Everything you need to run your workspace
Practical guides written by the people who build the product. Most teams are operational after the first three — and the diagrams below explain how the whole thing fits together before you open any of them.
Four jobs, done end to end
Everything in these docs serves one of these four. If you are trying to work out which guide you need, start by working out which job you are doing.
Every lead, every channel
Meta and Google Ads, WhatsApp, website forms, IVR calls, CSV and API all land in one inbox, de-duplicated on arrival.
Nothing goes cold
Overdue callbacks, stale leads, SLA breaches and unowned leads are detected and pushed back into someone’s queue.
Know before you call
Calibrated scores, conversion probability, demand forecasts and statistical insights — trained on your own data.
Work, don’t hunt
Call queue, pipeline, automations, campaigns and reports built around the next action rather than around data entry.
What happens to a lead, in order
Six steps between an enquiry arriving and a deal closing. Each one has its own guide below.
- 01CaptureAd, form, WhatsApp or call
- 02De-dupePhone and email matched on entry
- 03ScoreSignals become a probability
- 04DistributeRight agent, right centre
- 05EngageCall queue, WhatsApp, email
- 06ConvertDeal, revenue, attribution back
The first three guides
From an empty workspace to routing live leads. Most teams are operational by the end of these.
Your first 30 minutes
Get from an empty workspace to routing live leads, in the order that actually works.
Importing your leads
CSV import with column mapping and duplicate detection — and how to avoid the two mistakes everyone makes.
Lead distribution
Routing leads to the right person in seconds — round robin, weighting, territory and caps.
Getting every lead in, and refusing to let any of them rot
Capture is the easy half. The half that decides whether the CRM pays for itself is what happens to a lead nobody got round to calling.
Open leads nobody has touched. Ranked by score, so the most convertible stale lead surfaces first.
A follow-up somebody committed to and did not make. It goes back to the top of the queue, not into a report.
The most expensive failure in the list — a lead the scorer rates highly that no agent has dialled.
Per source and per centre. A lead that waited longer than you promised is flagged against the rule it broke.
Deliberately held leads wake themselves back up on their own date. Parking is not a place leads go to die.
Detected and re-distributed. A lead with no owner is a lead with nobody accountable for calling it.
Predictions you can defend, not a chatbot bolted on
Most of the intelligence here is statistical and costs nothing to run. The language model is used for the part only a language model can do — and it is cached, scoped and read-only.
- Lead scorer
- Insight engine
- Forecaster
- Business health score
- Note signals
- Logistic regression model
- Calibration curve
- Source conversion rates
- Nightly retrain
- Note feature extraction
- Enquiry enrichment
- Call-prep brief
- Ask AI
- Natural-language search
- Sourcewhich channel it came from
- Recencyhow fresh the enquiry is
- Engagementcalls made, replies received
- Note signalsintent, objection, urgency
- Re-enquirywhether they came back
- Profilebudget and timeline, when asked
- Lead scorer
- 0–100, and it will show its working
- Trained model
- logistic regression over your closed leads
- Calibration curve
- the historical rate of this score band
- Honest nulls
- no history means no number
probability
- Lead score82 / 100
- Source rate31%
- Next actionCall now
- 1LiveCallThe IVR or app call is logged with duration, talk time and direction
- 2LiveRecordThe recording is stored and playable, behind its own permission
- 3NextTranscribeSpeech to text, once per call, stored against the note
- 4LiveUnderstandSentiment, objection, intent and urgency extracted from what was said
- 5LivePredictThose signals feed the lead score and the conversion probability
- 1PlanThe model chooses tools from a fixed whitelist. It never writes SQL.
- 2GuardTool allowed? Admin-only tool? Row-level scope applied?
- 3ExecuteThe service builds and runs the query — the model only supplied arguments.
- 4AnswerThe result is narrated, with a real lead table and the tools it used.
Where the money came from, where it leaked, what happens next
Reports answer what happened, insights say what changed and why it matters, and the forecast says what is coming — with the error rate attached.
- Direct / walk-in27%
- Google Ads24%
- Meta Ads20%
- WhatsApp10%
- Website10%
- Other9%
How the platform is put together
Useful if you are integrating with the API, evaluating the product technically, or working out where a setting lives.
The complete feature map
Every area of the product in one place. If something here has no guide yet, ask support and we will write it.
It refuses to lose a lead
Duplicates are caught on entry. Cold leads, overdue callbacks, SLA breaches and unowned leads are detected nightly and pushed back into a queue with the reason attached.
Its predictions are defensible
Scores are calibrated on your own conversions. Forecasts ship their own error rate and are rejected when a naive baseline beats them. With no history, the answer is a blank rather than a guess.
The AI cannot go rogue
It never writes SQL, never writes data, picks only from a whitelist, and is filtered by the same row-level scope as the person asking. Every AI feature degrades to a deterministic path.
The full documentation index
Grouped in the order a new workspace works through them.
Getting started
Leads
Working leads
Team & access
Automation
Reporting
Integrations
Billing
Developers
Cannot find what you need?
Our team answers documentation questions the same way we answer support ones — properly, and within a business day.