What Flows and What Stalls:
Designing How Organizations
Make Sense With Their Data
On organizational responsiveness in a moment of constant change.
We are in the most data-rich moment in organizational history. Yet organizations still struggle to move insights through the business to drive decisions and create new opportunities. They have scaled their technology and data capture faster than their operational capacity to understand, contextualize, and act upon it.
And now, AI is accelerating signal production faster than the old linear models of dashboards and decision-making were designed to enable. Regardless of whether it’s in the energy transition, ESG mandates or healthcare modernization, the implications are far reaching.
The standard answer is technology: better tools, better pipelines, better models. But these are symptoms of a gap that more data, more dashboards, and more AI features have not closed. Organizations built data infrastructure and invested heavily in the roles that maintain it, without an equal investment in the capacity to integrate it.
The problem isn’t a shortage of data, it’s the abundance of it without the capacity to translate raw signal into meaningful decisions and outcomes.
This is the work of architecting an organization’s durable practice and structural capacity to orient, interpret, and act in changing conditions. It draws on service design, systems thinking, and modeling organizations as networks of people, data, and relationships, so the structure between them can be redesigned, not just the dashboards on top.
Try as we might for a single fix, responsiveness is a set of questions an organization has to keep answering. We’ve identified four we’re tracking here in this field brief. Each line of inquiry holds a hypothesis, examples across industries, and signals from the field.
Most modern technological transformations operate on a linear transmission myth: if you capture enough data, build a faster pipeline, and display it on a dashboard, the organization will become responsive. But the dashboard is a one-way transmission artifact with sender and receiver. When it doesn’t work, the default response is more signals, faster transmission, better compression, cleaner visualization.
Meaning is constructed across teams, not transmitted, and scaling the pipeline scales the noise, not necessarily the understanding. A data point is inert noise until it connects to a human who holds its context, can translate it to signal, and act on it.
When organizations overemphasize tech and data without the human component, they build lightning-fast data pipelines without the capacity to interpret that data. Instead of driving responsiveness, it creates information overload. AI is now repeating this pattern at scale.
Think of a recent decision that mattered. Did the data you’d already collected actually shape it — or did the decision happen alongside the data?
Where the gap shows up
Energy
A municipal utility managing decentralized grid assets and data logs millions of high-frequency meter signals as “information for record.” This information is maintained in internal IT systems, lacking robust routing to the frontline teams handling daily operational variance and delivering new dynamic energy services. The technology preserves the old structure while giving the appearance of transformation.
Philanthropy
A national foundation requires quarterly reporting from 200+ grantees. The data flows in, becomes board decks, and stops. Program officers, the people closest to what’s working, spend their hours chasing compliance rather than synthesizing valuable patterns across the portfolio. The fund has years of accumulated signal about what works and no practice for turning it into strategy.
Signals we're tracking
- How sensemaking distributes through AI-disrupted processes. If centralization was the answer for scale, what does distributed sensemaking look like, and how do you value and measure it?
- AI copilots deployed over fragmented databases, delivering polished text that strips the tacit context a real judgement call needs, while eroding the safety required for collaborative learning.
Data, once tucked away as a byproduct of enterprise software for technical teams, is now the primary medium through which services are constituted. The service is what the underlying systems do with information before, during, and after to create those touchpoints.
Data needs to be two things at once: a product, with ownership, accountability, and defined quality; and as dynamic service, orchestrated to reach the people and decisions that need it. Architectures like data mesh address the product layer at the technical level, and they leave the service layer completely blind.
The design conversation has stayed at surface UX. But in a digitized world, the service experience is constituted through backend information flows: eligibility is calculated by algorithms, queues are prioritized by data logic, and feedback loops are run automatically. Designing data structures is no longer backstage IT, it is an upstream, front-stage design capability that builds or erodes trust. A data model has become a strategic medium for what things mean, executed by systems.
If a customer or member asked why an automated decision affected them, could anyone in your organization trace the answer end to end?
Where the gap shows up
Public programs
A workforce program’s participant experience is constituted almost entirely by backstage data: eligibility determined by income verification across three agencies, services triggered by enrollment status, outcomes tracked in a state system that doesn’t talk to the program’s CRM. When a participant falls out of the program, no one can say where or why, because the journey exists in fragments across systems no one owns end to end.
Financial services
When a financial services company deploys an unmapped machine-learning model to evaluate credit risk, the system can quietly execute proxy discrimination or downrank profiles using invisible metadata layers. Because the data model was built without oversight, the executive team remains entirely unaware of their systemic exposure until regulatory or reputational harm hits scale.
What we're watching
- Legal rulings that treat automated AI outputs as a corporation’s own legally binding speech, and the responding accountability structures pulling data architecture into corporate governance.
- The shift from cosmetic UX or traditional journey mapping toward constructing data contexts and operational design directly within their architectures—making background data interpretation visible, ownable, and contestable before code is written.
Automating decisions carries false promise and new risk. Models are excellent at processing explicit, structured data patterns at scale, and weak at experience-based judgment, human relationships, and unwritten operational context that keep an organization safe. They risk furthering path-dependency on old data models, and on their own are not able to translate new dashboards they create into action.
Optimize purely for frictionless speed and you encounter the paradox of efficiency, stripping out essential unwritten context, creating automated blind spots, and driving customer or worker alienation. Rather than eliminating humans from the loop, responsiveness happens at the handoff. The design job is to let technology carry the administrative load so human judgement can step in where the model is limited.
Because of this lack of symmetry, to call this way of working “collaboration,” as some do, obscures accountability. A better frame is “co-performance:” a clear division of labor. (framing via Csertan, Touchpoint/SDN; “moral crumple zone” via Madeleine Clare Elish)
Where does a human nominally “review” automated output in your workflows? Under real conditions, would they catch an error?
Where the gap shows up
Operations / sensing
During an early-stage deployment of an environmental sensor solution, a background model routinely flags a “0” reading during late-evening cycles. Remote data engineers flag the data drop as a pipeline failure and expend hundreds of hours troubleshooting the code. A physical service design pass reveals that the room closes for a cleaning routine at that exact hour. The model recorded an accurate state but lacked the articulated boundaries to explain why the information was absent. The data point is noise until it is mapped to real-world human behavior.
Healthcare
A hospital clinical triage engine automatically routes patient flows. Nurses are given a nominal “override” option, but because they are under time pressure and have no visibility into how the model derived its recommendation, the override becomes symbolic. The nurse defaults to the algorithm’s choice and assumes liability when the system is wrong. Bolting automated routing or predictive analytics engines onto a strained operational floor exacerbates an existing “accountability gap” into a full operational risk.
What we're watching
- What is AI, deployed as the solution, conserving rather than disrupting? What needs disrupting?
- Teams using AI to process signals and communications from other teams, only to produce a response, with diminishing interpretation needed for robust collaboration.
- Where generative AI shifts cognitive loads and bottlenecks within an organization rather than removing them.
Traditional organizational development treats the company as a predictable machine, installing habits through training pipelines and compliance manuals. This approach breaks down in a post-generative environment where teams are building the plane as they fly it, and where software composes interfaces, routes resources, and makes background choices no human can supervise transaction-by-transaction.
The shift is from installing habits to designing the contexts that cultivate the social, physical, and technical practices where learning and sensemaking happen in the daily flow of work, and codifying design intentions and boundaries into parameters both people and machines can reliably act within.
When a decision goes wrong, do you have a practice for tracing why, or is the lesson only transferable as long as the people are there to tell it?
Where the gap shows up
Cross-team operations
A multinational operations team tracks customer activity data across an automated service flow. The machine flags an isolated 400% anomaly spike in product cancellations and automatically freezes account pipelines, treating it as a technical system glitch. A cross-functional analysis reveals the number was an accurate behavioral signal reflecting an unmapped external sequence: a key partner platform had updated its operational API, changing user workflows. The data point was structurally pristine, but its commercial meaning was invisible until a human mapped the relationship between the two systems.
Innovation programs
A fund stands up a measurement & evaluation function to “build a learning culture.” It operates as a compliance layer — rubrics, dashboards, annual reports — while the actual learning happens in hallway conversations between program officers that never get captured. The M&E infrastructure is technically sound and organizationally inert.
What we're watching
- Simple social rituals doing what taxonomies can’t, e.g. a “30-day decision autopsy” where teams trace a decision that went wrong, surface the unstated assumptions, and co-design heuristics for next time.
- The quiet collapse of off-the-side-of-the-desk “Centers of Excellence,” and the rise of embedded capability networks where central teams stop writing policy and start acting as curators, parameter architects, and network weavers.
Together, these lines of inquiry share a foundational thread: the gap emerging for organizations in this accelerating change will not be closed by the next platform, model, or dashboard. It requires an interpretive capacity that has to be deliberately built throughout the organization.
An organization that hasn’t built the capacity to interpret and act on the signals it receives doesn’t get to choose differently when it faces disruption. The option simply isn’t operationally available. Research on organizational learning calls this lockout (Cohen & Levinthal, absorptive capacity): by the time the new path is obviously valuable, the organization that didn’t invest in the adaptive capacity to understand it will struggle to take it. It foreclosed the option.
The work starts with legibility: making visible the network through which insight flows, and where it doesn’t. Who holds context. Where data travels and where it stalls. From there, the work maps the sensemaking relationships: how meaning is constructed, and where and when it breaks. Then it operationalizes: what skills, practices, and roles are needed to activate that meaning in context? It builds the bridge from data object to decision to value, drawing on data strategy, service design, and organizational change. What emerges is interpretive infrastructure: the capacity to turn the signals an organization gathers into a transformative asset.
We maintain this as an active log curating signals from practitioners, researchers, and rulings that are shaping how this dynamic is navigated and shaped.
-
Jun 2026
Three debts accumulate in AI-assisted teams: technical debt in the code, cognitive debt in shared understanding, intent debt in the missing rationale behind decisions. When AI generates without capturing reasoning, all three compound. The gap isn't technical, it's interpretive.
-
Jun 2026
“Can someone building without a designer in the room still produce something that's high quality and recognisably us? A component library can't answer that. It tells you what the button looks like. It can't tell you why we built it that way, when to break the rule, or what ‘good’ feels like for a flow nobody designed by hand.That judgment used to live in designers' heads. It has to live in the system now, in a form an agent can actually read… We call it Surge Intelligence: a maintained knowledge base of every component, pattern, token and design decision we have, written for AI to use… The point isn't the tooling. It's that the quality bar travels with the system instead of the person.”
-
Jun 2026
Enterprise identity standards now broker what an AI agent is allowed to do on a user's behalf, keeping the policy decision, consent, and audit trail with the organization's identity provider. Agentic systems force accountability to become explicit infrastructure rather than implicit practice, built simultaneously at the protocol level and the service-design level.
-
Jun 2026
If the interface is composed live, by an agent, in a moment you will never be in the room for, then your craft moves to two things. You build the environment the agent works inside… Then you feed in your design intentions… in a form a machine can absorb efficiently and accurately.
-
May 2026
Munich Regional Court rules Google's AI Overviews are Google's own speech, not search results. Once an AI system summarizes, makes claims, and speaks in a company's voice, legal accountability becomes part of the product architecture. The era of "it's just an algorithm" as a shield is ending.
-
Apr 2026
Organizations building human-in-the-loop responsibility checkpoints, model cards that trigger automatically when a system detects atypical parameters, forcing a pause before an automated output hardens into a decision. Intentional friction as a design feature, not a failure.
-
Apr 2026
Service teams adding visible intent layers and contextual operational design domains directly into blueprints, mapping system constraints, metadata lineage, and human override thresholds onto workflows. Making the backstage legible is becoming a design responsibility, not an afterthought.
The work behind these inquiries is a practice that starts by tracing the network: who holds context, where data moves, where it stalls. It maps the relationships where meaning gets constructed or lost. And it designs the structures, roles, and rhythms that let an organization act on what it’s making sense of.
We’re in ongoing conversations with leaders working this problem in energy, philanthropy, and AI adoption. If you recognized your organization somewhere in these pages, we’d like to talk.