Category strategy support agents
Research, structuring, and first-draft support for category strategies — market context, supplier landscape, spend framing — on the team’s own templates.
Impact · Case study
From manual Source-to-Contract work to measured AI systems.
This is the flagship engagement behind the numbers on this site: a portfolio of AI agents built inside a Fortune 500-scale energy environment, across category management and procurement process workflows. The employer is confidential under NDA — the method is not.
- ≈$4M
- Projected two-year value in hours returned
- S2C
- Agent coverage across Source-to-Contract
- 100%
- Of outputs human-reviewed before use
01 · The situation
Category managers and procurement specialists in a global energy operation were spending large parts of their week on high-volume manual work: gathering market and supplier information, compiling data from multiple systems, drafting category and process documents, and formatting reports. The expertise was in the decisions — but the hours were going to the preparation.
02 · The workflow, before and after
Same deliverable. Different week.
The pattern repeats across every workflow we automated: preparation collapses from days to minutes, and expert time moves to the part that actually needs an expert.
Before · Manual workflow
Days per deliverable, most of it below the expert’s pay grade.
- 1
Pull spend and supplier data from multiple systems
hours
- 2
Search for market intelligence and industry reports
hours
- 3
Chase stakeholders for inputs and context
days of lag
- 4
Compile everything into a working document
hours
- 5
Draft, format, and polish the deliverable
hours
- 6
Expert judgment: the actual decision
the small remainder
After · Agent-assisted workflow
A reviewed first draft in minutes; expert time goes to judgment.
- 1
Agent gathers, structures, and cites the source material
minutes
- 2
Agent produces a first draft on the team’s template
minutes
- 3
Expert reviews, corrects, and challenges the draft
about an hour
- 4
Expert judgment: the actual decision
the majority of the time
03 · What was built
A portfolio of narrow agents — not one big bot.
Procurement process agents
Support for recurring Source-to-Contract process work: document preparation, summarization, comparison, and status reporting.
Research and analysis assistants
On-demand briefs that read, compare, and summarize large document sets — the work that used to consume afternoons.
How an ApexAxiom agent runsInside your data boundaries
01 · Inputs
The real work arrives
- Documents & contracts
- System data & spend
- Requests & templates
02 · The agent
One narrow job, done well
- Retrieves & structures sources
- Analyzes and compares
- Drafts on your templates
03 · Human review
Your expert decides
- Reviews and corrects
- Approves or escalates
- Owns the decision
04 · Measurement
Value gets counted
- Time per task vs. baseline
- Volume × delta − review
- Adoption tracked weekly
04 · Why it worked
- Built by an operator who had run these exact workflows — the agents match how the work is actually done, not how a process chart says it is done.
- Each agent does one narrow job well, instead of one broad agent doing everything badly.
- Tested against real historical cases before anyone was asked to trust it.
- Adoption was designed, not hoped for: the tools live where the team already works.
05 · What stayed human
- Every agent output is reviewed by the expert who owns the work before it is used.
- Decisions — supplier selection, negotiation positions, strategy calls — remain entirely human.
- Sensitive data boundaries were mapped before the first build; what could not leave, did not.
- Escalation paths exist for every workflow: when the agent is unsure, a person is asked.
06 · Measurement
What was measured — and how the ≈$4M is calculated
Time per task was measured before and after, per workflow. The delta was multiplied by real task volume, discounted for expert review time, and projected across two years. That projection is the ≈$4M figure — team hours returned to higher-value work. No revenue attribution, no soft “productivity” multipliers.
Names, systems, and internal details are withheld under NDA. Everything else on this page — the method, the measurement logic, the guardrails — is exactly how we work.
The calculation, in plain terms
- 1Time per task: manual baseline vs. agent-assisted, per workflow
- 2Task volume: real counts, not estimates
- 3Review cost: expert checking time subtracted, not ignored
- 4Projection: two-year horizon, stated as a projection
Start with one workflow
Send us the process that is costing your team time.
Describe one workflow — what goes in, what comes out, who touches it, and how long it takes. We will tell you honestly whether AI is the right tool, and what a first build would look like.