RecoveryAgent
Your CRM tells you what your team recorded. RecoveryAgent helps you find out what actually happened.
Businesses invest heavily in CRMs, sales processes, marketing and reporting. Yet a CRM is only as reliable as the information entered into it. Busy teams take shortcuts. Loss reasons become approximations. Follow-up gets marked complete without capturing what the customer experienced. Over time, management dashboards can remain internally consistent while revenue quietly leaks.
RecoveryAgent is a human-controlled recovery and verification layer over the existing CRM.
Its job is not to replace Salesforce, HubSpot, a broker CRM or any other system of record. It finds unresolved opportunities worth investigating, reconstructs the available context, helps a skilled human reconnect with the customer, captures what the customer says actually happened, and turns those confirmed outcomes into evidence the owner can use.
The core idea: Verified Customer Truth
RecoveryAgent distinguishes between:
- Recorded Truth — what the CRM and internal systems say happened;
- Verified Customer Truth — what the customer directly confirms through a meaningful human interaction linked to the original opportunity;
- Truth Delta — a material difference between those two accounts;
- Operational Drift — repeated verified evidence that the process customers experience differs from the process management believes is being executed.
AI cannot create Verified Customer Truth. The customer provides the signal and the human draws it out. AI does what it is strongest at around that interaction: reconstructing context, prioritising cases, identifying contradictions, structuring evidence and finding patterns across many outcomes.
AI dissects. Human draws it out. RecoveryAgent verifies it.
How the current product works
- Find — import a governed pipeline export and identify unresolved opportunities.
- Prioritise — deterministic TypeScript scoring ranks recovery priority. Gemini does not control the score.
- Reconstruct — RecoveryAgent builds bounded historical context while treating CRM notes as evidence, not gospel.
- Diagnose — Gemini reasons over model-safe context, states uncertainty and recommends the recovery approach.
- Reconnect — a governed contact rail supports contextual outreach designed to establish a real human conversation, not automate a sales pitch.
- Stop on signal — any meaningful customer response immediately stops the old recovery sequence.
- Listen and confirm — the human makes the call and records the real disposition: deal now, deal later, prepare, educate, closed or suppressed.
- Recover and learn — the business can recover immediate or future opportunity while retaining structured evidence about why the original opportunity stalled or was lost.
A simple reconnect question can create more value than another dashboard:
“Good. Glad you got it done. Can I ask one thing before I let you go—what did they do that we didn’t?”
The answer may reveal a recoverable opportunity. It may also reveal that the CRM recorded “price” when the customer actually experienced a five-day follow-up gap. Both outcomes have value.
Human versus AI
The human owns the live customer conversation, regulated or professional judgement, and the final sale.
Gemini supports the human by reconstructing bounded context, distinguishing evidence from uncertainty, diagnosing likely loss or stall reasons, generating a concise 30-second reconnect script, and structuring the resulting learning.
Deterministic application code—not the model—owns consent precedence, suppression, scoring, contact sequence state, permissions and final recorded disposition.
Why this matters for small businesses
The fastest way to fix a commercial problem is often not another layer of technology. It is reducing the distance between the owner and the customer who paid them—or paid the competitor.
RecoveryAgent makes that investigation repeatable.
Lead recovery is the wedge. Revenue recovery pays for the investigation. Verified Customer Truth is the asset.
Current build position
RecoveryAgent currently includes CSV ingestion, deterministic scoring, organisation isolation, consent and suppression controls, Gemini diagnosis and reconnect support, an exact eight-attempt human contact state machine, response-stop logic, human-confirmed disposition, Gmail draft and calendar handoff foundations, and an auditable evidence trail.
The next evidence layer aggregates confirmed outcomes into Truth Delta and Operational Drift reporting across a cohort. That aggregate management dashboard is not claimed as complete production functionality yet.
The public product site is recoveryagent.au. Synthetic demonstrations are explicitly labelled synthetic; commercial, customer and revenue claims are only treated as verified when supported by real evidence.
Built With
- firebase
- firestore
- gemini-api
- google-cloud
- next.js
- typescript
Log in or sign up for Devpost to join the conversation.