Foundry Data & workflow platform
Connects records such as orders, stock and shipments so teams can review the same information and act on it in their daily operations.
Palantir Technologies
Taking operational AI into production.
Palantir builds data and AI software for businesses and public institutions. From 2021 to 2024, I led two PMs and coordinated 20 specialists across commercial accounts in four industries. Our work connected operating data to decisions employees could act on, then helped clients move accepted workflows into daily use. I set priorities, developed the PMs and agreed client delivery commitments for Foundry and, later, AIP.
Product Team Lead2021–2024
Product platforms in my scope
AIP Artificial Intelligence Platform
Introduced in 2023, AIP connects generative AI to business data so people can review proposals within their operating workflows.
01What was the friction?
A 13-hour wait delayed the next operating decision.
Planners responding to a shortage had to reconcile separate order, stock and shipment records before another team could approve a response. Across four industrial clients, a ready request waited a median of 13 hours for the next team to accept responsibility, delaying the operating decision. I traced handoff waits with the workflow PM and operators, agreeing a consistent method so we could identify the information and ownership gaps behind the delay.
Trace the handoff with operators.
Walkthroughs showed planners checking separate records and requests waiting for an authorised owner. The PM measured the handoff consistently across four clients, establishing a 13-hour median baseline.
Request ready · Records checked separately
Stock
Orders
Shipments
13 hmedian wait
Next teamAccept responsibility
Baseline: ready request → accepted responsibility · Four industrial clients
The walkthroughs defined one complete response: connect the records, name the approver and retain the decision. I shaped these requirements with the PM, using a nine-hour handoff target to assess the change.
Stock
Orders
Shipments
Supply plannerReview & approve
Success measure: median handoff ≤9 hours · Same four-client study
Find where approval responsibility was missing.
The PM and I reviewed approvals with client operations leaders, finding decisions waiting for an authorised owner. We agreed the named approver, permission checks and retained decision context for release.
Client operationsIdentify the authorised approver
Release requirementsOwner, checks and decision record
Release condition: an authorised action with a complete decision record
Assess the move from a pilot into daily use.
From 2023, AIP Bootcamps tested a client task through a short pilot. The adoption PM reviewed buyer agreement, an operating owner, live data access and support before I committed capacity for daily use.
The client had to accept the workflow and agree how daily operation would work. Each open requirement had a named owner and a next action, allowing the PM to resolve the gaps before the account entered production.
Client owner
Live data access
Ongoing support
Production condition: the client accepts the workflow and operating plan
02What approach and priorities did I choose?
Prioritising the complete response before wider reporting.
The handoff review made the complete shortage response the first priority: it addressed the waiting we had measured and could be tested against a nine-hour target. I deferred wider reporting, made approval records part of the release and assessed new requests by their expected benefit, effort and effect on other accounts. When AIP became available in 2023, the same discipline extended to pilots, with rollout capacity tied to client readiness.
01 / FirstOperating delay
Complete the shortage response
Walkthroughs linked waiting to scattered records. I brought the order, stock and response into one workflow, ahead of wider reporting.
02 / RequiredApproval ownership
Make approval traceable
Decisions also waited for an authorised owner. I made the approver, checks and a retained action record part of the complete workflow.
03 / ExpandClient readiness
Commit to supported daily use
AIP pilots needed client ownership, live data and support. I reserved rollout capacity when the agreed workflow could enter daily use.
Each choice addressed a finding from discovery and was weighed against benefit, effort and existing commitments.
- 2021Foundry
Resolve the shortage
The workflow joined orders, stock and the response so a planner could resolve the shortage in one task.
Target: median handoff ≤9 h - 2022Action Log
Retain approval context
Action Log retained who approved a change and its operating context across nine Workshop workflows.
Record: action, owner & context - 2023AIP Bootcamps
Prepare daily use
AIP Bootcamps tested client tasks. Rollout depended on an operating owner, live data and agreed support.
Ready: owner, data & support - 2024AIP evaluation
Validate AI recommendations
AI recommendations were retested on the same 840 cases to decide which tasks were ready for release.
Target: ≥90% expert acceptance
Each release followed the same sequence: investigate the task, define acceptance checks, then agree the operating plan.
03How did I lead the team?
By giving each PM the authority to deliver the plan.
To deliver this plan across competing accounts, I led two PMs who owned the operating workflow and client adoption, with 20 shared specialists supporting both. The PMs could negotiate within an agreed client scope; I coached their recommendations and decided changes to staffing or commitments. Weekly reviews followed waiting time and readiness, while monthly reviews with functional and commercial leaders resolved capacity before the next release.
Felipe Lowe
Product Team Lead
Account priorities, PM development and changes to staffing or client commitments sat with me. In one-to-one reviews, each PM tested a recommendation against the evidence, expected client benefit and delivery effort.
Operational workflow PM
The workflow PM owned the complete user journey and release checks, using waiting-time reviews to refine the solution within the agreed client scope.
Client adoption PM
The adoption PM agreed the client plan, secured an operating owner and confirmed support requirements before coordinating the move into daily use.
Weekly product reviews
Waiting time, usage and open ownership or data requirements informed the next product decision, with a named owner and next action for each issue.
Monthly capacity reviews
Account readiness and existing commitments informed staffing changes with functional and commercial leads, keeping capacity and delivery plans aligned.
How the 20 specialists supported delivery
Client-facing engineering7
Data engineering4
Applied AI2
Product design2
Implementation3
Product operations2
22 contributors including the two PMs. Specialist capacity was shared and agreed with functional leaders; my sponsor was the Director of Commercial Product.
04Which bottleneck shaped the delivery plan?
Client readiness determined the rollout commitments.
Client ownership, live data and ongoing support determined whether an accepted AIP pilot could move into daily operation. I assessed those conditions with the adoption PM, then agreed rollout scope, specialist capacity and support with functional and commercial leads for each client’s entry into daily use.
05How did we deliver and validate the product?
By turning the priorities into a workflow people could use.
We built the planner’s workflow in Workshop, Foundry’s application builder, connecting the order, available stock and proposed transfer to an authorised approval. The reconstruction below lets you review that transfer and inspect the record it creates. The release evidence explains how operator walkthroughs refined the screens and checks, and how later AIP evaluations helped define which recommendations were ready for supported daily use.
Workflow example
Release evidence
Resolve material shortage
Awaiting approvalOrder 1048 · Plant A · Supply planning
- Order requirement
- 160 units
- Available at plant
- 40 units
- Proposed transfer
- 120 units
At the plantProposed from the warehouse
Transfer 120 units from the regional warehouse to Plant A. The available stock and transfer window cover the order while the delayed shipment is rescheduled.
The planner checks the proposed transfer against the order, stock and delivery window before approval; the saved record keeps the action, owner and operating context together for supervisor review.
The example creates an approval record.
Approve the proposed transfer to inspect the record. It keeps the action, responsible owner and operating context together for a supervisor to review.
Example approval record
Created when you approved the transfer in this walkthrough.
- Action
- Transfer 120 units
- Approved by
- Supply planner
- Affected order
- Order 1048
- Route
- Regional warehouse → Plant A
- Context retained
- Stock, order and delivery window
- Recorded at
- At approval
The approval record links the 120-unit transfer to Order 1048, the supply planner and the stock and delivery window reviewed for the decision, giving supervisors a complete basis for tracing the action. The PM can use the retained context to investigate recurring gaps and refine the checks that guide subsequent transfers.
How operators tested the workflow.
Walkthroughs tested whether a planner could judge the transfer using the order, stock and delivery window. Review questions guided the PM’s screen changes to the operating evidence, proposed response and approval steps.
Order160 units required for Order 1048 at Plant A.Stock40 units at Plant A; a 120-unit transfer.Delivery windowThe transfer window covers the order’s delivery needs.
The walkthrough followed the same handoff used in the time measure: a ready request passed to the next responsible team with its order, stock and transfer window visible. Each review question informed the PM’s next screen change, while the saved approval let supervisors trace the owner, action and operating context.
Conditions required before release.
- Current limits
- Check current stock and the transfer window against the remaining order quantity, so the planner can confirm a feasible response before approval.
- Authority
- The supply planner’s permissions are checked before approval, keeping the transfer with a named owner authorised to act on the affected order.
- Decision record
- Keep the action, owner and operating context together, giving supervisors a complete approval record and the PM evidence to investigate recurring gaps.
How we tested AI recommendations.
92%accepted by experts
0%Target ≥90%100%
773 accepted67 to review
After corrections, 773 of the same 840 cases passed expert review, exceeding the 90% target. The remaining 67 cases still needed corrections or operator review.
First review: 613 accepted and 227 to review. Second review: 773 accepted and 67 to review.
Criteria used in both reviews
- Relevant operating information
- Stock and capacity constraints
- Clear approval responsibility
- Practical usefulness of the response
We kept the cases and criteria unchanged to compare both reviews. The PMs used the accepted task types to set the next release’s scope.
06What did the OKRs and KPIs show?
The review showed shorter waits and six live accounts.
At the 2024 review, the PMs compared handoff times, repeat AI tests, client readiness and signed agreements with the targets agreed before delivery. An eight-hour median wait and 92% expert acceptance met their targets. Six of nine client pilots reached daily use, one below plan, and five of those clients signed US$14M in three-year expansions. These results guided the next commitments, with readiness and quality gaps assigned for follow-up.
Less waiting between teams
Median handoff · Four industrial clients
13 → 8 h38% reduction
Before 13 hAfter 8 h
160
Target ≤9 hoursTarget met
What the result changed
The eight-hour median beat the nine-hour target. I used the measured reduction in waiting between teams to support extending the complete exception workflow.
See the measurement details
The PM compared waits before and after the workflow change across four industrial clients, measuring from a ready request to the next team accepting responsibility. The median fell from 13 to 8 hours: 38.46%, rounded to 38%.
More recommendations accepted
Expert review · The same 840 operating cases
73% → 92%+19 percentage points
First 73%Second 92%
613 → 773 of 840100%
Target ≥90% acceptanceTarget met
What the result changed
Acceptance exceeded the 90% target after corrections. The PMs used the 67 remaining failures to limit release scope and plan further operator review.
See the measurement details
The team corrected rejected recommendations and repeated the same 840-case review with unchanged criteria. Experts accepted 773 cases, up from 613 initially, raising acceptance from 72.98% to 92.02%, a gain of 19.05 percentage points.
Pilots entering daily use
Nine selected accounts · AIP Bootcamp programme
6 of 9production accounts
06070809
5 expanded1 existing scope3 in pilot
Target ≥7 accounts1 more to target
What the result changed
Six accounts reached daily use, one below the target of seven. I kept each remaining pilot tied to an owner and readiness action before committing capacity.
See the measurement details
The adoption PM reviewed nine selected AIP Bootcamp accounts. Production required an accepted workflow in daily use, an operating owner, live data access and ongoing support. Six met these conditions against a target of at least seven.
Signed enterprise expansion
Five agreements · Three-year total contract value
US$14MUS$2M above target
2.2
2.5
2.6
3.1
3.6
Three-year value per agreement · US$M
Target ≥US$12MTarget met
What the result changed
Signed expansions exceeded the value target by US$2M. I aligned each new commitment with an accepted scope, support plan and delivery capacity.
See the measurement details
Five production accounts signed expansions after accepting the workflow and delivery plan. Their three-year agreement values total US$14M against a US$12M target. The sixth live account remained within its existing agreement.
The four-client handoff study, nine-workflow record audit, 840-case evaluation and nine-account adoption programme each have their own measurement scope.
Product context & measurement notes
How the products fit together
Foundry connects operational data. Its Ontology links business records and actions; Workshop turns those connections into applications for employees. AIP adds AI to that context, with expert evaluation informing how recommendations are used.
Reading the measures
Results refer to the selected programme at its 2024 review. Handoff time is a median wait between teams, and both AI reviews use the same 840 cases. Expansion is the combined three-year contract value. The 400-record audit is a separate, ongoing KPI. Each result includes its measurement definition, review group and responsible owner.