the bottom Read the six questions from Level 1 upwards.
- Count only proof Cross out every felt yes: only what you can show with customers, numbers or concrete commitments counts.
- Mark the lowest level Mark the lowest level you cannot prove: that is where you work first.
- Prepare, do not scale Prepare the levels above it, but do not optimize, scale or automate there until the marked level is proven.
- Agent after proof Put an agent on a level only once a human has proven it in your own system.
After that you know what to work on first, and the next level’s results can be read once the marked level is proven. All six questions are in the self-check.
The full picture
Here are all six levels in the order in which they build on each other.
Each level builds on the earlier ones.
Its results are only valid when those are proven.
Otherwise a bad result may come from an earlier level, and a good one may not repeat: false signals in both directions.
Proven means you can show it. Where your system stands, the self-check shows: it asks the levels in order, and only what you can show counts.
Worth building once the level is proven
Level 1 Business Case
Do you have a business case that can win?
Proven when: ICP and solution fit. The sales cycle fits. The ticket size is plausible. The market is large enough. You can reach the decision maker, and your current approach actually gets you into the call with them.
No agent yet Here it is first established whether the case holds. An agent needs that result as its input.
Level 2 Growth Foundations
Do you have the growth foundations in place?
Proven when: Lead engine and tech are set up correctly. Funnel metrics are assessed. Sales collateral and processes are in place.
Then worth building: Company Brain Clean up your data ontology and implement a Company Brain. It now stores what is proven, not assumptions.
Level 3 Unit Economics
Do you have a working process that pays off, meaning the CAC does not eat the potential revenue?
Proven when: Lead generation, research and conversion work as a working process. Unit economics, CAC and offer are understood: the CAC does not eat the potential revenue.
Then worth building: Where agents pay off Understand which areas can be agentified and which cannot, and what these agents need to perform in order to be the right investment.
Level 4 Lead Engine Optimization
Is your lead engine set up correctly, and does it reach its KPIs on its own?
Proven when: It captures the right signals. Channels, tools and automations are exhausted in their potential to reach the right people at the right time with the right communication. The funnel is set up to lead them into meetings. The proof: Booking Rate 1 to 3 % (or Response Rate 10 to 30 %), Disco to Demo above 50 %, Show Rate above 90 %.
Then worth building: Lead engine agents Targeted, specialized agents for single jobs of your lead engine.
Level 5 Sales Process Optimization
Do you have advanced sales processes in place to shorten your sales cycle?
Proven when: Disco to Closed Won 10 to 30 %. (LTV/CAC) / sales cycle in months above 0.25, with an LTV/CAC of at least 3.
Then worth building: Closing agents Specialized closing agents. They repeat what real deals have already confirmed.
Level 6 Hiring & Development Funnel Optimization
Do you have the right training, reporting and tracking systems in place to scale your team?
Proven when: The hiring KPIs are met. New hires are productive in record time.
Then worth building: AI SDRs and AI AEs AI SDRs or AI AEs that take over whole roles, once every agentic step before is proven as well.
An agent on a level whose preconditions are not proven scales the uncertainty, not the result.
Frequently asked questions
Where do we start with AI in sales?
On the lowest precondition you cannot prove. An agent on an unproven level works on assumptions.
Why can the order of the levels not be swapped?
Because the results of a level are only valid when the levels below it are proven.
Are AI SDRs worth it for us?
Specialized agents for single lead engine jobs are worth it once Level 4 is proven. AI SDRs or AI AEs wait until Level 6 and every agentic step before it are proven.