Activity mining: separating system actions from human work in your order timeline.
Order timelines mix automated system events with human labor. Activity mining separates the two, because system actions are instant and free, and human actions take time and cost money.
- Every event in your order log has an actor: a system (Stripe, the cron job, the carrier scanner) or a human (the warehouse picker, the QA inspector, the support agent). Activity mining separates them.
- System actions are deterministic and instant. Human actions introduce variance, and variance is where cost accumulates. You can’t reduce what you can’t measure.
- The most common finding: human exception handling (manual overrides, re-routes, holds) accounts for 5–12% of order volume but 35–60% of total handling cost.
Halia automatically classifies events in your Shopify and 3PL event log by actor type, surfacing which activities are human, how long they take, and where they concentrate.
Why your timeline is two timelines.
An order event log looks like a single sequence: created, paid, fulfilled, shipped, delivered. But each event has an actor. Some actors are systems: Stripe captured payment, a cron job triggered fulfilment assignment, the carrier’s scan system recorded pickup. Others are humans: the warehouse picker pulled the SKU, the QA team flagged a defect, the support agent applied a manual override to a stuck shipment.
When you look at the timeline as one continuous stream, the two types of events blend together. A 45-minute gap between “order created” and “payment captured” looks the same as a 45-minute gap between “pick assigned” and “pick confirmed.” The first is system latency. The second is a warehouse floor problem. Activity mining makes the distinction explicit.
What activity mining actually is.
Activity mining is a subset of process mining focused specifically on the actor dimension of event logs. Where process mining asks “what sequence of events occurred?”, activity mining asks “which events required human work, and how long did that work take?”
The output is a breakdown of order handling time by activity type: automated processing time, human decision time, human exception handling time, and wait time (neither automated nor human, just an order sitting in a queue). This breakdown shows you where time and cost are actually going, separated from the automated events that run at no marginal cost.
Common patterns activity mining surfaces.
Four patterns that are structurally invisible without activity mining:
- Manual override concentration. A small number of SKUs or routes generate a disproportionate share of manual interventions. These are candidates for process standardization or system configuration fixes.
- QA hold clustering. Quality holds that cluster on specific days (Monday mornings, post-weekend restarts) suggest a shift-change problem, not a product problem.
- Support-to-warehouse loops. Orders where a support agent action triggers a warehouse action (re-routing, priority re-pick) represent a process gap, the customer contacted support because the order was already off-track.
- Exception handler concentration. If 80% of exception handling is done by 2 people on your team, you have a key-person dependency, not a scalable process.
Activity mining questions, answered.
How is activity mining different from process mining?
Process mining maps the flow of events through your operation. Activity mining classifies each event by who or what performed it. The two are complementary, process mining tells you the sequence; activity mining tells you the work content of each step.
Can I do activity mining manually?
Yes, but it’s tedious. You’d export your order event log, tag each event row by actor type, then build pivots to summarise human time per order. A senior ops analyst can do this in 1-2 weeks for a sample of 1,000 orders. Automated tools do it continuously across all orders.
Do all e-commerce platforms expose actor metadata?
Most do partially. Shopify’s Order events include user_id when an action is human-triggered (vs. null for system actions). WooCommerce exposes similar metadata in its REST API. ShipStation tags shipment events with the user who created them. The metadata is rarely surfaced in standard dashboards but it’s in the event payload.
What's the biggest gain operators see from activity mining?
In our data, the largest single wins are typically scheduled batch jobs running too infrequently, fixing a cron schedule can shave 4-12 hours off median order fulfilment time at zero operational cost. The second-largest category is concentrated manual review that should be automated.
Does Halia's activity mining work without a 3PL?
Yes. The minimum viable input is order events from your storefront (Shopify, WooCommerce) and your payment processor (Stripe). 3PL integration improves the analysis for fulfilment-stage activities but isn’t required to get started.
See these patterns on your store.
Connect in 5 minutes. First findings the same day. Free under 500 orders/mo.
Start free →