P&G has put three briefs in front of us — AI-Enabled Brand Building, Future of Retail, and Top 10 Strategic Questions for Baby Care. They can be read as three separate pitches, each competing inside its own crowded point-solution market, or as one connected ask that only an ecosystem platform can answer. The second reading is the one P&G is actually looking for, and it is the one Foundation-AI was built to satisfy.
This brief lays out the case in four parts. First, what the three briefs are really asking for once read together. Second, what Foundation-AI is and what is production-ready today. Third, three concrete offers — one per brief — that deliver evidence inside 90 days each. Fourth, the strategic claim worth repeating.
The window closed on two of the three briefs months ago and P&G is still talking to us. That fact, more than anything in the briefs themselves, tells us what they are looking for and have not yet found.
Read separately, the briefs look like three separate asks. Read together, they are three views of one underlying need.
Brief 1 — AI-Enabled Brand Building. The last paragraph in the use-case list reads, verbatim, *"Ecosystem solutions ie. Platforms that enable brands to move beyond point solutions to a better integrated AI-enabled ecosystem."* That is the entire brief in one line. The ten use-cases listed above it are not ten separate problems — they are ten cases that all flow from the same missing layer. P&G has bought all ten as point solutions over the last five years, and is now asking who can give them the connective fabric beneath.
Brief 2 — Future of Retail. The opening paragraph names the actual problem: *"Major retailers like [Walmart, Amazon] are investing heavily in AI, proprietary data platforms, and media networks — behaving less like traditional retailers and more like tech companies."* The brief asks for solutions in three areas, but the binding theme across them is agent-to-agent collaboration — P&G's agents and the retailer's agents need to be able to work together on shared problems under shared rules. This is an ask for a trust layer between two companies whose AIs are about to start negotiating with each other.
Brief 3 — Top 10 Strategic Questions. This one looks different on the surface. It is not framed as a technology brief — it is framed as ten strategic questions about Baby Care growth. But each of the ten questions is a long-horizon thesis P&G is testing. The brief is asking how P&G should be answering these questions as the world keeps moving. The temptation is to send back a strategy memo. The better answer is to send back a way to keep answering these questions — a living thesis register that re-scores as evidence arrives.
Across the three briefs P&G touches four parts of its value chain — brand creation, retail relationships, category strategy, and consumer engagement. The three briefs are written separately because they originate in different teams. The unmet need is the same across all of them: a connective layer where agents can work together, memory compounds across cycles, evidence is auditable by construction, and the boundaries between P&G and its partners are written down, enforceable, and respected.
That is what Foundation-AI is.
Foundation-AI is a multi-component agentic platform: software agents that read from existing systems and act on them under supervision; a memory layer that compounds across cycles; a control layer for AI model usage with cost limits, output safety, identity carrying, and audit receipts; per-organisation data-boundary controls with written, enforceable policies; audit by construction on every event; end-to-end observability; and a strategic-intelligence layer for tracking long-term investment theses.
There are four capabilities under that one fabric. They operate the same way regardless of which P&G surface they apply to.
Agents that watch and act, alongside the people who already do the work. Software agents — narrow, focused, supervised — read from the systems P&G already runs, surface patterns and recommendations to the people responsible, and learn from how those people respond. They do not replace the brand managers, the category leads, the retail negotiators, or the strategy teams. They make sure the right information reaches each of them at the right moment.
A memory that compounds, rather than starting from scratch. Every operational pattern, every decision, every cross-partner signal flow is captured as a queryable layer. The next time a similar situation occurs, the relevant prior context surfaces automatically. Institutional knowledge that today walks out the door with rotating brand teams or relocating PMs stays with the firm.
An audit trail built into how the system operates, not assembled afterwards. Every action the AI takes is recorded at the moment it happens — what was observed, what was decided, by which model, with what supporting evidence, under which policy. Brand-consistency claims, sustainability evidence, retailer-collaboration commitments, category research conclusions all carry their own evidence with them by construction.
Boundaries that respect each partner's rules. Each company in the network — P&G, a retailer, a creative agency, a media platform, a research provider — keeps its own data within its own boundary. Only what is explicitly permitted by written policy crosses between them, and every crossing is logged with the policy that allowed it.
These four capabilities, working as one fabric, are what distinguish a platform answer from a point-solution answer.
| Capability | Status |
|---|---|
| Software agents (signal-watching, content adaptation, orchestration) | Shipped |
| Memory and knowledge corpus (queryable, compounding) | Shipped |
| AI model control layer (cost gate, safety scrub, identity, receipts) | Shipped |
| Per-organisation data-boundary controls with written policies | Shipped |
| Audit by construction (provenance on every event, every decision) | Shipped |
| Tamper-evident proof anchoring for high-stakes decisions | Shipped |
| End-to-end observability (industry-standard tracing) | Shipped |
| Strategic-intelligence layer (thesis-as-graph, continuous re-scoring) | Shipped |
| Agent-to-agent collaboration between organisations | Architecture shipped at platform layer; first production deployment is the engagement we are proposing |
Anchor recommendation: Personal Health Care. Beauty and Fabric Care work too — the shape is identical.
Duration: 90 days.
The brief lists ten use-cases. We commit to four in 90 days, with explicit demonstrations of each.
1. From a consumer signal to a creative concept, in days rather than quarters. A brand director asks, *"What is the strongest emerging signal about this consumer segment that we have not yet acted on?"* The system reads consumer signal across the channels P&G already monitors — search, social, retail, brand surveys, sentiment data — and surfaces ranked patterns with the supporting evidence. The brand director picks one. The system drafts three creative-concept directions in the brand's voice. The brand team curates. The next round starts from accumulated learning rather than a blank page.
2. One asset, every platform, every variant, all under brand-consistency rules. A campaign concept becomes a message, an image set, three video formats, an animation, and platform-specific variants for Instagram, TikTok, YouTube, retail-media units, and connected TV — all in one pass. The brand team sets the rules — voice, claims, sustainability evidence, regulatory permits — once. The system enforces them on every asset, every variant, every cycle.
3. Brand visibility in the new generative-search world. P&G's consumers increasingly find brands through Google's AI overviews, Perplexity, ChatGPT, and Claude rather than through a list of blue links. The system continuously watches how P&G brands surface there. When a brand is described inaccurately, surfaced in a misleading context, or missing from a high-intent query, the SEO and brand teams see the pattern, the supporting examples, and a prioritised list of corrections.
4. The content supply chain, end to end, as one continuous record. A new asset originates in the brand brief, passes through compliance review, gets formatted for each destination, lands in each media platform, and tracks through to performance — as one continuous record with provenance at every step. Today this flow breaks at every junction. We make it one operation.
The end consumer is the P&G consumer — a parent buying Pampers, a customer choosing Olay, a household manager picking up Tide. Their experience changes in three ways.
1. Discovery becomes more relevant. They see content shaped by signal from people like them, in language that fits the platform they're on, with claims that have been verified before publication. The change is invisible by design — they don't see "AI-generated content," they see better-fitting content.
2. Personalisation respects their data boundary. Their data stays inside the perimeter the privacy policy permits. Every cross-boundary use is logged with the policy that allowed it.
3. Continuity across touchpoints. A consumer who interacts with the brand on social, then on retail, then in-store, doesn't get three disconnected experiences. The brand's understanding of them compounds — within the explicit consent boundaries they've given.
The brand-team experience: a brand director opens one screen instead of six. The screen surfaces what needs attention today, with the supporting evidence and the recommended actions. The brand director still makes every call. The screen makes sure the right calls reach them at the right moment.
Continuous consumer signal, ranked by relevance. Software agents continuously monitor the channels P&G has configured — social, search, retail telemetry, brand surveys, sentiment data — and surface the patterns that matter, ranked by relevance to the brand's named priorities.
Content generation under brand rules, with cost and safety controls. The platform runs every content-generation request through a control layer that picks the right AI model, enforces the brand's voice and consistency rules, applies a cost limit, and scrubs the output before the asset reaches the next step. Every asset traces back to its source brief, the model that produced it, and the approval chain that signed it off.
Brand visibility in AI-search, watched continuously. Software agents query the AI-search surfaces P&G's consumers use — Google's AI overviews, Perplexity, ChatGPT, Claude — on a configured cadence, capture how P&G brands appear, and rank the patterns. When a brand is misrepresented or missing from a high-intent query, the team sees the evidence and corrective actions on the next morning's screen.
End-to-end audit trail, built into how the work happens. Every observation, every decision, every asset, every approval, every platform delivery is one continuous record. When Legal asks "show me the chain of evidence behind this campaign claim," the answer is a query that runs in seconds.
Operator screens designed for the way brand teams already work. The brand team gets one screen rather than six tools. The screen mirrors the rhythm a brand director already runs on — the morning review of what's moving, the afternoon decisions on what to push live, the weekly review of what's compounding.
A working brand-building pipeline at one Personal Health Care campaign cycle, with full audit trail, measured against three metrics: cycle time, brand-consistency compliance, and content-quality lift against the brand's own scoring rubric. Phase 2 extends the pattern to the brand's other campaigns, then to other Personal Health Care brands, then to Beauty and Fabric Care.
Anchor recommendation: P&G picks. Walmart for the AI-as-tech-company shift, Amazon for agentic commerce maturity, Kroger for retail-media depth. The pattern is identical.
Duration: 90 days, plus a written boundary policy that becomes the template for every subsequent retailer.
The brief names three strategic areas with nine specific domains. We commit to one signal exchange in 90 days that delivers value on its own and stands up the working contract every subsequent exchange will reuse.
1. Real-time inventory across one retail relationship. P&G's category team needs to know — in minutes, not days — what inventory of which P&G product is in which of the retailer's distribution centres, what is moving at what velocity, and what is at risk of stocking out. The retailer has the data. The data has historically lived behind a wall. We make the visibility live without either side compromising its data perimeter.
2. Joint category management, working from the same numbers. When P&G and the retailer work a shelf reset, a planogram refresh, or a category review, both sides bring their own data, their own AI tools, and their own analytical frameworks. Today the work runs through working sessions where humans broker the data manually. We let P&G's category AI and the retailer's category AI work against the same shared view of the consumer-and-shelf data each is permitted to see.
3. Retail-media spend, traceable end to end. Every retail-media impression P&G buys carries a record of which targeting decision, which content variant, which model, which budget cap, which outcome. When the media team asks "did this retail-media campaign actually drive incremental sales for this brand at this retailer," the answer is a query, not a reconstruction.
4. The joint planning rhythm, automated where it should be. Quarterly business reviews, joint business plans, promotional calendars — today these run on email, spreadsheets, and meetings. We automate the structured parts (calendar coordination, document version management, action-item tracking, reminder cadence), so the human meetings focus on the judgement that requires humans.
Two layers. The first is the retailer-side category buyer (in effect, P&G's customer). The second is the end shopper.
The retailer-side category buyer experiences the change directly:
1. One shared view, instead of two competing views. Instead of P&G's account team showing up with one set of slides and the retailer's category lead with another, both sides operate against a shared view of the data each is permitted to see — same numbers, same definitions, same time-stamps. The conversation moves from "what is the truth" to "what should we do about the truth."
2. Continuous coordination, not quarterly cycles. Joint business plans become living documents that update as evidence arrives, rather than quarterly setpieces that are stale by the time they're reviewed.
3. Defensible evidence behind every joint decision. When the retailer's leadership or P&G's leadership later asks "why did we make this call," the answer is in the record.
The end shopper — the actual person buying P&G products at the retailer — experiences the change indirectly:
1. Better availability. Real-time inventory visibility means out-of-stock events drop. The shopper finds what they came for.
2. Better in-store and on-site experience. Joint category-management optimisation means shelves are organised the way the consumer actually shops, and on-site retail-media is more relevant to what they're actually trying to find.
3. Better post-purchase continuity. Retailer-side data and brand-side data, governed under consent boundary, mean the brand can offer follow-up that respects the consumer's relationship with the retailer rather than ignoring it.
Boundary policy as a written, enforceable contract. Each side declares what its AI can see, what it can act on, what it can share, and what is forbidden. The policy is a real document, reviewable by both legal teams. The platform enforces it at the data-flow layer — not as a guideline, as enforcement. Every cross-boundary access is logged with the policy that allowed it and a reason code.
A shared signal channel that respects both sides' rules. One side publishes a typed signal (an inventory update, a category-management proposal, a retail-media performance event); the other side subscribes; both sides see the same delivery, with delivery guarantees. The channel is durable, recoverable, and auditable. The architecture is shipped and tested at the platform layer; the first production deployment between two anchor organisations is what this Retail pilot lands. P&G can be the first reference deployment.
Standard ways for both sides' AI agents to use the tools P&G already runs. Whether the agent needs to read inventory, propose a planogram change, draft a creative variant, log a decision, or summarise a meeting, the platform provides a standard way to do it. The agents do not have to be hand-built for each new tool. The integration depth includes a deep CRM connection so the relationship history between P&G and the retailer is a queryable layer wherever the boundary permits.
Coordination when both sides act at once. When P&G's agent and the retailer's agent are simultaneously trying to update the same joint plan or category record, the platform resolves the contention without lost updates.
Tamper-evident proof of joint decisions. Joint decisions can be anchored as records that both sides can independently verify, today and a year from now. This becomes useful when, months or years later, someone asks "did we actually agree to this?" The record answers.
One working agent-to-agent signal exchange (recommend real-time inventory visibility), one working joint-category-management workflow, one working retail-media-provenance demo, plus the boundary policy document — the document that becomes the template for every subsequent retailer relationship. Phase 2 extends to a second retailer; Phase 3 to all P&G's strategic retailer relationships, with each new retailer faster than the last because the boundary-policy template and the agent-to-agent fabric carry forward.
Anchor: Baby Care, as named in the brief. Extension to other categories is Phase 2.
Duration: Four-week initial workshop plus six months of continuous re-scoring.
The brief poses ten strategic questions. We commit to landing all ten as living theses inside the four-week workshop, then keep them re-scoring continuously for six months.
1. Super-premium versus mainline segment viability — held as a thesis with sub-claims (price-tier elasticity, consumer trade-up willingness, private-label dominance trajectory) and named decay signals.
2. Adjacent-category value-creation potential — five sub-theses (toiletries and skincare, pet-care diapering, wipes, health-tech parenting, on-the-go), each with its own evidence chain.
3. Sustainability and ingredient transparency as a growth lever — a thesis tracking which materials, certifications, and claims are moving consumer preference, updating as research and regulatory signal arrives.
4. Future go-to-market models — five sub-theses (direct-to-consumer, marketplaces and registries, social commerce, online communities and gifting, disruptive trial), each with named cohort-economics signals.
5. Connected and smart parenting — a thesis about IoT and software architecture in the category, with the named bets as sub-claims.
6. Loyalty and community as strategic assets — build versus buy versus partner expressed as three competing sub-theses.
7. Beyond diapering — extension into potty training and early education, held as a thesis about category continuity and emotional brand reach.
8. Stress-free parenting solutions — a thesis about consumer pain points and the product-and-service combinations that resolve them.
9. Earlier and broader engagement — pregnancy, grandparents, nannies — as a thesis about expanded influence with named target-cohort signals.
10. Market dynamics — private-equity investment, consolidation, ownership-landscape evolution — as a thesis about industry structure with named M&A and fund-formation triggers.
The "consumer" here is the P&G strategy-team member working with the theses every week. The end consumer (the parent, the caregiver) is the subject of the theses, but they don't interact with the system. Their journey enters indirectly: as the theses score themselves better, P&G's bets in the category serve them better.
The strategy-team member's experience moves through three phases.
Phase 1 — workshop weeks 1 to 4. The strategy team spends roughly half a day per thesis in structured working sessions with our team. We help them break each question into its claims, its underlying assumptions, its evidence anchors, and the relationships between them. By week 4 each of the ten questions exists as a structured object the team owns.
Phase 2 — first 90 days of operation. Continuous evidence flow into each thesis. Research agents read external sources — research publications, regulatory dockets, consumer-panel data, M&A announcements, competitive launches — and write the relevant evidence into the right thesis. Strategy-team members see their theses re-score in days when major evidence shifts, with the supporting source links one click away.
Phase 3 — six months and beyond. The thesis register becomes the strategy team's working memory. Quarterly leadership readouts to P&G's Baby Care leadership and Strategy Office become direct queries against the register, with the supporting evidence one click away. New theses get added in days, not strategy cycles. The team keeps the institutional memory even as members rotate.
Each strategic question stored as a structured object that re-scores itself. Rather than the question living in a quarterly memo, it lives as an object on the platform — with its claims, its assumptions, its evidence anchors, and the connections between them. When new evidence arrives, the relevant claim updates, the connected claims re-score, and the team is notified.
Research agents that continuously read the world for the strategy team. Software agents read public filings, regulatory dockets, industry press, supply-chain trackers, consumer-panel data, and competitive signals overnight. Each agent's findings are written into the right thesis with structured evidence — what the source said, what the inferred implication is, and which of the team's named claims it reinforces or contradicts.
Operational signal from inside P&G, in the same evidence flow. The same agents that read external signal can monitor specific operational sources P&G already runs (their own retail telemetry, consumer-panel data, brand-tracking studies, M&A databases, regulatory feeds) and write the relevant findings into the thesis register. External signal and internal signal land in the same evidence layer.
A strategist's screen, designed for how strategists actually think. The team gets a screen showing the ten theses, their current confidence scores, the recent evidence flow, the alerts that crossed thresholds, and the open questions. Drill-down on any thesis surfaces the supporting evidence chain.
Integration with the tools the strategy team already uses. The team continues to work in the tools they know — Microsoft Office, Google Workspace, internal SharePoint, Outlook, the strategy team's own dashboards. Those tools can read into and write back to the thesis register through standard connections.
Memory that compounds. Every thesis update, every accepted evidence record, every team-member decision and the reasoning behind it becomes part of the platform's memory. When a new strategy-team member joins, the full history of every thesis is one query away.
In the four-week workshop: all ten theses as structured objects, evidence sources wired, the strategist's screen live, and the team trained on the operating model. Over the following six months: continuous re-scoring with monthly leadership readouts. Phase 2 extends to other P&G categories — Beauty, Personal Health Care, Fabric Care, Family Care, Home Care, Grooming, Oral Care.
Foundation-AI is not just one set of AI agents. It is the fabric that lets many agents — ours, P&G's own, a partner's, a retailer's, a media platform's, a regulator's — work together on shared problems while each company keeps its own data and its own rules.
The interoperability format that makes that possible is PACT — Protocol for Agent Collaboration and Trust. PACT is the contract language between two agents that belong to two different organisations. When a P&G agent and a Walmart agent need to share a piece of operational signal — say, a container's expected availability at a particular distribution centre, or a category-planogram proposal — PACT is what carries the message between them. It is the equivalent of a written agreement between two diplomats, but expressed in a form that the agents themselves can read, sign, and honour.
In plain language, PACT does three things:
PACT is what makes an agentic mesh possible without compromising either company's perimeter. Foundation-AI is the fabric that runs PACT-formatted exchanges across all the agents that need to collaborate — whether those agents are ours, P&G's own, a partner's, or a third-party vendor's. The next agent P&G adds — whatever it does, whoever builds it — drops into the same fabric, on the same trust contract, governed once and reused everywhere.