Talent profiles v2: findings and proposal
What v2 of the talent profiles could look like, based on a full walkthrough of the portal as a member. Covers the main problem v2 should solve, a proposed flow, what to borrow from foundit's member emails, lessons from building AI interviews at Iterative, and quick fixes.
Where it is today
Working well
- The profile generator gets a member from LinkedIn to a structured profile in minutes
- Personas, superpowers and the user manual describe how someone works, which a CV doesn't
- Consistent, calm UI with no errors on any page
- Coaching directory is clear and adds value for members
Gaps
- Profile quality depends on how complete the member's LinkedIn is
- Some generated facts are wrong, and nothing asks the member to check them
- After setup, members see no status, so there's little reason to come back
- Job Preferences asks again for things the profile already knows
The main problem: trust and depth
Profiles are what Fractional First puts in front of clients, so they need to be accurate and deep enough to match on. Today the generator can only work with what's on LinkedIn, and it sometimes gets facts wrong. One generated profile listed a French certification register (RNCP) as a Computer Science qualification from SUTD.
A client who spots one wrong fact will doubt the rest of the profile. v2 should focus on making profiles trustworthy and deep, more than on how they look.
Proposed v2 flow
Steps 1 and 2 exist today. Steps 3 to 6 are new.
- 1
Member connects LinkedIn
Same as today.
- 2
AI drafts the profile
Same as today, but the draft also lists what's thin: roles with no outcomes, missing team sizes, unclear scope.
- 3
Async AI interview fills the gapsNew
Short, targeted questions based on what's missing, e.g. "You list Partior but nothing on team size or results. What did you own?" This saves the most team time, because the human interview can start from a full picture. Lumina Aria runs async interviews like this for job applicants, so it's a useful reference for the format.
- 4
Member confirms the factsNew
Before publishing, the member checks roles, education and certifications. Each fact is marked confirmed or AI-written, so clients and the team know what to rely on.
- 5
Signals with evidence strengthNew
Instead of one score, show each signal with how well it's backed up. See the example below.
- 6
Member sees their statusNew
Reviewed, in matching, shortlisted. Even a simple status gives members a reason to come back and keep their profile current.
Example: signals with evidence strength
There's no universal formula for a good fractional CTO or CFO, and there doesn't need to be one. Fractional First can define the signals it cares about and show how strong the evidence is for each. That's the "confidence interval" idea, made concrete.
| Signal | Evidence | Source |
|---|---|---|
| Player-coach | Strong | Interview answer with a specific example, plus a matching LinkedIn role |
| Scaled a team from early stage | Medium | Stated in the interview, no team size given |
| Board or investor-facing experience | Weak | Inferred by AI from job titles only |
Illustrative example, not real member data.
Reference: foundit's member emails
foundit (formerly Monster APAC) emails me as a job seeker, about 200 emails so far, 31 of them in the last six days. Some of the mechanics are worth copying for steps 4 and 6. The volume and the matching are what Fractional First should avoid.
Worth borrowing
- One fact, one button. "Confirm your details" shows a single fact (current title and company) with one button to confirm it. That's step 4 at its smallest.
- The role at a glance. "You're invited to apply" lists skills, years of experience and location, then links to a one-click apply.
- Nudges that give a reason. "Recruiters often filter candidates by location" explains why a field matters. "You're shortlisted for 2 open roles" works as a status update.
Worth avoiding
- Repeats. The same confirm-your-details email arrived five days in a row, and job emails come back as "Follow up 1" and "Follow up 2".
- Matches with no reason. A Venture Engineer profile gets a Technician role at S$2,800 to S$3,000 a month and a Scrum Master role, and nothing says why either one fits.
- Pressure and upsell. Subject lines like "Your profile is taking a hit", plus a 20% off offer for a paid career package.
| Sent | |
|---|---|
| "…looks great, just confirm these details" | 21, 22, 23, 24 and 25 Sep |
| Technician role, S$2,800 to S$3,000 a month | 22 Sep, then "Follow up 1" and "Follow up 2" on 23 and 24 Sep |
| Cloud Engineer role | 21 Sep, then two follow-ups on 22 and 23 Sep |
| "Your profile is taking a hit" | 20 and 22 Sep |
| "Recruiters are noticing your profile" | 21 and 23 Sep |
From my inbox, 20 to 25 September 2026.
For Fractional First
Borrow the mechanics, not the volume. Ask members to confirm one fact at a time, show roles at a glance, and send a status update only when something has changed. Every match should say which signals it's based on, using the evidence from step 5. For a curated network, fewer and better messages are the point.
Lessons from building AI interviews at Iterative
At Iterative I built AI voice interviews that screened founders for the Validation Program and the Global Program, and scoped moving first calls in the main funnel to AI. The Global Program bot got 4.46/5 in founder feedback. What carries over to step 3 and step 5:
- Run the interview from a fixed agenda, not an open chat. Each step has a goal, a time limit and a structured output, so every interview produces the same fields.
- Score against a written rubric with weighted positive and negative signals. Every score cites quotes from the transcript, and with no quote the score defaults to the lowest level. That rule is what makes the evidence strength in step 5 believable.
- Calibrate on real examples. Test the scores against interviews the team has already judged before relying on them.
- Keep a human in the loop. Founders worry about being judged unfairly by an AI, so the team should review the scores and people should be able to ask for a human call instead.
- Write the results straight into the system of record. Scores, summaries and transcripts went into Airtable automatically, so nobody copied them by hand.
- Start on an existing tool, but own the rubric and the data. The first version ran on a no-code platform and shipped fast. Its limits on prompts, model choice and voice quality are why it's now being moved to a different setup. Keeping the rubric and data outside the tool is what makes that move possible.
Quick fixes
Small bugs found during the walkthrough. Each one is a short fix.
Rate range reads "$100 to $0"
Job Preferences › Compensation
With Maximum empty, the helper text ends the range at $0. Show "$100+ per hour" instead.
"Profile complete" while three steps are still open, plus a "Company" placeholder
Dashboard
The header says complete, but Publish, Preferences and Agreement still look undone and the checklist has no ticks. The profile card also shows the literal word "Company".
Preferences don't prefill, and the agreement refers to empty fields
Job Preferences
Location, work eligibility, preferred locations and industries start empty even though the profile lists them. The confirmation asks members to agree they can work "in the countries listed in my profile" when none are listed, and the tick box is drawn as a radio button.
Job Preferences overflows on phones
390px phone width
The page is 435px wide on a 390px screen, so text is cut off. The Edit Profile bottom bar also crowds the screen and its last button runs off the edge.
Other fixes for the backlog
- Label the icon-only edit buttons and the Job Preferences controls for screen readers. The public profile also repeats every heading from Education to Certifications.
- Add a confirmation before Regenerate, which sits next to Publish and looks like it would overwrite manual edits.
- Keep the sidebar on Edit Profile so members don't leave the app layout.
- Ask for the current password when changing it, and show password rules.
- Add per-page titles and a preview image and description to shared profiles, so links look right on LinkedIn and WhatsApp.
- Compress the coaching images (about 3.9 MB of PNG in total).
- Rename
/dashboard/brandingto match "Professional Coaching", and fill in the blank title on coaching detail pages.
Suggested next steps
- Ship the quick fixes above.
- Agree on the first 5 to 8 signals Fractional First cares about, and what counts as strong evidence for each.
- Prototype the gap-filling interview (step 3) with a small group of members, on an existing tool, with the rubric and data kept in Fractional First's own systems.
- Add fact confirmation and member status, which don't depend on the interview.
Pages were viewed as a logged-in member only. Nothing was submitted or changed. Screenshots are from a real member account.