• AI Solution
  • Product co-creation

Raising AI matching accuracy from 60% to 90% — a custom SFA for the SES industry

SES Provider Client C

Challenge and Context

Client C runs an SES (system engineering services) business and was planning to launch an SFA service for the SES industry, built around matching engineers’ skill sheets with projects. Two challenges stood in the way:

  • Matching projects with engineers’ skills required deep domain knowledge, leaving less-experienced sales staff dependent on veterans
  • An in-house attempt at AI-based matching plateaued at around 60% accuracy — short of practical use

They engaged Arsdot to raise the accuracy of the AI matching at the core of the service to a practical level.

Outcomes

Proprietary logic raised AI matching accuracy from roughly 60% to 90%.

We built the SFA service from scratch with Next.js, Python, LLMs, and AWS, covering project matching from skill sheets, automatic content enrichment, and skill visualisation. Less-experienced sales staff can now make solid proposals, and engineers can update their own skills easily, keeping skill sheets current with far less upkeep.

How Arsdot approached it

  • Broke down why accuracy was falling short and designed a pre-extraction data pipeline
  • Refined LLM prompt design to lift matching accuracy from 60% to 90%
  • Co-created the full service — matching, skill visualisation, and automatic enrichment

We started by technically dissecting why the existing AI matching underperformed. We then developed proprietary logic combining a pre-extraction pipeline for skill sheets with carefully engineered LLM prompts, raising matching accuracy from roughly 60% to 90%.

We also made it easy for engineers to update their own skills, so skill data stays fresh and matching accuracy holds over time.

Client voice

Head of Sales

Takahashi

  • Q

    What challenges were you facing before the project?

    A

    Matching projects with engineers depended on our veteran salespeople’s experience, and our own attempt at AI matching stalled at around 60% accuracy — not enough for real use.

  • Q

    What stood out after working with Arsdot?

    A

    They broke down exactly why the accuracy fell short and verified each improvement step by step. Matching accuracy reached around 90%, and it now anchors the service with confidence.

  • Q

    How is the service helping you today?

    A

    Even less-experienced sales staff can propose projects with confidence. Engineers find it much easier to update their skills, so skill sheets stay up to date.