← Back to Stories

Stories · FSI

Total Expert cuts pull request cycle time by 31% with AI-assisted software development

Mactores brought Claude to Total Expert’s engineers and sellers in the same rollout, with Claude Code on the codebase and Claude Cowork for commercial work, all inside a SOC 2 environment.

  • AI Agents for Apps
  • FSI
  • Claude Code
  • Claude Cowork
  • MCP
  • AWS
Total Expert homepage: the headline Deliver the Perfect Customer Journey above cards for an AI Sales Assistant call, a scheduled appointment and a sent application
On the platform 200+ financial enterprises ~47K users

What Mactores delivered for Total Expert

  • The clock starts when a pull request enters review and stops when it merges.
  • Sellers put the recovered hours into the conversations that move a deal.
  • That is 23 of the 25 engineers who received the first seats.
  • One enterprise setup holds them all, mapped to Total Expert’s SOC 2 and ISO 27001 controls.
  • Engineering · Claude Code

    Cycle time per pull request

    Before 100
    Now 69 Just over two thirds of the track. 69
  • Sales · Claude Cowork

    Outreach drafting time

    Before 100
    Now 40 Two fifths of the track. 40
Each track is the figure before the rollout, set to 100. Hover a Now bar for the value.

The customer

About Total Expert

Total Expert is the Customer Operating System for financial institutions. Its platform connects banks and lenders with the consumers they serve, so borrower data runs through it.

  • 200+ financial enterprises on the platform
  • 15 of 25 top US banks and lenders among them
  • ~47K users
  • ~1/4 of US mortgage production flows through it
Total Expert at a glance. Source: Total Expert

The challenge · 01

The problem

Total Expert is in the middle of an aggressive agentic product cycle, and two of its teams felt that pace most.

  • Engineering A roadmap paced by build speed Flagship AI initiatives such as AI Sales Assistant and Customer IQ move only as fast as engineers can build, test, and ship them.
  • Sales Hours of groundwork per account Sellers put in long stretches of preparation before they reach a real conversation with a bank or lender.

What stood between a seller and the meeting

  1. Prospect research Preparation
  2. Pipeline analysis Preparation
  3. A tailored proposal Preparation
  4. A real conversation Where deals move

Four conditions set by the environment

The environment raises the bar for any change.

  • Total Expert holds SOC 2 Type II and ISO 27001 certifications, so every configuration had to map to an existing enterprise control.
  • Borrower data and proprietary code had to stay inside Total Expert’s AWS boundary.
  • None of that data or code could ever be used to train models.
  • Claude already powered the Next Best Action sales agent in production, so adding seats could not interrupt live API usage.

The aim was to put both teams on Claude at once, with a repeatable way to provision an engineering surface and a commercial surface side by side.

How Mactores delivered · 02

Two new surfaces beside an agent that was already live

A single enterprise deployment carries both surfaces. OneLogin decides who reaches which one, and the agent already in production keeps a lane of its own.

OneLogin Provisioning and deprovisioning follow the directory automatically.

Engineering track engineering security group

Sales track sales security group

Kept apart separate
security
groups
Claude Code Agentic software development: feature work, unit tests, and pull request summaries.
2 CLAUDE.md standards Repository-level rules that steer the agent
3 Existing repositories The codebase engineers already ship from Inside the AWS boundary
Claude Cowork Prospect research, pipeline insight, account plans, earnings call summaries, and proposal drafts.
2 Custom GTM skills Built around the tasks sellers repeat most
3 MCP connectors Total Expert’s platform through a purpose-built connector, the content library through Atlassian’s official MCP Per-seller, role-scoped
Already in production
Claude API Next Best Action sales agent Live API traffic

OneLogin

Sign-in and SCIM provisioning from the company directory, Google Workspace

Provisioning and deprovisioning follow the directory automatically.

Engineering track engineering security group

  1. 1 Claude Code Runs directly against existing repositories Agentic software development: feature work, unit tests, and pull request summaries.
  2. 2 CLAUDE.md standards Repository-level rules that steer the agent
  3. 3 Existing repositories The codebase engineers already ship from Inside the AWS boundary

Kept apart: engineering and sales sit in separate security groups

Sales track sales security group

  1. 1 Claude Cowork Research and drafting for sellers Prospect research, pipeline insight, account plans, earnings call summaries, and proposal drafts.
  2. 2 Custom GTM skills Built around the tasks sellers repeat most
  3. 3 MCP connectors Total Expert’s platform through a purpose-built connector, the content library through Atlassian’s official MCP Per-seller, role-scoped

Already in production

Claude API → Next Best Action sales agent, carrying live API traffic

How Total Expert’s Claude deployment supports AI-assisted software development and AI sales prospecting next to an agent already in production. Hover OneLogin, Claude Code, or Claude Cowork for more.

The work

How Mactores runs it

This is how Mactores’ AI Agents for Apps work continues once the seats are live.

First-line support

Covered both activations, engineering first and then sales

Admin runbook

Handed over complete to Total Expert’s platform engineering leads

Quarterly review

The CTO office checks usage, security, and return on investment

Delivery · 03

A rollout that moved through agreed gates

Mactores already ran in production at Total Expert as the delivery partner for the Next Best Action agent, and that history shortened the path through security.

  1. Three-week vendor review

    Every data flow walked through with Total Expert’s security team against its SOC 2 audit scope

  2. 25 Claude Code seats

    The Platform and AI engineering teams go first

  3. Gate passed, 45 seats

    Utilization and quality both clear the bar agreed with the CTO organization

  4. Claude Cowork for 30 sellers

    Sales joins as the second surface, with custom GTM skills for its most common tasks

Every setting, tied to a control

The vendor review ended with a written mapping, so each setting can be traced to the control it answers to.

Control mapping Written, control by control

Area Configuration Control or scope
Access control Workspace policy and organization settings SOC 2 CC6 · ISO 27001 A.9
Data retention Zero data retention, enforced Whole organization
Model versions Pinned Each workspace

In production · 04

The outcome

Claude Code has become a daily tool for Total Expert’s engineers. In April 2026, the team shipped a Customer IQ feature ahead of schedule while the Next Best Action agent stayed up in production.

Engineering now has a clear path to 60 to 80 seats across the full organization.

A company-wide roadmap of about 250 seats takes Claude Cowork past sales into marketing, customer success, implementation, and operations.

Seats by stage, drawn to scale
  1. Jan 2026 25 seats, Claude Code
  2. Feb 2026 45 seats, Claude Code
  3. Apr 2026 75 seats: 45 Claude Code, 30 Claude Cowork
  4. Roadmap ~250 seats company-wide

What we ship · AI Agents for Apps

Want to know more about Mactores’ expertise in AI agents for regulated work?

Audit scope, borrower data, and an agent already in production all shaped this rollout. Mactores builds AI agent programs around the controls a business already answers to, and runs them on AWS.

Bringing Claude to more than one team? Mactores’ generative AI consulting practice plans that kind of rollout on AWS.