How a Real Estate Brokerage Analyzes 900+ Sales Calls with AI

Veer Real Estate runs a brokerage operation in Gujarat with 11 sales agents. Each agent handles property inquiries, schedules site visits, negotiates prices, and makes follow-up calls across multiple projects. On a typical day, the team makes 30 to 40 calls collectively. Over a month, that is 700 to 900 conversations about properties, pricing, possession timelines, and client requirements.

Before SalesEar, Sudhir Prajapati, the firm's admin, had no visibility into any of those conversations. He knew how many calls each agent claimed to make. He did not know what they said on those calls. Whether they quoted the right price, followed up on warm leads, handled objections well, or lost deals because of a careless promise. The only feedback mechanism was asking agents after the fact, which produced filtered, incomplete answers.

Four months later, Veer Real Estate has 909 calls analyzed, 11 agents fully covered, and a management process that runs on call data instead of guesswork.

The Google Dialer Problem

The first technical challenge was device compatibility. Several of Veer Real Estate's agents use phones running Google's stock dialer: Motorola, Realme, and Pixel devices. Unlike Samsung and Xiaomi phones, which have built-in call recording that third-party apps can access, Google Dialer restricts recording access at the OS level.

Most call analytics tools either do not support Google Dialer devices or require agents to use a separate VoIP dialer app. For a field sales team where agents use their personal SIM numbers, switching to a VoIP app is not practical. Prospects know the agent's number. Call quality on Indian mobile data is inconsistent. Agents refuse to switch and revert to the regular dialer within days.

SalesEar solved this with an automated helper app that captures calls on Google Dialer devices without requiring any action from the agent. Agents make calls using their regular phone dialer. The helper detects when a call ends and automates the capture process in the background. No manual recording, no file sharing, no workflow change.

For Veer Real Estate, this meant deploying call analytics across the entire 11-agent team in one afternoon, regardless of which phone each agent used. Samsung agents got auto-capture. Google Dialer agents got the helper. Every agent's calls started appearing in the dashboard the same day.

What 900 Calls Revealed

The first week of data produced the discoveries that always appear when a team starts analyzing calls but had never seen the patterns before.

Follow-up gaps were larger than anyone thought. Sudhir assumed his agents followed up on warm leads within a day or two. The call journey data showed a different picture. Several prospects who expressed clear interest on their first call did not receive a second call for 4 to 5 days. In Gujarat's competitive real estate market, a prospect talking to a brokerage on Monday is talking to two others by Wednesday. A 5-day gap is not a delay. It is a lost deal.

By grouping every call to the same prospect into a single timeline across all 11 agents, follow-up gaps became visible the morning after they happened. The daily review shifted from "did you follow up?" (which always gets a yes) to "the dashboard shows no call to this prospect since Tuesday" (which is specific and undeniable).

Pricing inconsistencies surfaced across agents. Different agents working the same project quoted different numbers on the phone. One agent included parking in the base price. Another quoted parking separately. Both were technically stating the pricing components, but the prospect heard "85 lakh all-inclusive" from one agent and "82 lakh plus 3 lakh parking" from another and perceived an inconsistency. That perception erodes trust in the brokerage.

Call analytics flagged every conversation where pricing language appeared. Sudhir reviewed the flagged calls each morning, a 15-minute process that replaced hours of recording playback. Pricing discrepancies were caught within a day instead of surfacing weeks later when a prospect complained at the booking stage.

Agent coaching became specific. Before call data, coaching was generic: "handle objections better," "be more professional on calls." After 200 calls per agent, the coaching became precise. Agent A talked 75 percent of the time on calls, leaving prospects no space to express concerns. Agent B handled price objections well but consistently dropped follow-ups on day 3. Agent C had the best conversion rate but made the fewest calls.

Each coaching conversation referenced specific calls, specific moments, and specific comparisons with colleagues who handled the same situation better. That level of specificity was impossible with only agent self-reports and CRM notes.

How Claude API Powers the Analysis

Every call that enters the SalesEar pipeline is processed through Anthropic's Claude API (Sonnet 4.6). The AI analysis layer extracts structured fields from each conversation: a summary, a quality score, objections detected, pricing commitments made, follow-up promises, next actions, and prospect intent level.

Two Claude capabilities were central to making this work for a Gujarati real estate brokerage.

Structured output consistency across hundreds of calls. Each call produces the same JSON shape: summary, score, objections, commitments, intent, next actions. When the dashboard displays scoring trends across 11 agents over 4 months, the underlying data must be formatted identically on call 1 and call 900. Claude's instruction-following maintained this consistency without format drift across the entire call history. A model that returns different structures on different calls would make the dashboard unreliable.

Multilingual reasoning on code-switched transcripts. Veer Real Estate's agents pitch in Gujarati, switch to Hindi for pricing discussions, and use English for real estate terminology: carpet area, floor rise, RERA registration, possession date, GST component. A call transcript where the agent says "sir, property ni basic price 82 lakh chhe, parking alag chhe, 3 lakh no charge lagse" contains pricing information in three languages. Claude processes this as one coherent statement and extracts: base price 82 lakh, parking 3 lakh separate, total commitment 85 lakh. A model that handles only one language would miss the financial specifics embedded in the code-switched sentence.

The Results After 4 Months

Veer Real Estate's deployment moved through three phases.

Month 1: Discovery. 11 agents deployed, first calls analyzed, initial patterns identified. Follow-up gaps and pricing inconsistencies surfaced for the first time. Sudhir started using the morning dashboard review before the daily meeting.

Month 2: Coaching. Specific interventions based on per-agent data. Talk-ratio coaching for agents who talked too much. Follow-up velocity targets for agents who let leads age. Pricing consistency checks became part of the weekly routine.

Month 3-4: Operational maturity. Call data became the default input for every management decision. Pipeline reviews referenced call journey timelines, not CRM notes. Agent performance reviews cited call scores across hundreds of conversations, not impressions from a handful of overheard calls.

The measured outcomes:

Call visibility went from zero to 100 percent across 11 agents. Every conversation is transcribed, scored, and searchable. No call is invisible to management regardless of which agent made it or which device they used.

909 calls analyzed in the production system. Each call produces a structured record: transcript, score, objections, commitments, intent classification. This data set grows daily, and compounds in value as patterns become more statistically significant.

Daily call review time dropped from ad-hoc listening (when it happened at all) to a structured 15-minute morning review of flagged calls. Pricing flags, follow-up gaps, and hot leads going cold are surfaced automatically. The manager reviews exceptions, not every call.

Follow-up velocity improved measurably after the first month of data-driven coaching. Warm leads that previously sat for 4 to 5 days without contact started receiving callbacks within 24 to 48 hours once the gap became visible in the daily dashboard.

Google Dialer device coverage, which would have been a deployment blocker with most other tools, was solved on day one with the automated helper. All 11 agents are covered regardless of phone brand, with zero ongoing manual steps from any agent.

"SalesEar gave us something we never had: a clear picture of what our agents actually say to prospects. The call data changed how we coach, how we follow up, and how we run our morning meetings." Operations Team, Veer Real Estate

The Technical Build

SalesEar's backend runs on NestJS with PostgreSQL. The AI analysis layer is powered by the Claude API (Sonnet 4.6). The mobile capture layer supports Samsung, Xiaomi, and Google Dialer devices through native recording and the automated helper application.

The same architecture that handles Veer Real Estate's 11-agent team scales to 50 or 200 agents. Call ingestion, transcription, AI analysis, and dashboard rendering are designed to run concurrently across multiple organizations.

Getting Started for Real Estate Teams

SalesEar's Plus plan at ₹1,999 per user per month includes call journey tracking, persona match, SOP scoring, manager role, custom analytics, and API access. For real estate brokerages with 10 or more agents needing full management visibility, Plus provides the complete feature set.

14-day free trial with full access. No credit card required. salesear.com/signup

For brokerages evaluating call analytics, the evaluation checklist covers the 7 questions that matter. For the ROI case, sales call analytics ROI breaks down the math for a 15-agent team.