Case files
The work, and what it moved.
Built by our forward deployed engineers inside your environment, each one runs on your definitions, your live signals, and a record of what the last decision returned. Nine production deployments across consumer goods, legal, engineering and enterprise software. Six documented in full below: the question each client brought, what we built, and the number that moved.
01 · Index of files
Six files, one build underneath
All six run on the same thing: your definitions, your live signals, your decision record. Open any file.
- /01Media management
“Which half of our media is wasted, and can we prove it in flight?”
- /02Business review
“The pre-read takes 15 people two weeks. Every month.”
- /03Social
“By the time we spot a trend, it's dead.”
- /04Sales & pricing
“Where is share leaking, and which lever gets it back?”
- /05Business review
“The monthly review explains what happened. Never why.”
- /06Media management
“Can the system move money, not just describe it?”
02 · The case files
File 01Media performanceFortune 100 CPG12 weeks · 2 engineers
Which half of our media is wasted, and can we prove it before the next flight, not after?
What we built
Reconciles the MMM read with in-flight platform delivery and agency actuals. The brand and its agency each did that reconciliation separately, on different definitions, six weeks apart. Waste surfaces weekly as dollars-at-risk by channel, each line carrying the recommended move and its predicted incremental sales.
Systems connected
- MMM outputs
- Meta
- TikTok
- Amazon Ads
- Agency reporting
- SOV
Team: a data-platform lead and an AI engineer, embedded with the media capabilities team · first waste report in week six
Paid social frequency capping is leaking $310K this flight: same audience, fourth exposure, zero incremental lift.
- Dollars at risk
- $310K this flight · $1.2M annualized at current pacing
- Driver
- Effective frequency exceeded in 2 of 5 audience cells; reach curve flat past 3.2×
- Recommended
- Cap at 3× and shift the excess to the under-delivered retargeting cell
- Predicted
- +$480K incremental sales over the remaining 5 weeks
By the time you're redeploying any of your spend, you're looking at data that's five or six months old. Which is crazy.
What moved
$4.3M /yr
Media waste identified and queued for reallocation
190 → 105 days
Insight-to-action cycle, measured across two quarters
File 02S&OP · supply chainFortune 100 CPG90 days · 2 engineers
The monthly S&OP pre-read takes fifteen people two weeks. What would it take to make it take nobody, no time?
What we built
A pre-read generator over retailer POS, shipments and the consensus forecast, with every KPI resolved to one validated definition first, because the two-week grind was mostly reconciliation, not analysis. Output lands as an editable deck in the company template, sources dated on every page.
Systems connected
- Retailer POS
- Shipments
- Consensus forecast
- Circana
- Company deck template
Team: two engineers, first insights inside 48 hours · shadow run matched the human pre-read for two consecutive cycles before anyone retired a meeting
4 / 22
Demand vs. consensus, exceptions only. Two SKU families are driving the entire national miss.
- Fill rate
- 94.1% (−2.3 pts vs plan)
- Driver
- DC-East allocation lag, weeks 2–3
- Ask
- Approve reallocation before wk 5 reset
Sources: retailer POS (Mon) · shipments (Mon) · consensus fcst v3 · definitions: DOS v2.1
80% of the time is spent coupling things together so that we can make a decision. Only 22% was left to actually think.
What moved
1-2 hrs → 90 s
Pre-read deck generation, in the company's own template
100% match
Replicating the analyst-built baseline, verified before switchover
File 03Social & trend intelligenceFortune 100 CPG8 weeks · 2 engineers
There's more conversation than any team can read, and the moments worth acting on die in days.
What we built
A round-the-clock scan across text, image and video, tuned to the brand's equity, voice and taxonomy. Trends arrive scored on relevance, momentum and brand fit, each with a ready-to-edit brief. Every accept, edit and pass trains the next week's ranking.
Systems connected
- Social listening
- TikTok Creative Center
- Brand guidelines
- Creator roster
Team: two engineers, embedded with the brand's social team · the first trend the team acted on surfaced eleven days before it peaked
"Damp styling" is breaking out of the haircare niche: 4.1× momentum this week, strong fit with the hydration equity claim.
- Momentum
- 4.1× week-over-week · 78% novel audience
- Brand fit
- High: maps to hydration claim; two roster creators already in the conversation
- Window
- Est. 9-14 days before saturation
- Attached
- Ready-to-edit brief · 3 format suggestions · draft creator list
I just sent it to my team to fact-check, because if it's right, I'm very impressed. It gave really detailed and pointed recommendations.
What moved
88K → 9 min
Posts read and ranked in one run, previously 2-3 weeks of manual reading
Eleven days
Lead time on the first trend the team acted on, ahead of its peak
File 04Sales & pricing strategyFortune 500 beverage90 days · 3 engineers
Category share shifts monthly. Reconciling sell-out against research takes weeks. So pricing calls run on instinct.
What we built
Category scorecards over POS sell-out and syndicated data, decomposed to the levers a key-account manager can actually pull: price gap, promo depth, distribution, velocity. Each share move arrives with its likely driver and a recommended action sized in revenue.
Systems connected
- Circana
- Numerator
- Retailer POS
- Sales forecasts
- Review transcripts
Team: three engineers across data, AI and category analytics · the $180M figure is the sum of sized, named moves the category teams accepted into plans
Share Δ over four weeks. Each move sized in revenue and named to an owner.
- Brand A −0.8
- Price gap vs PL widened past $1.10 → PPA: shrink gap to $0.85 · +$38M
- Brand B +0.4
- TPR depth working at 25%, not 30% → trim promo depth · +$21M margin
- Brand C −0.3
- Velocity fine; TDP loss at two retailers → win back distribution · +$44M
POS, competitor POS, market share, TDP and distribution, promo, sales: that is true performance output data. Pair it with the media data and you get really strong insights.
What moved
$180M
Revenue identified in 90 days, across tiered pricing, persona-based engagement and churn
10× faster
Category planning cycle
File 05Executive performance reportingFortune 100 CPG6 weeks · 2 engineers
The monthly review shows what happened, never why, and never what last month's decision produced.
What we built
One read over finance, sales and operations, on validated KPI definitions. Variances arrive already decomposed to drivers, board-ready slides generate in the company template, and every decision from the review is tracked to the outcome it produced, in time for the next one.
Systems connected
- Finance
- Sales
- Ops
- Pipeline
- AR
- Inventory
- CX metrics
Team: two engineers, six weeks to production · most of leadership's data questions are now answered live in the meeting instead of "we'll come back to you"
Gross margin −110 bps vs plan: 70 bps is input-cost timing that reverses next period. The real miss is 40 bps of trade-spend drift in two accounts.
- Decomposed
- Input costs −70 bps (timing) · trade spend −40 bps (structural) · mix +0
- Recommended
- Reset accrual caps on the two accounts before quarter close
- Last cycle
- Tracked: September's pricing action landed +2.1% net revenue vs +1.8% predicted
The ambition is to get everything into the central intelligence layer, because then everything gets triangulated and contextualized against each other.
What moved
2 hr → 90 s
Per board-ready deck, in the company's own template
2.5× faster
Time-to-insight across the review cycle
File 06Budget reallocation · the loop, closedSame client as file 012 quarters to policy
Why this file matters
Files 01 to 05 make work faster. This one is different in kind: a recurring capital decision now runs on the accumulated record, with a human holding the gate. This is what the other five compound into.
Fine, it can describe money. Can it move money?
What we built
A standing quarterly policy: take 10% of last quarter's spend from the three lowest-ROI channels and move it to the three highest, with predicted incremental sales attached to every shift. It moves $6.2M a quarter, and a person approves every move. The following quarter, the system scores its own prediction, and the next proposal builds on that record.
Move $6.2M (10% of Q2 spend) from the three lowest-ROI channels to the three highest, effective next flight.
- From
- Linear TV (late) · Display · Audio
- To
- Retail media · Paid social · Search
- Predicted
- +$9.8M incremental sales over two quarters
- Scored
- Q2 proposal: predicted +$8.1M, measured +$7.6M (94%)
- Approver
- VP Media Strategy; a person approves every move
We are blocking 10% of last quarter's spend from the three lowest-ROI channels to the three highest, with the predicted incremental sales.
What moved
10% of spend
Now reallocates by evidenced policy, every quarter
94%
Q2 reallocation scored after the fact: predicted +$8.1M, measured +$7.6M
03 · The system underneath
Nine files. One system underneath.
Every file above runs on the same build: your data resolved to validated definitions, live signals watched against them, and every decision recorded with what it returned. That shared operational context is why file 06 was possible at all, and why each new file starts further along than the last.
At install Day 1
- Sources connected
- 14
- Validated learnings
- 0
- Decision records
- 0
- Actions cleared
- 0
Recommendations: industry-generic. Any fresh install would say the same thing.
Metric definitions: contested. Three versions of ROI found across teams.
Twelve cycles later Day 90
- Sources connected
- 14
- unchanged
- Validated learnings
- 41
- ▲ from 0
- Decision records
- 26
- ▲ from 0
- Actions cleared
- 6
- ▲ from 0
L-041 · Validated ×3 cyclesProof-led creative outperforms feature-led in enterprise accounts.
L-017 · RetiredHeld only during promotion weeks. Conditions recorded so it isn't relearned from scratch.
ROI · UnifiedFinance and media now reconcile without a meeting. One definition, with lineage.
What the deployment leaves behind
Where it runs
Your cloud tenancy. Your access controls. Data stays in your estate.
What you keep
The repo, the connectors, and the decision record. Admin included at handover.
What's swappable
Models, switched per task as the frontier moves. The record isn't.
04 · The test of a specialist
We knew what TDP meant before the kickoff call.
The work above required knowing what these words mean and which fights they settle. The operating language of the teams whose decisions we were asked to change.
Every file in this dossier started with a conversation our engineers could hold without a translator. That is the difference between a build that ships in six weeks and one that spends its first quarter learning the business.
Media & marketing
MMM vs. MTA reconciliation · reach curves and effective frequency · flighting and pacing · SOV-to-share dynamics · creative wear-out
Shipped → files 01, 03, 06 · waste detection, trend activation, reallocation policy
Category & sales
Sell-in vs. sell-out · TDP and velocity decomposition · price-pack architecture · promo depth and TPR lift · price gaps vs. private label
Shipped → file 04 · category scorecards with sized moves
Supply chain & S&OP
Consensus demand planning · DOS and fill-rate exceptions · allocation logic · forecast value-add · the pre-read cadence itself
Shipped → file 02 · the 90-second pre-read
Finance & executive
Variance decomposition · driver trees · trade-spend accruals · outcome tracking on prior decisions · one validated definition per KPI
Shipped → file 05 · the review that explains why
05 · How a file gets opened
Twelve weeks, no mystery.
Every file above followed the same shape: the lighthouse. You keep something at the end of each step.
01Days 1–2
Scope
Pick one recurring decision. Two engineers, a scoping call, a written scope. First insights inside 48 hours of access.
- Commit
- One meeting
- You keep
- The written scope
02Weeks 1–2
Connect
Scoped sources in your tenancy. KPI definitions validated first, because that's where the two-week grinds live.
- Commit
- Read-only access
- You keep
- Validated definitions
03Weeks 3–6
Shadow
You can stop here — the findings are yours
The system runs beside your operators. You inspect the evidence, the caveats, and the misses. File 02 matched two full cycles before a meeting was retired.
- Commit
- Two cycles
- You keep
- The findings
04Weeks 6–12
Production
Stop any time — everything is yours
The loop goes live where your team works: Teams, Slack, the deck template. Authority widens one task at a time.
- Commit
- One workflow
- You keep
- Everything
05Day 90
Handover
Nothing to migrate — the system is already yours
Written go/no-go. You keep the repo, the connectors and the record. Most clients open a second file instead.
- Commit
- Written go/no-go
- You keep
- Repo, connectors, record
Appendix A · For the CIO and the data organization
The consumption layer your data platform has been missing
You built the lakehouse, the governance, the semantic ambitions. What stalled is trust at the point of consumption. Business teams don't act on answers they can't verify. Queries run through a semantic layer over your unified ground truths, with deterministic validators: an answer either reconciles to source, with lineage attached, or it doesn't ship.
That is why every file in this dossier was possible, and why the business teams in files 01 to 06 stopped arguing about whose number was right.
Audit
The full chain, queryable
Every output tracked: sources, validators passed, approver, and eventual outcome, for every AI-produced artifact.
Interop
Composes with what you run
A2A, MCP and API surfaces. Agents compose with the systems and copilots you already run, ours and anyone else's.
Models
Model-agnostic by design
Frontier models switched per task as the frontier moves. No fine-tuning lock-in.
Ground truth
Validated, or it doesn't ship
Deterministic validators reconcile every answer against source before it reaches a decision-maker.
Footprint
Runs in your tenancy
Over the platforms you already own. Data does not leave your estate.
Works with
- Databricks
- Snowflake
- SAP & legacy ERPs
- SharePoint
- Salesforce
- Circana
- Nielsen
- Kantar
- Numerator
- Meltwater
Portability
Instant data portability
The record exports in open formats. Repo, connectors and admin transfer to you at handover.
Security
SOC 2
Access through your existing identity and data-loss controls; full security packet at the scoping call.
Commercials
Fixed-scope build
No per-seat or per-token pricing. Code, connectors and the record transfer at completion.
“By the time you're redeploying any of your spend, you're looking at data that's five or six months old. Which is crazy.”
Bring us a file worth opening.
A 45-minute scoping call with the engineers who would build it. Pick the recurring decision that hurts and leave with a written scope and a 48-hour plan.
Book the scoping callSending this internally? Route your CMO to files 01-06, your CEO to file 06, and your data organization to Appendix A.