Your own AI agent that learns from your data.
Not generic marketing advice based on what works for everyone. An AI agent that accumulates knowledge from your campaign history, tested hypotheses, competitor intelligence, and your own business rules — and gets more accurate with every briefing cycle.
What the agent learns from
Your campaign history
The agent reads 30+ days of your ad performance data in every briefing cycle. It knows your ROAS baselines, your seasonal patterns, and which campaigns have historically over- or under-performed.
HADI hypothesis outcomes
When a sprint closes with a confirmed or failed outcome, that learning is permanently archived into the agent's memory. "Increasing bid cap above €3.50 on Retargeting reduces conversion volume for this account" — facts like this persist across every future recommendation.
Competitor intelligence
After each weekly competitor snapshot, the agent archives relevant findings: which formats competitors are scaling, which they're pulling, what messaging angles are gaining traction in your category.
Business profile & rules
Your performance targets, key products, budget rules, audience notes, and any context you give the agent in chat. These are stored as durable rules and applied to every briefing and recommendation.
Industry patterns
Cross-tenant knowledge — anonymised patterns from accounts in similar categories and ICPs — is curated manually by Prooflytics and made available to your agent as global context. Your agent benefits from patterns your account hasn't experienced yet.
Chat conversations
When you correct the agent in chat ("actually our ROAS target is 4.0, not 3.5"), those corrections are extracted and stored permanently. You tell it once — it remembers forever.
How the learning loop works
An event generates a learning
A HADI sprint closes with a confirmed outcome. A competitor scales a new format. You correct the agent in chat. A business profile field is updated.
The learning is archived
The event is processed into a structured memory entry with a type (confirmed hypothesis, competitor finding, tenant rule, industry pattern), importance score, and trust level. Entries with a validity window expire automatically.
The agent retrieves it
On the next briefing cycle or chat message, the most relevant entries are retrieved via semantic search over your knowledge base and injected into the AI context alongside your live performance data.
Recommendations improve
The briefing recommendation references your specific knowledge: "Based on your confirmed test in sprint #14, increasing bid cap on this campaign historically reduces volume. Consider adjusting the targeting instead."
Increasing Meta Retargeting bid cap above €3.50 reduces conversion volume for this account. Tested twice, both Confirmed.
BrandX has been scaling UGC talking-head format since April. Now their longest-running creative. Category is shifting toward this format.
ROAS target for Retargeting is 4.5, not 3.5. Higher due to warm audience quality. Apply to all Retargeting recommendations.
Frequently asked questions
What makes the Prooflytics agent different from generic AI marketing tools?
Generic AI marketing tools answer questions based on training data — general marketing knowledge from the internet. The Prooflytics agent answers based on your data: your campaign history, your confirmed and failed tests, your competitor landscape, and your custom rules. It knows that "increasing bid cap on Retargeting historically reduces your conversion volume" because it learned that from your own HADI sprint outcomes — not from a training set.
How does the agent learn? Is it fine-tuned on my data?
The agent is not fine-tuned. It uses a retrieval-augmented approach: a per-tenant knowledge base of structured memory entries (facts, rules, outcomes, patterns) that are retrieved and injected into the context of every briefing, recommendation, and chat response. This means learning is immediate — a new fact added today is used in today's briefing, not after a retraining cycle.
Is my data used to train a shared model?
No. Your tenant knowledge base is private. Cross-tenant patterns are curated manually by Prooflytics (anonymised, specific to your ICP and vertical) and offered as optional global context — never derived automatically from your data. Your account's specific facts never leave your tenant.
Where is the agent used within Prooflytics?
The agent's knowledge base is active in three places: the daily morning briefing (recommendations are calibrated to your rules and confirmed outcomes), the action recommendations (ranked by your historical response patterns), and the in-product chat (answers questions about your account using accumulated knowledge). The same knowledge base feeds all three.
Can I review and delete what the agent has learned?
Yes. The AI settings page shows all learned rules and memory entries. You can review each entry, see where it came from (HADI sprint, competitor finding, chat correction, manual business profile), and delete any entry that is incorrect or outdated. The agent's memory is fully transparent and controllable.
An AI agent that knows your business, not just marketing in general
Start building your agent's knowledge base from day one. 14-day free trial, no credit card required.
Independent reviews on G2.