BLR IST

FOOTNOTE 2, EXPANDED

RECRUITERFLOW · AI PRODUCT DESIGN2024 — ONGOING

AIRA

Building an AI suite from nothing — five products, one platform, and the discipline to say no to the ones that would have undermined the ones that mattered.

ROLE
LEAD PRODUCT DESIGNER · SOLE DESIGNER FROM DAY ONESCOPE
0→1 DESIGN · DESIGN SYSTEM · DIRECTION & STRATEGYWITH
NIHAL BHATIA, PM · AKSHAY, PRODUCT DESIGNER

IMAGE COMING

FIG. 2.0 — ASK AIRA, THE ALWAYS-ON INTELLIGENCE LAYER THE PLATFORM CONVERGED INTO.

THE CONTEXT

In 2024, Recruiterflow had a competitive problem. Customers were using five tools to do what the CRM should do in one. LinkedIn Recruiter for sourcing. ChatGPT for drafting. Zoom for calls. The ATS for job orders. Recruiterflow for the rest.

AIRA was the answer to that question — not a single AI feature, but a platform of them, built in sequence so that each one earned the next. The design challenge wasn't making AI feel smart. It was making it feel trustworthy enough to act on.

THE PLATFORM ARC

iIn-profile chatTAUGHT US WHAT RECRUITERS WANTED TO KNOW

iiNotetakerCAPTURED THE FRESHEST DATA — THE CONVERSATIONS OUTSIDE THE SYSTEM

iiiAgentsPUT THAT DATA TO WORK — AI THAT ACTS, NOT JUST INFORMS

ivAIRA SourceEXTENDED THE REACH TO MILLIONS OF EXTERNAL PROFILES

vAsk AIRABROUGHT IT TOGETHER — GENUINE INTELLIGENCE, NOT A COLLECTION OF FEATURES

FIG. 2.1 — FIVE LAYERS. EACH VERSION EARNED THE NEXT ONE.

THE IN-PROFILE CHAT

IMAGE COMING

FIG. 2.2 — FROM FOUR PROMPTS TO A CONVERSATIONAL SURFACE.

The first version gave recruiters four fixed prompts. Within a month, the top request was a text field.

WHAT THE DATA SHOWED

35k

AIRA TAB OPENS IN 90 DAYS

64%

OF OPENS CONVERTED TO A SENT MESSAGE — WORKING, NOT EXPERIMENTING

5,147

PROFILE SUMMARIES — A THIRD OF ALL ACTION CLICKS

WHERE THE DESIGN SAID NO

Every AI feature discussion included someone suggesting a confidence score. We said no. Not because confidence scores are wrong — because these users couldn't act on a number. They needed a reason.

The design constraint became a product principle: show the reasoning, not the rating. Every AI output in AIRA surfaces why it made the recommendation, not how certain it is. Recruiters can interrogate the reasoning; they can't do anything useful with a percentage.

AIRA SOURCE

IMAGE COMING

FIG. 2.3 — SOURCING BROUGHT INSIDE THE PRODUCT, FOUR TOOLS COLLAPSED INTO ONE.

The average enterprise recruiter was paying for LinkedIn Recruiter, Lusha, Apollo, and at least one ATS. We collapsed all of that into one tab.

AI AGENTS

Agents are AIRA putting its knowledge to work: sourcing candidates, drafting outreach, and moving them through stages without manual triggers. The design challenge was trust — recruiters needed to feel in control of something that was acting on their behalf.

The answer was visibility: every agent action is logged, every decision is shown, and every output can be edited before it touches a candidate. The agent acts; the recruiter approves.

ASK AIRA

Ask AIRA was where the five products became one sentence: tell me everything I need to know about this candidate, right now, so I can make a decision.

IMAGE COMING

FIG. 2.4 — REASONS, NOT SCORES. THE AI SHOWS ITS WORK.

The test wasn't whether Ask AIRA answered correctly. It was whether recruiters trusted the answer enough to act on it.

P.S. — the most valuable thing that happened to Ask AIRA was a failure users named out loud← BACK TO THE LETTER