Home Use Cases Work Insights About Contact
Back to Our Work

How a Growth-Stage Data-Driven Company Built Forecasting Every Operating Number with Human Approval Before Money Moves

Operations1
Open the walkthrough

Five-Step AI Workflow with Human Oversight

Each step shows the workspace, what the AI completed, and what still needs human review.

Steps 4 and 5 need a person. Step 4 is waiting for approval right now. Click it to open the full walkthrough.

Estimates are directional and based on stated assumptions. All names, organizations, and identifying details have been anonymized in accordance with our confidentiality agreements.

The Transformation

How a Growth-Stage Data-Driven Company Built Forecasting Every Operating Number with Human Approval Before Money Moves

Before
A person used to re-key the numbers between systems every Monday morning.
A person used to build next week's forecast in a spreadsheet, one afternoon a week.
A person used to find a problem in Friday's report, days after it started.
After
every feed in, cleaned, with the broken one held
every line forecast with a range, or withheld
today against the forecast, hour by hour
the week's calls, scenario-tested, waiting on your pick
last week scored, the model changes waiting on you
Executive Summary

Every operating number the company plans on, from the overnight data pull to the decision it pays for. The agents collect and clean the numbers, forecast every line with a range, watch the day against the forecast hour by hour, and hand a person only the calls that move money. Five stations, run as a loop: what Friday's scoring learns about a model or a feed changes what Monday's forecast says and what the intake holds at the door.

What Was Broken

A person used to re-key the numbers between systems every Monday morning
A person used to build next week's forecast in a spreadsheet, one afternoon a week
The real cost
A person used to find a problem in Friday's report, days after it started.

What We Built

Five stations and 11 subagents. Each subagent carries its own tasks and its own refusal.

A4
Feed Sweep
Collect every feed off the order, warehouse, roster, supplier and returns systems
A3
Data Check
Check each field against its own history, so a decimal slip cannot pass as a real number
A4
Signal Builder
Extract the calendar, lag and season signals from every cleaned series
A4
Model Runner
Run the model set against every series, several models per line, never one for all
A4
Live Compare
Compare each hour's actuals against the band the forecast drew for it
A3
Alert Desk
Route each real anomaly to the person who owns that line, with its numbers attached
A2
Scenario Desk
Run each pending call through three futures: demand up, flat, down
A1
Allocation Writer
Draft the week's stock and hours recommendation off the forecast and the scenarios
A3
Decision Record
Record each decision with the forecast and scenarios that were on screen when it was made
A4
Score Keeper
Score every published forecast against what actually happened, line by line
A2
Tuning Proposal
Propose the model or threshold change that would have cut the week's worst misses

How It Runs

1

Data Intake

Nobody is asked anything here

Agents collect every feed from every system, match the same order across systems that name it differently, merge duplicates, and hold any feed whose totals disagree with the system it came from. Nothing waits on a person here. A held feed never enters a forecast and is never quietly patched over.

2

Forecast Building

Nobody is asked anything here

Agents shape the cleaned data into signals, run the model set against every line, pick the best model per line on weeks it never saw, and publish each forecast with the range it honestly carries. Nothing waits on a person here. A line no model can beat guessing on gets no number, and the withholding is flagged, not hidden.

3

Anomaly Watch

Nobody is asked anything here

Agents compare each hour's actuals against the band the forecast drew, wait out a single odd hour, and turn a swing that holds into one alert carrying its own numbers. Nothing waits on a person here. A real anomaly reaches the person who owns that line as an alert, days before it would have surfaced in a weekly report.

4

Allocation Decisions

A person answers here

Agents run each pending stock and staffing call through three futures, draft a recommendation where the futures agree, and put the forecast line under every number. A person picks. This is where a person works. Everything before it exists so that three calls, not forty lines of data, reach this screen.

5

Forecast Scoring

A person answers here

Agents score every published forecast against what actually happened, trace each big miss to its cause, and propose model changes with their downsides written beside their gains. A person accepts or rejects each one. A person decides here. How the company forecasts is not something an agent changes quietly.

Where a Person Decides

Step 4, Allocation Decisions. Pick the Contoso buy, the three futures and their costs are on screen. Decide the Wide World Importers reorder and the new line, or hold them a week. money moves only on your say.
Step 5, Forecast Scoring. Accept or reject the two proposed changes. Say whether the drifting Adventure Works line gets re-fit now or watched another week. the models are yours to change.

Operating Model

This changes how work flows through the team.

Role
Responsibility
Allocation Decisions owner
Pick the Contoso buy, the three futures and their costs are on screen. Decide the Wide World Importers reorder and the new line, or hold them a week. money moves only on your say.
Forecast Scoring owner
Accept or reject the two proposed changes. Say whether the drifting Adventure Works line gets re-fit now or watched another week. the models are yours to change.
Data Intake, when it goes wrong
Look here only if a number on the board looks wrong at the source.
Forecast Building, when it goes wrong
Look here only if a number on the board surprises you.

What Transfers, What Must Be True

What transfers
A person is in the loop wherever money moves or the models change how the company plans. Two stations refuse to proceed on their own, Allocation Decisions and Forecast Scoring, and those are the two a person carries. Everywhere else the agents clean, forecast, watch and alert, and reach you only when the call is yours to make.
Allocation Decisions stops for a person, and money moves only on your say.
Forecast Scoring stops for a person, and the models are yours to change.
Every subagent says what it will not do. 11 of them do.
What must be true in your environment
The agent can read the systems your records already live in. This one reads 17.
Somebody owns Allocation Decisions and has time for it.
Somebody owns Forecast Scoring and has time for it.

Failure Modes

What breaks this pattern:

✗ Bad totals enter the forecast

A day's feed arrives with a total that disagrees with the system it came from, and it goes in anyway. Every forecast built that day inherits the error, and nobody can say later which numbers it touched.

✗ Guesses published as forecasts

A model that cannot beat the naive guess publishes its numbers anyway. The board plans stock and hours against a figure no better than guessing, and treats it as a prediction.

✗ Alerts people learn to ignore

Every odd hour sends an alert, with no pattern behind it. People learn that alerts mean nothing, so the real event gets the same silence.

✗ Money moves without a person

An allocation executes before any person picks it. Stock and hours move on a model's output, and the first person to see the call is the one who has to unwind it.

Directional Outcomes

What the agent counts, and the station that counts it.

These counts are the tallies from one monitored run of the agents. They are not monthly or annual totals.

Feeds cleaned and through
Counted at Data Intake
4,830
Broken feeds that reached a forecast
Counted at Data Intake
0
Feeds held for a person anyway
Counted at Data Intake
22
Forecasts published
Counted at Forecast Building
14,350
Numbers published without a model behind them
Counted at Forecast Building
0
Forecasts withheld as unforecastable
Counted at Forecast Building
61
Hours of lines watched
Counted at Anomaly Watch
118,000
Real swings that surfaced first in a weekly report
Counted at Anomaly Watch
0
Anomalies sent to a person
Counted at Anomaly Watch
84
Our measurement policy: We do not publish precise ROI without baseline methodology. Every figure above carries its basis.

What Runs Where

Every step names the subagent that does the work, the record it writes, the thing that raises a question for a person, and what it is allowed to touch. This is drawn from the source, not from a diagram somebody kept in sync by hand.

1Data Intake
subagentfeed-intake
writesdata/clean/<feed>/<date>.json
raiseshold-broken-feed
may touchdata/**, the source systems read-only
2Forecast Building
subagentforecast-run
writesforecasts/<week>/<series>.json
raisesno-model-beats-guessing
may touchforecasts/**, data/clean/** read-only
3Anomaly Watch
subagentlive-watch
writesanomalies/<date>/<event-id>.json
raisesraise-anomaly-alert
may touchanomalies/**, forecasts/** read-only, the live feeds read-only
4Allocation Decisions
GATE
subagentallocation-desk
writesdecisions/<week>/<call-id>.json
raisesawait-your-allocation
may touchdecisions/**, scenarios/**, everything else read-only
5Forecast Scoring
GATE
subagentscore-and-tune
writesscores/<week>.json, changes/proposed/<change-id>.json
raisesaccept-model-change
may touchscores/**, changes/proposed/**, the history read-only

Stack

Every system this agent reads or writes.

System
Read at
Stations
the alert channel people already read
Anomaly Watch
1 of 5
the cleaned overnight data
Forecast Building
1 of 5
the day's forecast bands
Anomaly Watch
1 of 5
the decision record
Forecast Scoring
1 of 5
the forecast board
Allocation Decisions
1 of 5
the forecast record
Forecast Scoring
1 of 5
the live order and warehouse feeds
Anomaly Watch
1 of 5
the model library
Forecast Building
2 of 5
the order system
Data Intake
1 of 5
the price and promotion calendar
Forecast Building
1 of 5
the purchasing system
Allocation Decisions
1 of 5
the returns system
Data Intake
1 of 5
the scenario screen
Allocation Decisions
1 of 5
the shift roster
Data Intake
2 of 5
the supplier delivery records
Data Intake
1 of 5
the warehouse system
Data Intake
1 of 5
the week's actuals
Forecast Scoring
1 of 5
Next Step

Want to see if this pattern fits your predictive analytics?

No build commitment·Real samples, not a demo·Estimate in writing