Case Studies
Built. Measured. Live.

A 12-person B2B sales team in Delhi NCR was holding 3–4 meetings per day across their team — internal pipeline reviews, client calls, and cross-functional syncs. After every meeting, the same process played out: someone had to manually write the MOM, figure out who owns which action item, assign tasks via WhatsApp or email, and then follow up when deadlines approached.

On average, this was consuming 45 minutes per person per day — time that belonged to selling, not administration.

The issue wasn't effort — the team was working hard. The issue was that approximately 40% of tasks committed to in meetings were not being tracked anywhere. No one had a definitive view of what was committed to, who owned what, or when things were due. Deadlines were missed. Follow-ups were ad hoc. The MOM quality varied entirely by who wrote it that day.

A tool like Google Keep or a shared Notes doc had been tried. It required manual effort at exactly the moment (post-meeting) when attention was already shifting to the next thing. It didn't hold.

Pravah AI is an offline meeting recorder that runs on any device during the meeting. After the meeting ends, it transcribes the conversation, identifies key decisions and action items using an AI model trained on structured meeting formats, generates a clean MOM, and assigns each action item to the correct person in their Google Calendar or Microsoft Calendar — with full deadline context.

Reminders are automatically scheduled at 48 hours, 24 hours, and 6 hours before each deadline. They go to both the task owner and the meeting host. No one has to remember to send them. No one has to check a separate app.

The entire system runs offline during the meeting and syncs when connectivity is available. It requires no new app installations for the team — they continue to work in Google Calendar or Outlook as before. The output lands there.

The Workflow Audit took 90 minutes and identified the meeting admin workflow as the highest-cost manual process in the team's day. The proposal was delivered within 48 hours. Build and deployment took 3 weeks, with weekly working demos against the team's real meeting data throughout. No training was required — the output format was agreed in week one and has not changed since.

Tracked over 30 days against the prior 30 days using the team's own calendar and task data as the baseline. Task miss rate dropped from approximately 40% to near-zero. Average post-meeting admin time dropped from 45 minutes to under 3 minutes (reviewing and confirming the auto-generated output). The system has been running continuously since deployment with no manual intervention required.

A high-volume consumer business was receiving hundreds of leads per day across multiple channels. Every lead — regardless of quality, intent, or readiness — was being routed directly to a human counsellor for L1 engagement. The model was expensive and the throughput was constrained by the number of counsellors available.

Counsellors were spending the majority of their time on leads that showed no realistic conversion potential. The high-potential leads were not receiving faster or better attention — they were in the same queue as everyone else.

The L1 cost per lead was high because human time was the only resource being applied. There was no mechanism to pre-qualify, segment, or route leads before a counsellor engaged. The business was effectively paying human conversation rates for what was, in many cases, initial data collection and intent assessment.

The challenge was that fully removing the human from L1 was not acceptable — the brand required a warm, personalised experience. The solution needed to automate L1 without feeling automated to the prospect.

Taskentis is an AI counsellor built and trained on over 22,000 real call recordings from live counselling sessions. It handles initial lead engagement, qualification, and routing autonomously — collecting structured information through a natural conversational interface, assessing readiness signals against a defined qualification framework, and routing only high-potential leads to live counsellors with full context. The training dataset is what separates it from generic AI: the model has learned from real human counsellors handling real prospects, not from scripted flows.

When a lead is routed to a human counsellor, the handoff includes complete context — the lead's responses, assessed readiness level, and suggested conversation entry point. The counsellor starts from informed ground, not from scratch.

Leads that do not meet qualification thresholds are handled by the system through nurture sequences, with re-routing to human counsellors triggered by intent signals.

The pilot ran for 6 weeks with a control group receiving the standard human L1 process and a test group handled by Taskentis. Both groups were drawn from the same lead sources with equivalent lead quality distribution. Cost per qualified lead, conversion rate through L2, and overall funnel cost were the primary metrics tracked.

L1 cost per lead reduced by 60% in the pilot versus the control group. Overall funnel cost (end-to-end) reduced by 7%. Conversion rate through to L2 was maintained — high-quality leads were not being lost in the automated layer. Human handoff quality, assessed by live counsellors, was rated higher with Taskentis than with the standard process — because counsellors were receiving structured context on every lead rather than starting cold. The 22,000+ call recording training set is the primary reason for this: the model understands how real counsellors probe, handle objections, and qualify intent.

A counsellor team handling live inbound calls was working with a generic script — the same pitch, the same niche suggestions, the same framing regardless of who was on the call. Initial interest conversion was flat, and there was no clear lever to move it without a major training overhaul or script restructure.

The counsellors were skilled — the problem was not their ability, but their information. They were making the same pitch to every person because they had no real-time way to personalise it in a 5–7 minute live call.

Personalisation requires knowing something about the person you're talking to and having a relevant recommendation ready before the conversation loses momentum. In a live call environment, there is no time to research, think, or look things up. Counsellors were defaulting to what they knew — which was the generic script — because it was the only thing they could reliably access in real time.

Changing the script or running training programmes was considered but would have required significant time, consistency of execution was uncertain, and it wouldn't adapt to the individual person on each call.

Counsellor Bhaiya is a custom AI assistant that generates a personalised niche career path and a tailored pitch line in real time, from 4 simple inputs entered by the counsellor during the call. The inputs take under 30 seconds to enter. The output — a specific career niche recommendation and a one-line personalised pitch — is returned in seconds.

The counsellor uses this output immediately in the live conversation. No script change. No new process. No disruption to the call flow. The existing workflow was kept entirely intact — only the quality of information available to the counsellor in the moment changed.

The system was built as a simple, fast interface optimised for use during a live call — minimal inputs, immediate output, no cognitive overhead for the counsellor.

Tracked over 30 days with a control group using the standard script and a test group using Counsellor Bhaiya. Both groups were matched by lead source and inbound volume. The primary metric was initial interest conversion rate — the percentage of calls where the prospect expressed interest in proceeding to the next step.

Initial interest conversion rate increased by 19% in the test group versus the control group over the 30-day measurement period. This improvement required zero changes to the existing workflow, zero additional training, and zero disruption to the live call process. The counsellors using Counsellor Bhaiya reported higher confidence on calls and lower cognitive load — they felt better prepared for the specific person they were talking to.