What it does
Four jobs, all of them specific
Briefs you on a lead
Six call notes, three updates and two site visits, read and reduced to where the lead stands, what could lose it, and the single next action. It comes with a draft message the agent edits before sending.
Decides who takes a new enquiry
A website enquiry is routed the moment it arrives, to a telecaller who is punched in. The model weighs service area, how many leads they already hold today, and how long since their last call, then says why it chose them.
Writes listing copy from the form
The description, the page title and the search snippet, built only from facts already entered. No invented views, no invented distances, no 'nestled'.
Tags leads for searching
Area, property kind, budget band and urgency, as short tags produced by the same call that does the routing. They are searchable across the whole CRM, so 'apartments in Rajbagh under a crore' is a search rather than a conversation.
The rules we hold it to
Constraints are the feature
An AI you cannot predict is an AI you cannot delegate to. These are the limits it runs inside, and they are enforced in the code rather than promised in a policy.
Your clients' numbers stay here
Phone numbers and email addresses are excluded from every prompt by construction, not by policy. The model sees requirements, budgets, stages and outcomes.
Every request is audited
Who asked, which model, how many tokens, what it cost and whether it worked, including the failures. A button that quietly stopped working looks identical to one that was never pressed unless the misses are written down too.
It suggests; a person decides
Nothing it writes is published without somebody reading it. Routing is the one exception, because that has to happen in under a second at three in the morning, and it is recorded, reversible and explained.
It is allowed to say it does not know
A thin record produces a brief that admits the record is thin. A model that always sounds confident is a model you stop being able to use.
The lead score is not AI, and the screen says so
BluPlots scores leads and forecasts the pipeline. Both are arithmetic: budget, timeframe, temperature and whether there is a way to contact them, weighted and added up. The forecast is a stage-weighted average. There is no model in either, and no training data.
We could have called it AI. Most products would. But a number is only useful if you believe it, and the fastest way to stop believing a number is to discover it was sold to you as something it was not. The page it lives on is titled “arithmetic, not a model” for that reason.
What that buys us is the right to be believed about the parts that are a model.
When it is unavailable
What happens when the model cannot answer
Every AI feature has a defined behaviour for the day the account runs dry or the provider has an outage. None of them is a spinner that never stops.
Routing keeps working
It falls back to deterministic workload rules, so a lead arriving during an outage still reaches a human immediately. The assignment records which path chose it.
Briefings say so
The panel reports that it could not run and why, whether no credits, rate limited or provider down, rather than showing an empty box.
Nothing is silently skipped
A failed call still writes its audit row, so a month of quiet failures is visible on the usage page instead of invisible.
Questions
The AI
What does the AI actually do?
Three things. It briefs you on a lead by reading its calls, updates and visits. It drafts listing copy from the facts already on the form. And it decides which telecaller should take a new enquiry, from service area, current load and how long since their last call.
Does it see our customers' phone numbers?
No. Phone numbers and email addresses are excluded from every prompt, deliberately and by construction. The model sees requirements, budgets, stages and call outcomes.
Does it decide anything on its own?
It routes leads on its own, because that has to happen in under a second at three in the morning. Everything else suggests and a person decides. Nothing it writes is published without somebody reading it.
What happens if the AI is unavailable?
Routing falls back to deterministic workload rules, so a lead arriving during an outage still reaches a human immediately. Every assignment records which path chose it. The briefing and copywriting buttons simply say they could not run.
Are the lead scores AI?
No, and the screen says so. The score is arithmetic: budget, timeframe, temperature and whether there is a way to contact them. Calling that machine learning would be a lie, and a number you do not believe is worse than no number.
What does it cost to run?
It is included. Every call is logged with its model, its tokens and its cost, and an administrator can see the running total for the month on the AI page.
See what it says about your pipeline
Fifteen days free, the whole product, no card. Bring a hundred real leads and ask it what is going wrong.