Mind the Delivery Gap

Two teams in the same organisation, solving the same problem, completely unaware of each other. They commission separate work, reach divergent conclusions, implement incompatible solutions. Six months later a third team starts from scratch.

I have watched that happen more times than I care to count, and the thing worth noticing is that it is never a knowledge management failure. The organisation almost certainly has a SharePoint site, a lessons-learned repository and a community of practice that convenes quarterly. None of it helps, because knowledge does not travel through systems. It travels through trust, and trust is a human problem that resists technical remediation.

The previous article I wrote argued that AI money leaks at the join between what an institution intends and what it actually buys, and that closing that join is a demand capability rather than a platform. This one is about why the capability keeps failing to find a home, which turns out to be a much older story than AI.

Forty years, four rewrites

The Business Relationship Manager has been rewritten three times in four decades, and each rewrite began the same way. A technology arrived that the operating model had no route for. Value started leaking at the join. Somebody had to stand in the gap while the institution caught up, and that person kept getting a new job title while the work underneath barely changed.

The translator, 1980s to 1990s. Computing moved out of the machine room and into the business. Departments acquired systems they did not understand and IT acquired stakeholders it had never had to speak to. The mandate was narrow and almost entirely reactive: capture the requirement, translate it, pass it down the line, report back. Success meant the request arrived intact. Nobody asked whether the request was the right one, because asking that was outside the role as written. This is the era that still shapes how many senior leaders picture the job, and it is the source of a good deal of present-day frustration.

The relationship manager, 2000s to 2010s. Service management industrialised IT into processes and service levels, and the relationship stopped being a translation problem and became a satisfaction problem. The mandate widened, and it acquired a flaw that took a decade to surface. Measuring the relationship by satisfaction made the incentive to keep the stakeholder comfortable, and comfort is not the same thing as value. A well-managed relationship can sit on top of a badly-directed portfolio for years without anyone noticing.

The demand shaper, 2015 to 2025. This rewrite was forced by procurement rather than by technology. Once cloud made capability purchasable on a corporate card, the gatekeeper model collapsed. A department that wanted a system no longer had to ask; it could buy, deploy, and involve IT only when something broke or an auditor asked a question. That removed the leverage and replaced it with something better, which was influence earned upstream. If you cannot control what gets bought, the only useful position is the one occupied before the buying decision is made. This is the decade in which the discipline became genuinely strategic, and also the decade in which it was most consistently misunderstood, because a demand shaper who succeeds prevents work rather than delivering it, and prevention is invisible on a delivery dashboard.

The fourth rewrite is happening now, in real time, and most organisations are treating it as a staffing question rather than a structural one.

The numbers are a verdict on operating models

Gartner's public position is that at least half of generative AI projects were abandoned after proof of concept by the end of 2025, and that more than forty per cent of agentic AI projects will be cancelled by the end of 2027. The stated causes are worth reading closely: escalating costs, unclear business value, inadequate risk controls. Model capability does not appear on the list. The same analysis estimated that of the thousands of vendors claiming agentic capability, only around 130 were building anything that deserved the label.

Read that list again as a practitioner rather than as a technologist. Unclear business value is a demand shaping failure. Inadequate risk controls is a governance and decision rights failure. Escalating costs without a defined outcome is a value realisation failure. Every one of them is a relationship and alignment problem wearing a technical costume.

The public sector version has a sharper edge. When an AI initiative fails quietly in a department, the money is public, the accountability sits with an accounting officer, and the failure eventually becomes a select committee question rather than a lessons-learned slide. We do not get to write off a portfolio and move on.

And here is the part I find most revealing. The market has responded to this by inventing new roles, and those roles look extremely familiar. The AI Value Orchestrator centralises demand intake, prioritises by business value, sets measurement frameworks upfront and governs investment. The AI Product Manager shapes demand around real business pain, manages adoption and justifies value continuously. Strip the AI prefix from either job description and you are reading competencies as written in the BRM Body of Knowledge a decade ago.

Two conclusions follow, pointing in opposite directions. The first is validating: when an industry independently reinvents a discipline under a new name, the discipline was correct about the problem. The second should worry the profession. Organisations are reaching for new titles because they do not recognise that they already employ people who do this. That is a positioning failure, and it belongs to us rather than to them.

Three symptoms, one habit

This is where the separate complaints I have been making for a year turn out to be the same complaint. I want to set them side by side, because individually each looks like an administrative irritation and together they look like a structural diagnosis.

The framework maps the wrong neighbourhood. In August 2025 the Government Digital and Data Profession Capability Framework was updated. The Business Relationship Manager role received a refreshed description, new level descriptors and two skills removed, which is good evidence of a living document. The role page also carries a shared skills table listing the roles that share BRM capabilities: IT Service Manager, Application Operations Engineer, Incident Manager, Command and Control Centre Manager, End User Computing Engineer. Every one sits within the IT operations role group. The shared skills themselves are strategic thinking, user focus and stakeholder relationship management, which are not niche IT operations competencies. They are the skills that define effective Product Managers, Service Designers and User Researchers working elsewhere in the same framework. The table maps the neighbourhood. It does not map the city, and the most valuable collaboration a BRM has, with the people who hold user evidence, service design authority and product strategy, sits entirely outside it.

The perimeter was drawn before the roles existed. AI Solutions Architect. Prompt engineer. MLOps specialist. AI product owner. Assurance and ethics lead. None of them appear in the supplier contracts, the RACIs, or the forums most organisations still run on. Consider where they actually sit: outside a governance perimeter that nobody chose, because it was drawn years earlier around contracts, change cadences, decision rights and service levels designed for a slower world. So when an AI Solutions Architect makes a decision today, it may pass through no formal governance route at all. Nobody is dodging scrutiny. There was never a route to dodge.

The posting cycle is shorter than the trust. The traditional civil service generalist served the state well for decades: mobile, adaptable, intellectually quick, able to pick up a brief across widely different policy domains. Those are not small virtues, but they were cultivated for a world where problems stayed in their lanes. AI governance, economic inactivity, health inequality, climate adaptation: not one of these belongs to a single directorate. They live in the white space between disciplines, departments and communities of practice that rarely share a vocabulary. A generalist can navigate that space. Navigation is not enough. What these problems need is someone who can convene it, and convening rests on legitimacy in several worlds at once. That legitimacy cannot be manufactured through rotation. It takes time, sustained presence, and a kind of relational investment that a two-year posting does not permit.

Three separate complaints. One habit underneath all of them.

The institution draws a boundary at a moment of its own convenience, and then treats the boundary as a description of the work. The framework's shared skills table was a reasonable classification in 2020 and became a map of where BRM used to sit. The governance perimeter was a sensible enclosure around a stable supplier estate and became a fence with the new capability on the wrong side of it. The posting cycle was designed to build breadth and became a mechanism that resets relationships just as they start paying. In every case the boundary was drawn in good faith, for good reasons, at a moment that has since passed. And in every case nobody noticed, because a boundary that has stopped describing the work does not announce itself. It just quietly stops routing things.

Nobody owns the map

Which raises a question I think is the practical heart of this, and I have not seen it asked directly.

Both the capability framework and the governance perimeter are maps. Neither is anybody's deliverable. Nobody in either structure is assessed on whether the map still corresponds to the territory, and no review cycle exists whose specific purpose is to ask. Frameworks get updated when someone has capacity. Perimeters get redrawn after an incident. Neither mechanism is capable of noticing a boundary that has aged badly but not yet failed loudly.

That is a fixable governance gap, and it is cheaper to fix than almost anything else in this essay. Give each map an owner and a review cadence, and make the review question explicit: which roles now do work that this boundary does not route, and what changed since we last looked? That is a half-day exercise once a year. Set against the cost of a cancelled AI programme, it is free.

Why the habit is more expensive now

Consider a realistic scenario. Someone shapes the investment case for an AI-powered triage tool in a public-facing service. The value plan is sound. The business case is approved. But without a shared evidence base with user research, the model trains on assumptions about user behaviour that a discovery phase would have surfaced and challenged. The AI does not fail loudly. It performs exactly as designed, against the wrong understanding of the user. By the time that surfaces in complaints data or a service assessment, the investment is in production and the cost of remediation dwarfs the cost of the original gap.

This is the general case, not the exception. A defective process, automated, is a faster defective process. A poorly understood workflow furnished with an AI tool produces erroneous outputs with greater velocity. The disorder does not dissolve; it compounds, and it becomes harder to perceive because the speed of execution generates an illusion of competence.

The cross-government AI governance infrastructure, including the Algorithmic Transparency Recording Standard, the AI Safety Institute's guidance and the Civil Service AI Playbook, all assume that the professionals feeding investment decisions have access to shared, validated evidence. They are built on the premise that strategy and user reality are already connected. A BRM and a User Researcher operating from separate framework architectures, with no formal signal that their roles are designed to interlock, is a gap in that premise. The governance tools exist. The professional infrastructure to feed them consistently does not.

McKinsey's finding is the scale of it: while 92 per cent of organisations are augmenting their AI investments, only 1 per cent believe their digital transformations have reached genuine maturity. Government is not insulated from that gap. It may be more exposed to it than most. The technology is arriving. The human infrastructure to make it work is not.

What the work actually is

That human infrastructure has a definition worth being precise about. The BRM Body of Knowledge defines a relationship not as a process or a platform but as the state of connectedness between people that determines how they interact and behave in pursuit of shared purpose. Which means a connection problem cannot be solved with a system. It can only be solved by a person whom multiple teams genuinely trust.

The learning theorists Etienne and Beverly Wenger-Trayner gave the senior version of this a name in 2021, drawing on interviews with forty practitioners across six continents working on exactly these intractable challenges. They call it systems convening: a leadership practice that takes a landscape view, sees the whole ecosystem of knowledge, institutions and communities bearing on a problem, and holds enough legitimacy in several of those worlds to bring their inhabitants into genuine dialogue. A systems convener does not stand above the landscape. They are embedded within it, trusted by its inhabitants, and they create conditions for conversations that would not otherwise take place rather than facilitating conversations that already exist.

The BRMBoK supplies the operational counterpart. Demand shaping is influencing demand towards higher-value outcomes rather than fulfilling the requests that arrive, which in practice means helping partners articulate what they actually need rather than processing what they asked for. Value harvesting is the sustained work of ensuring that what was built actually produces the outcome it was meant to produce, long after go-live. Neither is a procurement gate. Both require presence over time.

And that phrase is where two of the three symptoms collide, in a way I think deserves stating flatly.

The two-year posting is not only a workforce policy. It is a value policy. Value harvesting requires presence over time. Systems convening requires legitimacy that only accrues with tenure. If the rotation cycle is shorter than the accrual period, then the institution has chosen, without ever deciding to, a staffing model that makes its own value model unreachable. Every officer in that cycle is doing their job properly. The outcome is still guaranteed.

Why the fourth rewrite needs a measurement change, not just a role change

There is a second reason organisations reach for a new job title instead of the role they already employ, and it is not only positioning.

Prevention is invisible on a delivery dashboard. A demand shaper who succeeds stops a bad investment before it becomes a programme, and nothing appears anywhere to show it. A person with a prototype produces a demo. In any budget conversation between the two, the demo wins, every time, and the institution learns to fund demos.

So the fourth rewrite will not land on role design alone. Something has to make prevention legible. The mechanics are not complicated: count the requests that did not become projects and record why, log the duplicate initiatives found and stopped, track the cases where an existing capability was reused instead of rebuilt, and put the cost avoided next to the cost spent in the same paper for the same committee. None of that requires new tooling. It requires deciding that avoided cost is a reportable outcome. Until that decision is made, this discipline will keep being reinvented under new names by organisations that cannot see the value of the version they already have.

Three repairs, at three altitudes

The habit shows up at three levels, so the repair does too. None of these requires a reorganisation.

At the level of the profession, publish a cross-role interaction matrix. This is the existing shared skills logic applied across the whole framework rather than only within role groups. Map where a BRM's strategic thinking meets a Product Manager's, where user focus connects to a User Researcher's practice, where stakeholder relationship management aligns with a Service Designer's stakeholder mapping, where value realisation connects to the Delivery Manager and Service Owner roles already defined. The effect is practical rather than theoretical. Career conversations get a richer map, communities of practice get a framework-backed rationale for joint working, and collaboration that currently depends on personal initiative becomes something the framework actively points people toward. This is not a criticism of the framework authors. Classifying BRM within IT operations in 2020 was a reasonable starting point when the profession was less mature. The framework has been updated since. The classification has not kept pace.

At the level of the operating model, build the route in. The instinctive response to new roles arriving outside the perimeter is to treat it as a governance problem: new people, no route, tighten the perimeter. The opposite argument is stronger. Their arrival is the clearest available signal that value is being created in new places and the operating model has not kept up. The connector mindset was built exactly for this. You do not own the contract, the data or the risk, but you know who does and you can put them in the same room. Take the AI Solutions Architect: brilliant at the art of the possible, moving at a speed our models were never designed for. A gatekeeper slows that person down without making anything safer. A partner who knows the terrain makes the possible land. The architect brings pace, new capability and a working prototype; the other side brings the map, contract fluency, the routes to a decision and clarity about who owns the risk. Separately, the architect ships pilots that stall at the governance boundary and the BRM curates a pipeline of yesterday's requests. Together they deliver capability the organisation can actually run, safely funded and supported, at pace. The mature question in a multi-supplier world is no longer what does the business want. It is how does this ecosystem absorb new capability safely and at pace. Every organisation needs someone whose job that is. Very few have named them.

At the level of career structure, let tenure and alignment do their work. Align the role to value streams, outcomes, journeys or missions rather than pairing one practitioner to one executive, because person-pairing reinforces exactly the silos that fragment AI demand into three uncoordinated pilots, and the duplication lives across the org chart rather than within it. And where convening is the job, protect the tenure that convening requires. That is a decision about which posts are exceptions to the rotation norm, not a wholesale change to how the civil service develops people.

The fourth rewrite, in outline

What the next form of the role looks like is already visible.

It is distributed rather than held: a competency embedded across enterprise architects, service owners, product managers and delivery leads, each shaping demand in their own domain, with dedicated practitioners orchestrating across them. The single point of focus stays. The assumption that it means a single person does not.

It runs on outcomes, obstacles and options rather than requirements. What is the outcome, what is genuinely blocking it, what are the credible routes through. A department that opens by asking for an agent to automate a process is usually describing a symptom, and the real constraint is often a sign-off chain rather than a processing speed.

It is modally aware. The same practitioner needs three registers: broker when the job is operational stability, account executive when the job is growing an existing service, incubator when the job is co-creating something that does not yet exist. Applying the wrong register is one of the more common ways a good practitioner loses a room.

And in government specifically it deepens into joint governance: formal steering with mission units, shared definitions of success, technical people embedded inside operational areas, with the role orchestrating rather than reporting.

Aaron Barnes, co-founder of the BRM Institute, has a name for the posture underneath all of that: anticipatory value orchestration. Someone operating at full capability is not waiting for a team to present an AI investment proposal. They are already building the picture. Which teams face pressures AI might address. Which of those have the relational maturity and data provenance to receive a capability without it failing quietly. Which capability built in one part of the organisation could, with the right connection, serve a cognate need elsewhere. They are orchestrating across institutional boundaries before the formal investment cycle starts, which is to say before governance calcifies around a decision nobody consciously made. That is not a capability invented for the AI era. It is an existing capability made urgent by it.

The through line

The reason this discipline keeps getting rewritten is that its actual subject was never technology. It is the recurring gap between what an organisation can now do and what its operating model will let it do. That gap opens every time capability moves faster than governance, which is to say continuously, and it is where the value leaks out.

I do not expect it to settle. New roles will keep arriving and operating models will keep lagging, because a prototype takes a weekend and a contract variation takes a quarter. No amount of redrawing changes that. A permanent gap needs a standing role to bridge it, and a standing role needs standing: the accumulated trust that lets someone stop a room rather than merely object in it. Which is the point at which this essay meets its companion. The authority to pause is the scarce good in an AI-enabled organisation, and the three habits described here are, between them, a fairly efficient machine for destroying it.

Every other role in the profession has a primary orientation. The Product Manager is oriented towards delivery, the User Researcher towards evidence, the Service Designer towards experience. All indispensable. None of them constituted to hold the space between institutional strategy and human readiness.

Government is about to invest very substantially in artificial intelligence. Whether that investment changes something worth changing will not be determined by the technology. It will be determined by the human infrastructure around it, and by whether the institution can bring itself to name the work rather than redraw the boundary around it for a fifth time.

Forty years in, the job title has changed four times. The work has not. The only open question is what we call it next, and whether the people already doing it are in the room when the naming happens.

References

• Government Digital and Data Profession Capability Framework, updated August 2025, including the Business Relationship Manager role page and its shared skills table; and the 2025 to 2030 roadmap commitment to safe, responsible AI adoption.

• BRM Institute, Business Relationship Management Body of Knowledge, on demand shaping, value harvesting, strategic partnering and the definition of a relationship; and the BRMP and CBRM certifications accredited through APMG International, covering the domains Evolve Culture, Build Partnerships, Drive Value and Satisfy Purpose.

• Aaron Barnes, co-founder of the BRM Institute, on anticipatory value orchestration.

• Etienne and Beverly Wenger-Trayner (2021), on systems convening, drawing on interviews with forty practitioners across six continents.

• Gartner, on generative AI proof-of-concept abandonment, the cancellation of agentic AI projects by 2027, and the estimate of genuine agentic vendors.

• McKinsey, on the proportion of organisations augmenting AI investment against those reporting genuine digital maturity.

• Algorithmic Transparency Recording Standard; AI Safety Institute guidance; Civil Service AI Playbook.

• Scopism, SIAM Body of Knowledge; ITIL 4 change enablement; ISO/IEC 42001 AI management systems.

This essay develops and combines arguments first published across the GOVBRM Newsletter on LinkedIn, in the editions on the capability framework's shared skills table, the human infrastructure problem, systems conveners, new roles needing a route in, and the four rewrites of the Business Relationship Manager.

About the author

Sebastian Moore is a Business Relationship Manager working in the UK public sector, at the point where government ambition meets delivery. He was named in Apolitical's Government AI 100 and recognised as a Global Top BRM by the BRM Institute, and he co-founded the Civil Service Care-Experienced Network.

These are personal practitioner views, written in a personal capacity, and not a statement on behalf of any employer