Introduction

Writing code is rarely the slowest part of enterprise delivery. For an 18-person SAP team, the real bottleneck sat in the ticket itself: vague requirements, missing technical context, and endless back-and-forth with consultants. Working as a fractional AI lead, Marian Kamenistak helped the team introduce a simple gate—no specification enters development unless it scores at least 8 out of 10. The result was a 71 percent reduction in median implementation cycle time, falling from 19 days to roughly 5.5 days over five weeks.

What Happened

The team began by admitting that AI could not rescue a bad ticket. An ABAP developer handed an unclear request would wait for answers, guess, and rework—only faster with AI. To fix this, they built an AI evaluator that reads a Jira ticket and scores it against a 10-point rubric created for SAP objects.

The rubric requires:

  • The business goal and who benefits
  • Affected SAP objects, such as tables, function modules, BAdIs, and interfaces
  • Current versus target behavior
  • Testable acceptance criteria using given/when/then
  • Test data and edge cases
  • Authorizations and personal data handling
  • Dependencies and side effects, including batch jobs and transports
  • What is explicitly out of scope

Acceptance criteria and affected objects carry two points each, reflecting where the team had struggled most. At the start, the average spec scored just 3.3. Any specification below 8 is returned to its author with a list of missing items; it does not enter the sprint.

Once the spec gate was in place, the team layered on AI coding skills. One skill scores incoming tickets; another searches the actual Z-codebase for relevant contracts and bugs rather than relying on pasted snippets. The team also built an incident categorization agent in about three hours, replacing a process that previously depended on one person’s knowledge. SLA-related ticket resolution time also fell 49 percent, from 2.72 days to about 1.4 days.

Knowledge sharing happened through deliberate pairing: every developer new to AI was paired with an AI-savvy colleague so skills spread across the team instead of staying with a few individuals.

Why This Matters

The lesson is not that AI writes better ABAP. It is that delivery speed improves when requirements are clear before coding begins. In SAP environments, a missed authorization or a forgotten transport dependency can cost far more than a few lines of code. Because the scoring happens in Jira and does not require AI to connect directly to SAP, any team can start without touching production systems.

The approach is not without critics. Some engineers argue that heavy specification is overkill for small fixes. For a two-line change, that may be true. For SAP implementation work, where technical context and business rules are complex, the team found the trade-off worthwhile.

Key Takeaways

  • Score every specification before development begins, and block anything below 8 out of 10.
  • Measure your baseline median cycle time in Jira so you can prove improvement.
  • Weight acceptance criteria and affected technical objects heavily in your rubric.
  • Pair AI beginners with experienced users to spread skills quickly.
  • You do not need AI integrated into SAP to start; you need clean Jira data and a shared definition of a good spec.

Conclusion

Spec-driven development treats the specification as the source of truth for both people and AI. For this SAP team, the biggest gain came not from generating code faster, but from stopping work on poorly defined tickets. Teams that want to replicate the results can start on Monday by scoring their last 20 specs, fixing their baseline metric, and refusing to let anything below an 8 enter development.