Algebra IT captured 64K work records in five weeks with BackContext

The Challenge

Algebra IT wanted a model trained on its own staffing workflows, but had no clean record of how the work actually got done. Intake, screening and submission happened across several agent tools, and none of it was linked to outcomes.

‍

The Approach

Applied Crew started with Algebra IT in mid-September 2026. BackContext rolled out across the team’s agent tools in a day, recording every task with who did it, where, and what came of it.

‍

64K records in five weeks

BackContext captured 64K work records in the first five weeks. Each record is attributed and linked to its outcome, and only verified examples are kept as training and eval data.

‍

Three workflows mapped

Working inside the team, our engineers mapped three workflows for fine-tuning: intake, screening and submission. Each one is being built up with the verified examples it needs to train and evaluate a model.

‍

First model in training

The first model is in training. Before it ships, it will be benchmarked on Algebra IT’s held-out tasks against Claude Opus 5.5 and GPT-6.1 Sol, and it goes live only if it clears the quality gate.

‍

The Results

In five weeks, Algebra IT went from scattered agent sessions to 64K attributed, outcome-linked work records and a clear path to its first model. Our typical time to a first production model is 4–6 weeks.

‍

Key Outcomes

• 64K work records captured in five weeks
• 1 day to roll out BackContext across the team’s agent tools
• 3 workflows mapped for fine-tuning: intake, screening, submission
• First model in training, with a benchmark against Opus 5.5 and GPT-6.1 Sol scheduled
• Every outcome recorded so each model version learns from the last

‍

Customer Feedback

“BackContext was running across every agent tool in a day. Five weeks later we had the training data most teams spend a year collecting.” — Head of Delivery · IT services