- Delivery side owned generation logic, freshness handling, and product integration
Case studies
AI & ML development
AI Summaries in ERP Workflows
WORKFLOW:AI summaries for client history and Enerlytics analysis inside ERP workflows
SYSTEM STATE:Live ERP workflow with saved summaries, freshness checks, async generation, and admin AI settings
01
Get Energy turned AI summaries into reusable workflow artifacts inside ERP
The case addressed a common ERP problem: useful context was spread across long client history, operational records, and analysis flows. Reading and rebuilding that context manually took time and increased cognitive load.
The product did not need another AI text box. It needed summaries that could be generated, stored, reviewed, refreshed, and reused inside real ERP workflows. The solution became a small AI summarization workflow inside the product, with client history summaries, Enerlytics summaries, downloadable artifacts, freshness logic, saved database state, and admin controls for providers, prompts, and limits.
02
Context volume made manual reconstruction slow and inconsistent
Users worked with long histories and changing operational information. Rebuilding context from raw records every time a decision or review step was needed was slow and inconsistent.
The product constraint was clear: an AI summary was useful only if it stayed part of the system after generation.
What shaped the workflow
Client history was long enough to create manual review load
Summary quality mattered inside an ongoing product workflow
Old summaries could become misleading after new activity appeared
Heavy analysis needed a separate path from lightweight history summary
Admin teams needed control over provider, prompts, and limits
03
The first release focused on client history summary
The first slice summarized client history. A user could request a summary, view it in the interface, and use it to understand what had happened across multiple history entries faster.
That gave the team a production baseline before adding saved summaries, stale-state logic, and heavier Enerlytics analysis.
What was inside the first slice
•Backend generation of client history summary
•Frontend entry point through icon and modal
•Summary output visible inside the ERP interface
•Direct use inside a live product workflow
What was added later
Save summary data in the database
Freshness and stale-state checks
UI states and indicators
Enerlytics summary workflow with stored artifacts
Admin settings for provider, model, prompts, and limits
04
The summary became more useful once it could persist inside the product
A summary has more value when it survives the first read. Saving it in the database turned it into a reusable product artifact instead of temporary AI output.
That made it possible to compare the current summary against new history and decide when regeneration was needed.
What changed with saved state
01Summary became reusable across sessions
02The product could track whether it was still current
03Users could rely on UI status instead of guessing freshness
04The workflow gained continuity inside the ERP system
05
A stale summary can be more dangerous than no summary
Once summaries are reused, freshness becomes part of product quality. A summary that misses the latest history can distort understanding and create false confidence.
That is why stale-state handling became part of the first production logic, not a later cleanup task.
Where risk was concentrated
New history entries making the summary incomplete
Users treating an outdated summary as current context
Missing visibility into whether regeneration was required
Summary usefulness dropping while the UI still looked stable
06
The product later split lightweight summary from heavier AI analysis
Client history summary and Enerlytics summary did not behave the same way. Enerlytics needed heavier generation, separate artifact handling, and an async path for result creation and retrieval.
That split made the ERP workflow automation more practical because each use case had the right operating shape.
What the Enerlytics workflow added
•Separate entity for AI-generated analysis
•Stored artifacts such as PDF or blob output
•Queue-based generation flow
•List, open, download, and delete behavior in the UI
•Live refresh around result availability
07
The AI layer became easier to manage once configuration moved into admin controls
Production use improved when provider, model, prompt settings, and input limits became configurable. The team could manage AI behavior as an operating surface inside the product, instead of relying on hidden hardcoded logic.
That also made the workflow more durable as providers and deployment choices evolved.
What admin controls covered
AI provider selection
Model and endpoint settings
System and user prompt configuration
Max history entry limits
Separate settings for heavier analysis workflows
Direction toward self-hosted or local AI infrastructure
08
Manual context reconstruction dropped and workflow reuse improved
The biggest product gain was less friction around long history and analysis context. Users could move faster from raw records to usable context, and the product could keep that AI output inside the workflow instead of treating it as disposable text.
This made AI summaries in ERP operational, not ornamental.
What improved
Less manual reading of long client history
Faster access to usable context
Better reuse of AI output inside the product
Clearer visibility into whether a summary was still current
Stronger separation between lightweight summary and heavier async analysis
09
The workflow stayed useful because summary quality and controls remained owned
After release, the AI layer still needed supervision around freshness, admin configuration, prompt logic, and the difference between summary use cases.
That kept product behavior aligned with actual workflow needs as the AI layer expanded.
Ownership boundaries
- Client-side product ownership judged whether summaries remained useful in live workflow
- Admin settings provided an operational control surface for ongoing tuning
- Workflow quality depended on continued review as new AI paths were added
AI summaries become production-useful when they behave like product artifacts, not disposable output
This enterprise AI workflow case study shows how Get Energy turned summary generation into a live ERP workflow with saved state, freshness checks, async analysis, and admin controls. That delivery shape fits products that need faster context reconstruction while preserving trust, reuse, and operational clarity after generation.