Week one
Shadow
It watches. You work.
Tensvia drafts replies you never have to use. You keep running your business exactly as you do today while we measure your real response time.
- Sends without you
- Nothing
Speed-to-lead, managedYour next customer messaged you at 8pm.
Your next customer messaged you at 8pm.
Answer them at 8:01.
Tensvia answers every inquiry across Instagram, text, WhatsApp and email — in your voice, with your prices and your hours. You approve every reply until you decide you don't need to.
No cold spam. No bans. Nothing sends without you.
Instagram · new inquiry
8:04 PMhey do you do balayage? how much for shoulder length
Drafted · pricing question · high confidence
Hi! Yes — balayage on shoulder-length is $180 and takes about 2.5 hours. I have Thursday 10am or Saturday 1pm open this week. Want me to hold one?
Priced from your service list · availability from your calendar
Nothing leaves this screen until you tap.
One queue. Every channel your customers actually message you on.
- Instagram DMInbound + reply
- MessengerInbound + reply
- Text / SMSConsent-gated
- WhatsAppConsent-gated
- EmailTwo-way threading
How it works
Four steps, and you can stop at any of them.
- 01
Connect
Link the channels you actually use. About two minutes each, and nothing is read until you say go.
- 02
Teach
Your services, prices, hours and policies become the only facts Tensvia is allowed to answer from.
- 03
Approve
Every inquiry gets classified and a reply drafted. You read it, edit it, send it. Nothing moves without you.
- 04
Autopilot
Once you trust it, let the routine questions answer themselves — category by category, on your terms.
Deliberately not included:Voice callsAutonomous quotingTaking paymentCold mass campaignsReplacing your CRM
You stay in control
Autonomy is a dial, not a switch.
The real worry isn't whether AI can write a reply. It's whether it will say something stupid to your customer. So you decide how much rope it gets, and you can take it back at any point.
The pilot
Approval
It drafts. You send.
Every reply lands in a queue with the reasoning attached. One tap sends it, one tap rewrites it. This is where most owners stay for a while, and that's fine.
- Sends without you
- Only what you tap
When you're ready
Autonomy
It handles the easy ones.
You pick which categories can answer themselves — hours, pricing, availability. Anything outside those lines still comes to you.
- Sends without you
- Only your approved categories
16 hours
how long a pricing DM sat unanswered on our own account
Why this exists
We built it for our own business first.
Not a hypothetical. That message was a booking, and it went somewhere else. We built Tensvia for our own business first, then made it something you can buy.
You won't find invented percentages on this page. When the pilot numbers are real, they'll go here — and until then the audit is how you get a number that's actually yours.
Safeguards
The tools that get people banned skip all of this.
Mass-DM tools get shut down because they treat consent as a setting. Here it's enforced below the application, where nothing can route around it.
Consent enforced in the database
Not in a prompt. An opted-out contact cannot be messaged, even if something upstream tries.
Every decision is logged
What was proposed, why, who approved it, when it sent. You can always answer 'why did it say that?'
It has no credentials
The model drafts and classifies. It cannot send, spend, or reach another business's data. Sending is a separate, gated step.
Reversible by design
Drop back to Shadow mode at any point. No migration, no lock-in, and your CRM stays your CRM.
20:04:11 PROPOSE send_reply · lead#4821 · intent=pricing · confidence=0.91
20:04:11 POLICY consent=ok · window=open · risk=low → needs_approval
20:06:39 APPROVE by owner · edited=false
20:06:40 SEND channel=instagram · delivered · 2m29s after inbound
A real trail from a real reply. Every message you send has one.
Pricing
Start with the measurement, not the software.
If the audit says you don't have a response-time problem, we'll tell you and you can stop there. That happens, and it's a fine outcome.
Workflow audit
$400one time
We measure what a slow reply is actually costing you.
- Two weeks of real response-time measurement
- Every inbound channel reviewed
- A written before-picture with your own numbers
- If the number isn't bad, we tell you and stop
Implementation
$750–2,000one time
Your channels connected, your knowledge loaded, your queue live.
- Channel setup and consent configuration
- Services, pricing, hours and policies loaded
- Approval queue tuned to how you work
- Shadow → Approval rollout with you in the room
Ongoing
$75–250per month
Monitoring, tuning, and someone who picks up the phone.
- Drafting quality reviewed monthly
- Knowledge kept current as prices change
- Deliverability and consent monitoring
- Model and messaging usage billed at cost
Messaging and model usage billed at cost, never marked up.
FAQ
The questions owners actually ask.
No. Tensvia only replies to people who messaged you or opted in. Consent lives in the database, not in an instruction the model could talk its way around. Opt-outs are permanent and enforced before any send.
In Shadow and Approval mode it cannot send anything at all — you do. It also answers only from the services, prices and hours you loaded, so it has nothing to invent from. When it isn't confident, it escalates instead of guessing.
Keep them. Tensvia plugs into what you have and writes back to it. Replacing your system of record is explicitly out of scope, and we'd rather say that up front than discover it in month two.
ChatGPT can't see your Instagram DMs, doesn't know today's prices, doesn't log what it sent, and won't stop you from texting someone who opted out. Those four things are the entire job.
Automation fires a fixed sequence. Tensvia reads the message, decides what kind of message it is, and routes the ones it shouldn't touch to you — with its reasoning attached so you can check its work.
The audit is $400. Implementation runs $750–2,000 depending on how many channels you use. Ongoing is $75–250 a month. Messaging and model usage are billed at cost, not marked up.
Safeguards
The guardrails are the product.
Anyone can generate a reply. The work is in what stops one from going out. These are mechanisms, not badges — we hold no SOC 2 report or ISO certificate yet, and we would rather say so than imply otherwise.
Consent lives in the database
Not in a prompt the model could be talked around. Consent state is a row, checked before every send, and it fails closed — a missing record is treated as no consent, never as unknown.
Opt-outs are permanent
STOP, unsubscribe and revocation are recorded and enforced before any outbound send, on every channel. There is no campaign setting that overrides them.
The agent proposes, a person disposes
Until you move a workflow to autonomy, every message is drafted and held. Approval records who approved it. Execution is a separate step after approval, not part of it.
Every draft carries its provenance
Replies cite the records they were drawn from — your services, prices, hours, policies. Nothing ungrounded is presented as grounded, and you can open the source behind any sentence.
Credentials are references, never values
Provider keys are stored as vault references. They are never written to a log, never placed in model context, and never sent to the browser.
One tenant cannot read another
Isolation is enforced at the database with row-level security, not by application code remembering to filter. If the application logic fails, the database still refuses.
Refusals are logged beside successes
Held, skipped, opted out, retried. A feed that only shows what worked teaches you nothing about whether the guards are working.
Your data leaves when you do
Export on request in a machine-readable form, and deletion within 30 days of termination except where we are required to retain records.
Messaging compliance is shared: we build the enforcement, you are responsible for how contacts entered your list. Our Acceptable Use & Messaging Policy sets out where that line sits.
Your next customer is in your inbox right now.
Two weeks of measurement, one honest number, and a straight answer about whether this is worth doing. $400, and we'll tell you if it isn't.
- No contract past the audit
- Nothing connected until you approve it
- You keep every number we produce