What it does
Your company's institutional memory. Whenever you finish something with a real result — a GTM experiment, a logged decision, an investor deal that closes, a hire, a completed outcome workflow — it becomes a learning event. The engine finds similar past actions, distills the lesson, and proposes source-linked recommendations: "from similar past actions, this kind of move tended to work (or underperform)." Every claim links back to the record it came from — the engine never makes an unsourced claim.
Who can use it
- Anyone in your organization with read permission for the Learning Engine module can browse events and recommendations.
- Update permission is needed to accept/reject recommendations and record realized outcomes.
- Everything is scoped to your organization — the engine only learns from your own data, never from other companies.
Where to find it
From the sidebar, open Strategy & Planning and click Learning Engine.
Find your way around
The page has two sections in the left rail:
- Recommendations — "Source-linked, accept or reject." Proposals generated from your history, waiting on your decision. The badge shows how many are pending.
- Learning Events — "What happened, and the lesson." The raw record of completed actions and what each one taught.
Stat chips show Pending (recommendations awaiting a decision), Events (total learning events), and Worked (recommendations you later confirmed paid off). Click any row to open its detail; use the refresh icon to reload.
Feed the engine
You don't create anything here directly — learning events flow in automatically when you finish work elsewhere:
- Complete a GTM experiment with a result.
- Record the outcome of a decision in Decision Loop.
- Close an Investor CRM deal (won or passed).
- Hire or reject a candidate in Jobs & Recruitment.
- Complete an Outcome Workflow.
Each shows up as a learning event with the action, the Expected vs Actual result, the Metric, a distilled Lesson, and a suggested Next step, plus a "View source in …" link back to the original record.
Review a learning event
- Open Learning Events. Filter by Action type (Campaign, Pricing change, Investor outreach, Product launch, Hiring, Runway action, Experiment, Decision) or Outcome (Win, Loss, Mixed, Inconclusive).
- Click an event to see its lesson and the Similar past actions panel — matched past events with a "% match" score, their outcome, their lesson, and a link to their source.
Act on a recommendation
When the engine finds a strong enough pattern, it proposes a recommendation.
- Open Recommendations and click a proposed item. The detail shows the summary, the rationale, and a Sources box linking to the exact records the claim is based on.
- Click Accept or Reject:
- Accept — confirms in a dialog: "You'll be able to record later whether following it actually worked, which feeds back into the engine."
- Reject — marks it rejected. Nothing is executed automatically either way.
- Later, come back and click Record realized outcome on an accepted (or rejected) recommendation. Add an optional note and choose It worked or Didn't work. That answer becomes a new learning signal — "the engine just learned from this."
Tips & limits
- Recommendation statuses: Proposed (needs your decision), Accepted, Rejected, Superseded (replaced by a newer proposal). Realized outcomes show as Worked / Didn't work badges.
- The engine distills lessons with a short AI call in the background; it uses a small amount of AI credits automatically. If your organization's AI toggle for the Learning Engine is off (or credits run out), events still flow in with their raw details — you just won't get AI-polished lessons.
- Recommendations are pattern-matched from your own history, so the module is quiet at first. The more experiments, decisions, and deals you close with recorded results, the smarter it gets.
- Similarity is shown as a "% match" — only sufficiently similar past actions generate recommendations.
Frequently asked questions
Why is everything empty?
The engine only has something to say once you've completed actions with recorded results. Start by finishing a GTM experiment with a result — the empty state says exactly that: "Complete a GTM experiment with a result. When the engine finds similar past actions, it proposes a source-linked recommendation here."
Does accepting a recommendation do anything automatically?
No. Accepting just records your decision. Nothing is executed — the reject dialog says it plainly: "Nothing is executed automatically."
Where do the lessons come from?
From your own completed work: GTM experiments, Decision Loop reviews, closed investor deals, hiring outcomes, and completed Outcome Workflows. You can't write standalone lessons here — the engine's value is that every lesson traces to something that actually happened.
Why should I record whether a recommendation worked?
It closes the loop. Your "It worked" / "Didn't work" answer is stored as a fresh learning event, so future recommendations get sharper.
Is the engine learning from other companies' data?
No. Matching and recommendations are computed only from your organization's own records.
Can I trust a recommendation?
Check its Sources box — every recommendation links to the specific past events it's based on, with match scores. If the sources don't convince you, reject it.
Still need help?
Our team answers every message. Already a customer? Open a support ticket from inside Nautis.