Launching today
DodoForm
Turn talking, pics, or scribbles into clean, structured data
129 followers
Turn talking, pics, or scribbles into clean, structured data
129 followers
Voice, photos, messy notes — DodoForm turns however people communicate into clean structured data. 100+ templates, AI-powered analytics, native integrations, and branded forms. Done in seconds, not minutes. 14-day Pro trial, no card.
DodoForm
The “last-statement wins” rule is smart, especially because it matches how people usually listen. The part I’d be careful with is fields where the correction itself is useful context, not just noise.
For example, in sales or hiring intake, “actually no, use Wednesday” may be the final answer, but the earlier Tuesday mention can explain availability, urgency, or uncertainty. I’d love to see a lightweight audit trail for high-impact fields: final structured value, confidence, and the snippet that caused the value to change. That would make messy input feel safer without forcing everyone back into rigid form behavior.
DodoForm
@jim_jeffers You're right — "last-statement wins" is a simplification that
quietly throws away signal in exactly the high-stakes fields where
context matters most. A sales rep saying "Tuesday — actually
Wednesday, their CFO is back from leave" is telling you two things,
and only one of them ends up in the deal row today.
The audit trail you're describing is the right shape:
→ Final structured value
→ Confidence score
→ The exact transcript snippet that produced it
→ Earlier candidates that were superseded, with timestamps
For high-impact fields (close_date, budget, decision_maker,
medical history, anything regulated), that should be one click
away from the structured row — not buried, not opt-in. For low-
stakes fields (favorite color in a wedding RSVP), the trail is
overkill.
Adding "high-impact field flag" to the field schema and surfacing
the audit trail in the response detail view is now on the next
two-week list.
That field-level flag is the right call. I’d also think about a “why this mattered” label, not just the raw audit trail.
In your CFO example, the superseded Tuesday value is useful because it explains dependency/risk, not because anyone needs Tuesday in the structured row. If DodoForm can distinguish “correction because I misspoke” from “correction that reveals a constraint,” the analytics side gets much more interesting: you start learning where people hesitate, negotiate, or expose hidden blockers while still keeping the final form clean.
DodoForm
@jim_jeffers spot on. A correction is only useful if you know why it happened.
- "Tuesday — actually Wednesday" because the CFO is back = reveals a dependency
- "Tuesday — actually Wednesday" because you misspoke = just noise
If we tag corrections as `constraint` vs `misspoke`, form owners get two wins:
1. Clean final data
2. Signal on where their process actually breaks
The "why this mattered" label is going on the field flag. We'll ship the audit trail + confidence first, then layer in correction-reason tags.
This is the difference between a log and real intelligence. Thanks for the push.
the voice input angle is interesting but voice to structured data has a confidence problem. people speak in fragments, change direction mid-sentence, use filler words. curious how the AI decides what's signal versus noise when someone rambles their way through a form field
DodoForm
@ansari_adin
Three things we do, in order of impact:
1. Field-aware extraction, not transcribe-then-parse. The model knows
upfront that field X expects a datetime, field Y expects a phone,
field Z is open-ended. So when someone says "yeah Tuesday-ish,
actually no, Wednesday morning works better" — the prompt is anchored
to "what's the final intended datetime?" not "what did this person
say?" Filler words and false starts get filtered as noise because
they don't match the field's schema.
2. Last-statement wins for contradictions. If someone changes
direction mid-sentence, we bias toward the most recent declarative
claim. "Email is maya at gmail — wait no, maya at acme dot com" →
maya@acme.com. This matches how humans listen too.
3. Confidence-scored confirmation step. Every extracted field comes
back with a confidence value. Above ~0.85, it auto-fills and the
respondent sees it as pre-filled (still editable). Below that, we
show "we think you meant X — is that right?" Low confidence never
silently writes wrong data; it asks.
the “humans don’t communicate like APIs” angle is strong. i’d be curious how you handle cases where the messy answer contains ambiguity that should not be silently cleaned up, like “maybe Tuesday unless Sam replies” or “use the old address for now”.
do you surface that as a confidence/review step somewhere, or does the form owner define which fields are allowed to be inferred vs. need explicit confirmation?
DodoForm
@kar_re Good question. We do both.
1. The AI doesn't just guess a value — it also keeps the "but"
part. "Maybe Tuesday unless Sam replies" → it picks Tuesday, but
saves "unless Sam replies" as a separate note. "Use the old
address for now" → it uses the old address, but flags that "for
now" is part of the answer. Nothing gets quietly thrown away.
2. The form owner decides what happens next, per field:
- just fill it in
- fill it in but ask the person to confirm
- don't let them submit until they clear it up
- always make them type it themselves
Default is "fill it in and ask to confirm." Safe by default, and
the owner can loosen it where it doesn't matter.
Still working on showing those side notes ("unless Sam replies")
to the form owner as real data, not just a warning. Same idea as
the audit trail someone asked about earlier - they'll ship
together.
love that split between the cleaned value and the “but” note. the per-field defaults make a lot of sense too, especially starting with confirm-by-default.
making those side notes queryable/reportable feels like the really valuable bit imo, since that ambiguity is often the actual signal.
mailX by mailwarm
100+ templates is a lot, I really want to know what has been the most popular use case of this product so far.
DodoForm
@thamibenjelloun we just went live today, so
"most popular" is still early data. But the patterns we're seeing in
beta and across early signups skew heavily toward 3 buckets:
1. Sales / CRM intake — voice memos after a call → structured deal row
(the example in our hero). This was the unexpected hit.
2. Job applications & hiring intake — resume + GitHub + essay questions
with AI parsing the messy stuff.
3. Event RSVPs — wedding, conference, team offsite. Branded themes
convert noticeably better here.
The templates list will quickly evolve based on what people
actually reach for.
Stripo.email
Congrats on launch day! The free tier with a generous quota is a good call for a product like this.