A grocery chain was losing margin and could not see why. Our AI agent found the cause on day 14 of 42, two weeks before the books closed. Finance then signed off $240k in yearly savings.
Multi-format grocery, APAC
$240kSaved each year, signed off by the client's Finance team within 30 days
14 daysUntil the AI agent raised its first alert
42 daysWhole project: six weeks, fixed scope, fixed fee
+2.4ppMargin won back (in percentage points), from Q3 close to the Q4 board call
EngagementSix weeks, fixed scope, fixed fee. It began with a Week 1 audit: two calls and a written one-pager the client keeps.
BuiltAn AI agent that watches costs, and a trail that proves every number
The day-14 drop
One alert, two weeks before close.
What matters here is timing. The same problem, found after the books close, only teaches a lesson.
Contracted priceActual invoice
Agreed price vs invoiced price: the gap nobody watchedschematic, no values
Day 0Kickoff
Day 14First alert, acted on (fuel surcharge)
Day 30Finance signs off, by day 30
Day 42Project ends
Contracted priceActual invoiceschematic, no values
Agreed price vs invoiced price: the gap nobody watched
Day 0Kickoff
Day 14First alert, acted on (fuel surcharge)
Day 30Finance signs off, by day 30
Day 42Project ends
$240ksaved each year, signed off by FinanceAfter close it is a lesson. Before close it is $240k.
The problem
Margin was leaking where nobody was looking.
Margin was falling and the dashboard showed it, but not why. Costs passed through several systems before reaching a report, so the cause only became clear after the quarter had ended. The leak sat in the gap between the price agreed with suppliers and the price actually invoiced. There were too many suppliers to check by hand.
Landed cost starts here, and crosses several systems before any report
What was broken
The dashboard showed margin falling, but not why.
Costs passed through several systems, so answers arrived after the quarter closed.
Nobody compared the agreed price with the invoiced price.
With this many suppliers, checking by hand "was not humanly possible".
You might be here if
Your dashboard shows margin falling, but not why.
You only find causes after the quarter has closed.
Nobody compares agreed prices with invoices.
You have too many suppliers to check by hand.
Why it mattered. A cause found after the books close explains the quarter. It does not change it.
The approach
Watch costs as they happen, and prove every number.
Instead of waiting for reports, the AI agent read the cost data directly and watched costs as they moved. Next to it we built a trail from every number back to its source. The customer calls this the hard part. It is why Finance could check the result and sign it off "rather than take it on trust".
Every inbound load carries a contracted price, an invoice and a surcharge
The existing route
Landed costPasses through several systems
DashboardShowed the fall, not why
→
Wait for the report, reconstruct after closeThe path the agent does not take
Cost fields, read directly
Contracted price
Actual invoice
Fuel surcharge
Landed cost
→
woodfrog built
Claude-powered margin agentWatches costs as they move
Trail back to sourceUnder every number the agent used
→
The customer's side
Day-14 alertFuel surcharge
Procurement teamActed on the first alert
FinanceChecks the trail, signs off
$240k saved each year
Finance signed the number off because every field behind it traced back to source.
The solution
A working agent in 14 days, with proof built underneath.
What was built
Week 1 audit: two calls, a map of the systems, the top three decisions, where things got stuck, and a written one-pager the client kept.
A Claude-powered margin agent, live in 14 days.
It reads four numbers directly: agreed price, invoiced price, fuel surcharge and landed cost (the full cost of goods delivered).
Its first alert, acted on by the procurement team, was a fuel surcharge anomaly on day 14, two weeks before close.
In weeks 4 to 6, a trail under every number the agent used, so Finance could trace each claim to its source.
In the customer's words"The agent did the easy part. The hard part was the lineage discipline. That's what Woodfrog built first, and it's why Finance signed off on the savings in the first month."Head of Data, APAC retail customer
How it ran, six weeks counted in days
Wk1Wk2Wk3Wk4Wk5Wk6
Audit
Audit
Agent
Ship the margin agent
First flag
Day 14, agent live and first alert acted on
Lineage
Trail to source under every number
Sign-off
By day 30 at the latest
Engagement ends
Day 42
Wk 1Wk 2Wk 3Wk 4Wk 5Wk 6
Audit
Audit
Agent
Ship the margin agent
First flag
Day 14, agent live and first alert acted on
Lineage
Trail to source under every number
Sign-off
By day 30 at the latest
Engagement ends
Day 42
The impact
A flag in two weeks. A saving Finance signed off.
The first alert came a third of the way in
First alert acted on
14
Finance sign-off at the latest
30
Engagement ends
42
Days from kickoff, all three from this engagement.
Margin protected on the shelf, signed off in Finance
See where your margin leaks.A short call, or record a message. A person replies within one business day.
