Every company has a few numbers it can't get wrong and can't leave unwatched. Here is what teams do with each capability, step by step: what starts it, what runs, what gets checked, and what lands in someone's inbox.
There is nothing to set up per question and no use case to define. If the measure exists in your semantic layer, you can ask about it, and you can ask why it moved.
Finance
What changed in order margin this week?
Which accounts moved more than 250k since last close?
Supply chain
Which suppliers missed their SLA twice this quarter?
Where is on-time in-full heading this month, and why?
Quality
Which lines ran under 95% first-pass yield yesterday?
What do the last five deviations on line 3 have in common?
Revenue
Which accounts fell below 60% seat use this month?
What is driving contraction in the mid-market?
ANTICIPATE
Numbers that can't be left unwatched.
Nobody asks and nobody assigns. AgentFlow watches a metric and the whole tree beneath it, catches trouble and openings early, and tells the owner what to do about it.
ANTICIPATE · FINANCE AND COMMERCE
Order margin, watched.
Margin was reviewed on Fridays. By then the week's orders had shipped, the freight was booked, and the only thing left to do was explain it.
Margin leaks caught within a day, not a week. 1.3 points of order margin recovered over two quarters.an omnichannel retailer
Who owns it
The VP of finance owns the number. The pricing and logistics leads own the fixes.
What starts it
Nothing. It is always on: every hour, and whenever new orders or carrier invoices land.
What it watches
Order margin, and the tree under it: revenue per order, cost of goods, fulfillment cost and returns, and the fifteen inputs beneath those.
What runs
measure margin and every driverruns as code
data is freshcheck
spot a move outside the normal rangeruns as code
walk the tree to the causecalls a model
the cause explains the movecheck
size the cost by Fridayruns as code
match the playbook, draft the fixcalls a model
the owner approvesa person decides
make the change, record itchanges something
runs as codecheckcalls a modela person decideschanges somethingfans out
What is checked
It speaks up only when the drivers explain the move. It recommends only what the playbook allows. Nothing changes until the person who owns routing approves it.
What arrives
A note to the people who can act: what is slipping, why, what it will cost by Friday, and what to do about it, with a button.
Why AgentFlow
A dashboard would have shown this on Friday. AgentFlow knew the carrier had changed, knew the two-day promise, and knew who owns routing. So it could say what to do, on Tuesday.
ANTICIPATE · REVENUE
Churn, caught early.
Usage slipped in March. The renewal was lost in June. In between, the signal sat in a dashboard nobody had a reason to open.
At-risk accounts flagged five weeks earlier, on average.a B2B software company
Who owns it
The head of customer success owns the number. Account owners and revenue operations act on it.
What starts it
Nothing. It is always on: every hour, and whenever new usage data lands.
What it watches
Net revenue retention, and the tree under it: churn, contraction, expansion and service quality, and the thirteen inputs beneath those.
What runs
measure retention and every driverruns as code
data is freshcheck
spot a move outside the normal rangeruns as code
walk the tree to the causecalls a model
the cause explains the movecheck
size the renewals at riskruns as code
match the playbook, draft the fixcalls a model
tell the account ownerchanges something
runs as codecheckcalls a modela person decideschanges somethingfans out
What is checked
It speaks up only when it can name the cause. A movement it can't explain is reported as unexplained, not dressed up.
What arrives
A note to the account owner: what moved, why, which accounts, how much renewal revenue is exposed, and what the playbook says to do next.
Why AgentFlow
Nobody has to ask. It knew which accounts shared a release, which renew soonest, and which playbook applies to each.
ANTICIPATE · SUPPLY CHAIN
On-time in-full, protected.
The retailer's scorecard arrived on the fifth of the month. It said 91%, and the chargeback came with it. By then everyone knew why and nobody could do anything about it.
Chargebacks avoided in the first two quarters: $1.3M.a consumer packaged goods maker
Who owns it
The head of operations owns the number. The quality lead and the planners own the fixes.
What starts it
Nothing. It is always on: whenever supplier confirmations, production counts or shipments land.
What it watches
On-time in-full, and the tree under it: supplier delivery, schedule attainment, inventory availability and carrier performance, and the thirteen inputs beneath those.
What runs
measure on-time in-full and every driverruns as code
data is freshcheck
spot lead times stretchingruns as code
trace the SKUs and lines exposedcalls a model
check who else is qualifiedruns as code
draft the switch and the resequencecalls a model
the quality lead approvesa person decides
update the plan, record itchanges something
runs as codecheckcalls a modela person decideschanges somethingfans out
What is checked
An alternate supplier is proposed only for the SKUs it is qualified for. The switch needs the quality lead's approval. The plan is frozen before it is sent, so what runs is what was signed.
What arrives
A note to operations and quality: which SKUs are exposed, on which lines, where the month will close, what it will cost, and the switch that prevents it.
Why AgentFlow
It knew the retailer's penalty starts at 95%, so it spoke up at 93 and falling. It knew which supplier feeds which SKU, and who is qualified to replace them.
AUTOMATE
Processes that have to be right every time.
Someone assigns the work once. After that it runs on a schedule or an event, the same way every time, with its checks, its approvals and its record.
AUTOMATE · QUALITY
The out-of-spec batch.
A batch fails its spec at 2:40 a.m. The investigation is fourteen steps across eight systems, and it used to wait for the one engineer who knew all fourteen. By the time she arrived, the next two batches had run.
Event to containment: a morning down to 38 minutes. Every deviation closed with a complete record.a specialty chemicals producer
Who owns it
The plant quality lead wrote it and owns it. Quality engineers refine it. The plant manager is asked when the schedule is at stake.
What starts it
An event from the MES: a batch has failed final QC.
What runs
pull the batch recordruns as code
pull the lab resultsruns as code
results signed by the labcheck
compare to spec and the last 30 runsruns as code
check maintenance and calibrationruns as code
instrument error ruled outcheck
trace the raw material lotsruns as code
check every batch that shares them, 11 at oncefans out
draft the root causecalls a model
every claim cites a recordcheck
propose containmentcalls a model
the quality lead approvesa person decides
place the holdschanges something
the plant manager decidesa person decides
update the schedulechanges something
open the supplier corrective actionchanges something
file the reportruns as code
runs as codecheckcalls a modela person decideschanges somethingfans out
What is checked
Lab results must be present and signed. Calibration must be valid, or the run branches to a re-test. The root cause has to cite a record for every claim. The containment plan is frozen with a fingerprint before it goes for approval, and an unanswered approval escalates after fifteen minutes.
What arrives
Four batches on hold, one lot in quarantine, two production orders moved to a clean lot, a corrective action open with the supplier, and a deviation report with every record ID. All before the morning shift.
Why AgentFlow
It consults eight systems and changes four of them, so nothing here can be improvised. The path was fixed before the batch ever failed, and what ran is exactly what a named person signed.
AUTOMATE · FINANCE
The close, explained.
The first week of every month used to go to explaining the last one. Three analysts, a wall of spreadsheets, and a controller who got the answers on day five.
Flux review: five days down to one morning.a $3B industrial distributor
Who owns it
The controller. The accounting team uses it and changes it.
What starts it
A schedule: two days after each month-end close.
What runs
pull balances and journalsruns as code
ties to the trial balancecheck
find accounts over the thresholdruns as code
explain each one from the journalscalls a model
drivers cover 90% of the movecheck
write the summarycalls a model
deliver to the controllerruns as code
runs as codecheckcalls a modela person decideschanges somethingfans out
What is checked
The totals have to tie to the trial balance. The data has to be less than an hour old. No explanation is published unless the documented drivers account for 90% of the movement; anything short of that is marked unexplained and sent to a person.
What arrives
A flux package in the controller's inbox: every account over the threshold, its explanation, and the journal and document IDs behind it.
Why AgentFlow
Auditors ask how a number was produced. The ledger shows every query, every source and every approval. The 90% rule was written by the controller, in one sentence, without a ticket.
“We used to spend the first week of this month explaining last month. Now the explanations are waiting for us, with the journal entries attached.”
AUTOMATE · SUPPLY CHAIN
Supplier risk, every Monday.
Forty-two suppliers, one category team, and a review that happened quarterly because nobody had the hours to do it weekly. SLA breaches surfaced when the credits had already lapsed.
Breaches caught a week sooner. $1.1M in credits recovered in two quarters.a global auto-parts maker
Who owns it
The head of procurement. Category managers read it and act on it.
What starts it
A schedule: Monday, 9:00. It could as easily be an event, an API call, or a row landing in a table.
What runs
query delivery, quality and SLA dataruns as code
rows present, under an hour oldcheck
one agent per supplier, 42 at oncefans out
fold 42 results into one ranked tableruns as code
write the summary from the aggregatescalls a model
post to Slack under a run idruns as code
runs as codecheckcalls a modela person decideschanges somethingfans out
What is checked
Nothing moves until the data is there and fresh. The arithmetic is code, so the numbers are exact. The summary step reads the aggregates, never the raw rows.
What arrives
One message in Slack: the ranked table, the three suppliers who need a call this week, and why.
Why AgentFlow
Forty-two reviews at once, finished in 14.8 seconds for about a cent on the benchmark run. Cheap and fast enough to run every Monday without anyone thinking about it.
Customer results as reported by the customers, who are described rather than named. The runs shown are illustrative. Conditions on request.
YOUR PROCESS
Yours probably isn't on this list.
It doesn't need to be. If a process has steps, numbers and consequences, it fits. The best first candidate is one that is done by hand today, matters every week or every month, and makes someone nervous. If it qualifies, the first $50,000 is on us.
TALK TO US
Bring us the process you worry about most.
We will show you how it would run on AgentFlow, what it would check, and what each run would cost.