The corridor
The four moments where a signed-up customer goes quiet
Drop-off is not evenly spread. It clusters at the moments below — and at each one the assistant’s intervention is different, because the reason the customer stalls is different.
The half-finished sign-up
The customer starts the form, hits a field they cannot answer from their phone, and leaves to find it — often for good.
The assistant saves the attempt, says exactly which one thing is still needed, and reopens the same form rather than a blank one when the customer returns.
- Sign-up flow
- CRM
The document that never arrives
A request for “supporting documents” lands in email, unclear about which documents, what format, or why they are needed at all.
The assistant asks for each item by name with the reason attached, accepts a phone photo where policy allows, and confirms receipt so the customer knows the ball has moved.
- Document store
- Messaging
The verification silence
Identity or approval checks run in the background for days, and the customer hears nothing — the single largest driver of abandonment and of “where is my application” calls.
The assistant reads real status from the systems of record and narrates it — received, in review, one item outstanding — without ever performing or promising the decision itself.
- KYC provider
- Case system
Approved, then never used
The account is open, the app is installed, and nothing happens — the customer never reaches the first action the product was bought for.
The assistant proposes the one next step that produces first value for this customer’s stated purpose, and hands anyone who hesitates to a human welcome call.
- Product
- CRM
Why silence loses customers
Applicants rarely quit a process; they quit not knowing where they are in it
Watch a real onboarding funnel and the pattern is consistent: the steps that lose customers are seldom the hard ones. They are the quiet ones — the days after documents are submitted, the wait after a check begins — when the customer cannot tell whether the process is moving, stuck, or lost. People forgive effort; they do not forgive uncertainty. A competitor whose sign-up says less but answers faster wins applicants your process had already qualified.
The assistant’s real product is therefore not conversation but certainty: an honest narration of where the customer is, drawn from the systems that actually know, plus a chase at each moment the data says people stall. That division of labour is deliberate. The assistant reads status, explains and reminds; the identity check, the credit view and the approval remain with your compliance function and your systems of record, which is where a regulator, and a customer, would expect them to be.
The deployment
How an assistant is fitted to a funnel that already exists
The assistant is shaped by your funnel’s evidence, not by a generic onboarding template. The work runs in this order.
Measure where applicants actually stall
We instrument the existing funnel and rank the drop points by lost customers, because the intuition about where people quit is usually wrong by one step. A ranked list of the moments worth automating.
Wire the reads before the words
The assistant is connected read-only to the case system, the document store and the verification provider’s status, so everything it says about progress is drawn from record rather than guessed. Status the assistant can narrate without inventing.
Script the asks and the chases with your team
Each document request carries its reason, each follow-up its timing, and each escalation its trigger — written with the people who today answer the “where is my application” calls. A playbook the assistant executes and your team recognises.
Pilot on one segment, judge on completions
One customer segment runs through the assisted corridor while the rest continue as before, and the comparison is completed onboardings — not conversations held or messages sent. Evidence before the rollout, on the metric that pays.
Our role
What the assistant carries and what compliance keeps
- 01
The guided corridor
We build the assistant that collects, explains, narrates and chases across the funnel, in Arabic and English, on the channels your customers already use.
- 02
The boundary, engineered in
Identity verification, credit views and approvals are wired as reads and referrals, never as actions the assistant performs — the permission model makes the boundary structural rather than behavioural.
- 03
The handover to a person, timed early
A confused applicant, a rejected document or a stalled case beyond its expected window goes to your team with the full history, because the moment before frustration is when a human call still converts.
What your process requires of a customer, and who is approved, remain decisions of your compliance function and your systems of record; we build the assistant that makes meeting those requirements the easy path instead of the abandoned one.
Common questions
What buyers ask
Does the assistant verify identity documents itself?+
No, and that is a design decision rather than a limitation. The assistant collects documents, checks the obvious mechanical things — the file arrived, it is readable, it is the type that was asked for — and tracks the case. The verification itself, and any approval or credit decision, runs in your existing verification provider and case system, and the assistant only reads and narrates the status those systems report. That keeps the audit trail where your compliance function already looks for it, and it means adopting the assistant changes the customer’s experience without changing your regulatory posture.
Which stretch of the onboarding funnel should the assistant cover first?+
The stretch your own data says is bleeding most, which we measure before building anything. In most funnels the answer is one of two places: document collection, where vague requests and format problems stall applicants who intended to comply, or the verification wait, where days of silence convert pending customers into lost ones. Starting with a single stretch is deliberate — it produces a clean before-and-after on completed onboardings within one pilot, and that evidence decides whether extending the assistant across the rest of the corridor is worth the next investment.
How many reminders are too many before a human should call instead?+
Fewer than most sequences send. Our working rule is that each follow-up must add something — a clearer explanation, an easier way to submit, an honest status change — because a repeated identical nudge reads as nagging and hardens the silence. When two value-adding chases have not moved a case, or a customer expresses confusion or irritation, the assistant stops and routes the case to a person with the history attached. The applicant a human call recovers at that moment is usually one more reminder would have lost for good, and the assistant’s job is to know the difference.
In this cluster
Scope it with us
A working session that maps your funnel from sign-up to first value, marks where applicants actually stall, and scopes the assistant against those points rather than the whole journey.
Start the conversation
