The dead zone — and the copilot that lives inside it.
Most SaaS companies have customers who pay enough to hurt when they churn, but not enough to deserve a human. The product team never finds out why they left. Here's what to build instead.
DraftFor sparring with friends5 min readWorking sketch inside ↓
01 · The woundCustomers in the middle
Every SaaS product has a dead zone. Customers pay enough that losing them hurts — $100 to $1,000 a month — but not enough to justify a human CSM holding their hand. So when they hit friction inside the product, they're on their own.
They Google. They open a chat bubble and get a doc snippet. They give up. They churn quietly, and the company finds out three months later from a dashboard.
The friction is rarely "I have a question." It's "I'm trying to do a thing and I'm stuck." A connection string that won't connect. A permission setting that doesn't behave the way the label suggests. A workflow that needs three settings turned on in the right order.
These are not support tickets. They are activation and adoption problems — and they decide whether the customer ever gets to value.
02 · The status quoWhy today's tools don't solve it
Chat widgets (Intercom, Crisp) wait to be asked, reply with FAQ snippets, and have no idea what's on the user's screen.
Product tours (Pendo, Appcues) are scripted and brittle. They assume the user follows the happy path. They don't react to a user actually struggling.
Docs chatbots (Inkeep, Kapa) ground answers in documentation but can't see what the user is doing, can't act on their behalf, and can't tell when help is even needed.
All three are reactive, generic, and blind to context.
The product is the entire support surface. Treating it like a help center bolted onto an app is why activation stalls and quiet churn keeps showing up at QBRs.
— what I keep telling myself when I see the third "where do I…" message this week
03 · The ideaAn in-product copilot
A copilot that lives inside the customer's product and behaves like the smartest CSM you've ever hired — except it's there for every user, on every page, at all times.
It does three things no existing tool does together:
It sees the screen. Not as a screenshot — as structure. It knows which page the user is on, which fields they've filled, what errored, what they've retried five times in a row.
It knows the user. Plan, role, history. What they've already done. What they were supposed to do next.
It acts, not just answers. "Want me to walk you through it?" → it highlights the right field, narrates the steps, and where it makes sense, does it for them with confirmation.
And it leads with voice — because typing while stuck inside an app is the friction. Talking is the natural escape hatch.
i
The widget below is a real, interactive sketch — not a screenshot. Try entering a bad connection string and clicking "Test connection" twice. Or just click the helper to start a session.
· · ·
Tip: click Test connection a couple of times.
app.quanta.io / integrations / new
IntegrationsNew connection
AK
Add a data source
Choose where your data lives. Quanta will sync incrementally every 15 minutes once connected.
P
Postgres
Most common
❄
Snowflake
OAuth
B
BigQuery
Service account
S3
S3 bucket
IAM role
Connection stringRequired
Standard libpq URI. We'll never log this.
SchemaOptional
Connection failedHost db.internal is unreachable from Quanta's network. This usually means the database is on a private network and needs IP allowlisting.
04 · The compounding side effectA friction telescope
Every session is a labeled data point: "this user got stuck on this screen, doing this thing, because of this gap." Aggregated, it becomes a friction telescope for the product team — exactly which steps are losing users, what they're confused about, and what to fix next.
The copilot saves the customer. The data improves the product. Both compound.
Here's what the company sees on their side:
Copilot insightsQuanta · production
Last 7 days
Sessions
1,247
+22%
Resolved without human
89%
+4 pts
Avg. session
2m 18s
−18s
CS hours saved
~62
+11 hrs
Top friction points
Volume · last 7 days
/integrations/newPrivate network · IP allowlist
287
/billingMid-cycle plan switch
164
/api-keysScope vs. role confusion
121
/onboarding/step-3Workspace setup skipped
78
✦
Suggested fix
Add inline IP allowlist instructions on the Postgres connect screen. 79% of these sessions are private-network customers reaching the same conclusion the Copilot did.
Voice models are finally fast enough for real-time conversation (sub-400ms round trip).
Multimodal LLMs can reason over a live DOM the way a human reads a screen.
Agent infrastructure is mature enough to take safe, scoped actions inside a third-party app.
The PLG cohort — thousands of $100–$1,000 MRR SaaS companies — has the budget, the wound, and a CSM headcount they cannot grow proportionally.
06 · The betWhat this is, really
The companies in the dead zone don't need cheaper support. They need an in-product copilot that turns confused users into successful ones — automatically — and tells the product team exactly where to dig next.
That's the wedge. Build the copilot. Earn the data. Compound both.