← Back to Stories

Stories · Internet & Software

Hibu gets answers from 1.82M calls a month with conversational analytics on Claude

Mactores built Hibu a conversational analytics platform on AWS. Every inbound lead call runs through Claude, and business users question the full call base in plain language. Answers that waited 2 to 3 days in an analyst queue now arrive in under 5 minutes.

  • AI Agents for Apps
  • Internet & Software
  • Claude API
  • Claude Cowork
  • AWS

What Mactores delivered for Hibu

  • Every inbound lead call Hibu’s clients receive, at a volume no manual review process could reach.
  • Questions used to join a 2 to 3 day analyst queue because call content had no BI or SQL path.
  • 17 of 20 first-wave users, asking about 23 questions each per week.
  • April security review, May first users, June and July champions, August company-wide approval.

Baseline

2 to 3 days

in an analyst queue for any question about call content, with no BI or SQL path into transcripts

Outcome

Under 5 minutes

to an answer drawn from the actual transcripts of 1.82M calls a month

Time to value

4 months

from the first security review in April 2026 to approved company-wide expansion in August

Time to an answer, drawn to scale
Before 2 to 3 daysTwo days is 2,880 minutes and three is 4,320. The queue ran anywhere in that range, so the solid bar stops at two days and the hatched part runs to three. 2 to 3 days in an analyst queue
After Under 5 minutesFive minutes is 0.12% of a 3-day track. At true scale it is a hairline, marked here with a dot. Under 5 minutes in Claude Cowork

Each track spans 3 days (4,320 minutes). Hover a bar for the minutes.

The customer

About Hibu

Hibu provides digital marketing to small businesses across the United States, connecting local companies with customers through search marketing, social advertising, display, and website services. Roughly 4,600 people work at Hibu, and the demand it generates for its clients arrives largely as inbound phone calls captured through Hibu One.

  • Search marketing
  • Social advertising
  • Display
  • Websites
Hibu homepage describing the Hibu One digital marketing platform for small businesses Hibu OneOne platform for a small business’s digital marketing. The demand it generates for Hibu’s clients arrives largely as inbound phone calls. WebsitesWebsites Hibu builds for local businesses, like this veterinary clinic. They succeed when the phone rings. Social advertisingSocial ads for local businesses. Like search and websites, they are judged by the calls they bring in. Search marketingSponsored search ads for a plumber, a dental practice, a heating company. When they work, the phone rings.
Hibu One runs a small business’s search, social, display, and website marketing on one platform. Hover the numbers. Source: hibu.com
  1. 1Hibu One. One platform for a small business’s digital marketing. The demand it generates for Hibu’s clients arrives largely as inbound phone calls.
  2. 2Websites. Websites Hibu builds for local businesses, like this veterinary clinic. They succeed when the phone rings.
  3. 3Social advertising. Social ads for local businesses. Like search and websites, they are judged by the calls they bring in.
  4. 4Search marketing. Sponsored search ads for a plumber, a dental practice, a heating company. When they work, the phone rings.

Results

With Claude + Mactores, Hibu does

  • Answers questions about call content in under 5 minutes, down from 2 to 3 days in an analyst queue
  • Analyzes 1.82M inbound lead calls a month through Claude
  • Reached 85% weekly active use in its first wave of Claude Cowork users, at about 23 questions per person each week
  • Approved company-wide Claude Cowork expansion across its roughly 4,600-person workforce in August 2026
  • Keeps the full transcript corpus inside its own AWS environment, with read-only, role-scoped access

The challenge · 01

The problem

Hibu’s product generates phone calls. Search marketing, social advertising, and websites built for a plumber or a dental practice succeed when the phone rings, and at Hibu’s size that adds up to roughly 1.82M consumer lead calls a month across its customer base. Every call produces a transcript with lead classification, rating, disposition, and satisfaction signals.

The questions Hibu’s business teams most wanted answered sat inside those transcripts.

  • Lead quality How good are the leads, and how do they convert? Waited 2 to 3 days
  • Market signals Which competitors and prices come up? Waited 2 to 3 days
  • Trends Where do disposition and satisfaction trend by product, vertical, and geography? Waited 2 to 3 days

None of it had a BI or SQL path, so each question joined a 2 to 3 day analyst queue, and many were never asked at all.

Hibu homepage counter showing 27,154,421 calls from search campaigns Calls from search campaignsHibu’s public count of calls its clients’ search campaigns have produced. Inbound lead calls like these are what the platform Mactores built analyzes: 1.82M a month.
Calls are the product. Hibu’s homepage counts more than 27 million calls from search campaigns alone. Source: hibu.com
  1. 1Calls from search campaigns. Hibu’s public count of calls its clients’ search campaigns have produced. Inbound lead calls like these are what the platform Mactores built analyzes: 1.82M a month.

The constraints were strict. Call content is sensitive consumer data, and the transcript corpus could not leave Hibu’s AWS account. Answers had to be grounded in what callers actually said, not in summaries of summaries.

How Mactores delivered · 02

Claude-powered conversational analytics lets business teams question the whole call base

Mactores built Hibu a transcript analytics platform on AWS that runs every call through the Claude API for enrichment and analysis, then put Claude Cowork on top of it as the way people reach the results. A purpose-built MCP connector gives Claude hybrid keyword and semantic search across Hibu’s OpenSearch index, plus full-call retrieval, all inside Hibu’s own VPC. A business user signs in through Microsoft Entra ID, asks a question in plain language, and gets an answer drawn from the actual transcripts. No SQL, no ticket to the business intelligence team, no waiting for someone else’s reporting cycle.

Every call enrichment path
Inbound lead callsInbound consumer lead calls for Hibu’s clients. Every call produces a transcript with lead classification, rating, disposition, and satisfaction signals.
Claude APIEvery call runs through the Claude API for enrichment and analysis, a volume no manual review process could reach. Anthropic commercial terms keep the data out of model training.
OpenSearch indexThe enriched corpus lives in an OpenSearch index inside Hibu’s VPC. It never leaves Hibu’s AWS account.
Every question answer path
Business userAccess follows Entra ID. App roles map to read-only OpenSearch roles scoped by call type, and when someone leaves Hibu, their Claude access goes with them the same day.
Claude CoworkBusiness teams ask in plain language and get answers drawn from the actual transcripts in under 5 minutes. No SQL, no ticket to the business intelligence team.
MCP connectorGives Claude hybrid keyword and semantic search across the index, plus full-call retrieval. Claude receives only what each query retrieves.

Every call enrichment path

  1. 1 Inbound lead calls 1.82M a month, captured for Hibu’s clients through Hibu One Inbound consumer lead calls for Hibu’s clients. Every call produces a transcript with lead classification, rating, disposition, and satisfaction signals.
  2. 2 Claude API Enrichment and analysis of every call transcript Every call runs through the Claude API for enrichment and analysis, a volume no manual review process could reach. Anthropic commercial terms keep the data out of model training.
  3. 3 Inside Hibu’s VPC OpenSearch index The full call corpus, enriched and searchable The enriched corpus lives in an OpenSearch index inside Hibu’s VPC. It never leaves Hibu’s AWS account.

Every question answer path

  1. 1 Business user Signs in with Microsoft Entra ID and asks in plain language Access follows Entra ID. App roles map to read-only OpenSearch roles scoped by call type, and when someone leaves Hibu, their Claude access goes with them the same day.
  2. 2 Claude Cowork Answers from what the connector retrieves Business teams ask in plain language and get answers drawn from the actual transcripts in under 5 minutes. No SQL, no ticket to the business intelligence team.
  3. 3 Inside Hibu’s VPC MCP connector Purpose-built by Mactores: hybrid search plus full-call retrieval Gives Claude hybrid keyword and semantic search across the index, plus full-call retrieval. Claude receives only what each query retrieves. Read-only Scoped by call type No write path No raw queries

The MCP connector searches the OpenSearch index. Both sit inside Hibu’s own AWS VPC, and the transcript corpus never leaves it.

How a question gets answered. The enrichment path runs on every call; the answer path runs on every question. Hover a step for detail.

The work

What Mactores built and ran

This is Mactores' AI Agents for Apps work in practice: Claude answering against Hibu's production call data, inside Hibu's own AWS account, with read-only access.

Transcript analytics platform on AWS

AI call analysis at full volume: every one of 1.82M monthly calls enriched and analyzed through the Claude API

Purpose-built MCP connector

Hybrid keyword and semantic search across the OpenSearch index, plus full-call retrieval, inside Hibu's VPC

Access model

Microsoft Entra ID sign-in; app roles mapped to read-only OpenSearch roles scoped by call type

Claude Cowork rollout

Role-based enablement and prompt patterns for the 12 questions users asked most

Adoption program

Five champions, a shared playbook, per-team prompt templates, and a 45-minute onboarding gate

Data curation

Schema descriptions and data curation improved from user feedback and domain knowledge

Run and handover

First-line support through August, then a runbook and admin training for the Hibu IT service desk

Delivery · 03

From security review to company-wide in four months

  1. Security review

    Three weeks with Hibu’s security team, walking call content end to end through Entra ID and AWS

  2. First 20 users live

    Role-based enablement and prompt patterns for the 12 most-asked questions

  3. Champions and playbook

    Five champions build a shared playbook; each team gets its own prompt templates

  4. Company-wide approved

    Cowork expansion approved; runbook and admin training go to Hibu IT

The work started with a three-week data-handling review with Hibu’s security team in April 2026, walking consumer call content end to end through the Entra ID and AWS setup. Each objection had a direct answer.

  • Data residency

    The corpus never leaves Hibu’s VPC, and Claude receives only what each query retrieves.

  • Model training

    Anthropic commercial terms keep the data out of model training.

  • Access

    Entra ID app roles map to read-only OpenSearch roles scoped by call type.

  • Agent actions

    The agent has no write path and cannot send raw queries.

In May, the first 20 business users went live with role-based enablement and prompt patterns for the 12 questions they asked most. Over June and July, five champions across marketing operations, sales enablement, and product turned their own query patterns into a shared playbook. Early questions were often too broad, and “what do customers think?” produced vague answers, so each team got its own prompt templates. Mactores also improved data curation and schema descriptions based on user feedback and domain knowledge.
Demand outran enablement by the third week, which is the clearest signal the platform was working. New seats were gated behind a 45-minute onboarding session, and in August 2026, Hibu approved expansion of Claude Cowork across the company.

In production · 04

The outcome

Hibu's business teams now answer questions about call content in under 5 minutes instead of waiting 2 to 3 days. In the first wave, 17 of 20 users were active every week by week six, asking about 23 questions each per week. With conversational analytics in place, conversation data that nobody had time to read is now a record the business can question directly.

Mactores ran first-line support through August and handed a runbook and admin training to the Hibu IT service desk, with Mactores as the escalation path. Access follows Entra ID, so when someone leaves Hibu, their Claude access goes with them the same day.

With company-wide expansion approved, Claude Cowork is set to reach teams across Hibu's roughly 4,600-person workforce. Row-level scoping by geography, product, and business is the next step, so each new group of people can bring the questions the platform hasn't been asked yet.

17of 20

first-wave users active every week by week six

~23

questions asked per active user, every week

Company-wide

Claude Cowork expansion approved in August 2026 across a roughly 4,600-person workforce

What we ship · AI Agents for Apps

Want to know more about Mactores' expertise in production AI agents?

Hibu's conversational analytics platform shows how Mactores ships AI agents: Claude answering against real business data inside the customer's own AWS account, with read-only, role-scoped access.

Know more about Mactores' AI agent expertise

Scoping an LLM program on your own call, chat, or ticket data? Read about Mactores' generative AI consulting engagements on AWS.