Perspective / Strategy & transformation
The transformation operating model: who owns architecture, delivery and service performance?

A transformation operating model defines how an organisation makes, implements and reviews decisions about digital capability. Its value becomes visible when priorities conflict: a commercial deadline meets a reliability problem, an architectural constraint affects a customer promise, or a project ends while the service continues. Clear accountability matters more than the number of committees or job titles. This article develops an illustrative model connecting executive sponsorship, product ownership, technical authority and service operation, with evidence that can support practical decisions.
The executive view
- Assign authority to investment, product, architecture and service decisions rather than relying on broad job titles.
- Use architectural records and service objectives to connect evidence with practical action.
- Fund lifecycle responsibilities and review the model when dependencies or business exposure change.
Separate responsibilities without creating artificial silos
Begin with the decisions that must be made rather than an idealised organisation chart. Someone must own the investment case, order the work, assess technical consequences, accept operational obligations and resolve conflicts. Those responsibilities need to connect even when different teams or suppliers perform the work.
A smaller organisation may assign several responsibilities to one person. That can work if authority, competence and conflicts are understood. A larger organisation can still have weak accountability if every issue is passed between functions without an agreed decision-maker.
The current GovS 005 Digital standard provides a public-sector reference for lifecycle ownership. Commercial organisations can use its emphasis on clear responsibility as a design input without adopting every government role or control. The operating model should fit the business’s scale, criticality and actual capacity.
Distinguish responsibility for recommending a decision from authority to accept its consequences. An architect can explain technical exposure without controlling the commercial launch budget. A sales sponsor can set priorities without being authorised to bypass a security boundary. Record who must contribute evidence, who decides within delegated limits and where unresolved conflicts go. Include suppliers in this map: a contract that makes a vendor responsible for delivery does not remove the client's need to own product priorities, data use and acceptance. The operating model should remain intelligible when the supplier or individual role holder changes.
Give each role a decision and an evidence obligation
A role description should identify what the holder can decide, the constraints on that authority and the evidence needed. “Accountable for delivery” is incomplete if the person cannot resolve scope, funding or dependency conflicts.
| Responsibility | Typical decision | Evidence needed |
|---|---|---|
| Executive sponsor | Continue or change the investment | Benefit assumptions and exposure |
| Product owner | Order the next increments | User needs, value and constraints |
| Technical authority | Accept an architectural option | Decision record and trade-offs |
| Service owner | Accept operation of the capability | Readiness and service evidence |
| Delivery lead | Coordinate a workable plan | Dependencies, capacity and progress |
This is an illustrative allocation, not a mandatory staffing model. A business should adapt it and document delegated limits. The essential test is whether a material decision has one understandable route to authority, with contributions from the people who understand its commercial and technical consequences.
Delegation needs thresholds and an escalation path. A product team may change a reversible interface detail locally, while a new data processor, cross-service contract or material operating-cost increase needs wider review. Avoid thresholds based only on project budget; a small configuration change can have a large operational consequence. State the evidence required for each decision class and how quickly a reviewer should respond. When the organisation has no qualified decision-maker available, acknowledge the capacity gap rather than allowing an approval label to conceal that nobody has examined the technical or commercial implications.
Set authority explicitly for contractor and supplier personnel. A delivery lead may recommend a release while only the client's service owner can accept the ongoing support obligation. A specialist reviewer may provide assurance without holding the budget to fund remediation. Document these distinctions in the working agreement and contract handover, including escalation contacts and cover during absence. Avoid delegating accountability by implication because a supplier is technically capable. The organisation still needs a person able to reconcile commercial priorities, residual exposure and the funding required to operate the resulting service.
Make architectural governance proportionate and persistent
Architecture decisions should record the context, options, chosen approach and consequences. Focus attention on choices that are expensive to reverse or affect multiple services: identity, data authority, integration contracts, availability assumptions and supplier dependence.
An architecture decision record should remain useful after the meeting that approved it. Include the assumptions that could change the decision and the conditions for review. A temporary exception needs an owner and expiry condition; otherwise, it can quietly become permanent design policy.
Avoid requiring central approval for every implementation detail. Delegate decisions within agreed boundaries and escalate changes that cross them. For example, a team might choose a local library while a change to the authoritative customer identifier requires wider agreement. The objective is coherent change with visible reasoning, rather than a review process that delays work without improving the quality of decisions.
Consider a hypothetical portal whose customer-status lookup was approved on the assumption that the supplier permits sufficient request volume. After a product launch, peak traffic exceeds the tested limit and cached information becomes too old for credit decisions. That is a review trigger for the original architecture decision. The product owner assesses affected commitments; engineering compares capacity, caching and alternative integration options; the service owner defines temporary operating limits. Record the new evidence and whether it changes the decision. This gives the review a concrete purpose and avoids treating the original approval as permanent authority under materially different conditions.
Worked example: a commercial launch meets a reliability concern
Consider an illustrative customer platform preparing a new self-service journey. The commercial team wants a launch aligned with a campaign. Testing shows that a downstream account service degrades under the expected demand.
The delivery lead presents the evidence and practical options. Technical authority explains whether caching, demand control or a staged release changes correctness or support obligations. The service owner evaluates the operating position. The product owner assesses what capability remains valuable in a reduced first release. The executive sponsor resolves a conflict that exceeds delegated authority.
The decision record might approve a limited cohort with explicit monitoring and a reversal trigger, or postpone the campaign-dependent feature. Neither result is automatically correct. What matters is that the trade-off is owned and evidenced. A project status labelled green should not substitute for a decision about known operational exposure.
Make the options comparable in the worked launch example. A restricted cohort may reduce peak load but also affect the campaign's economics and customer eligibility. Caching may improve response time while creating freshness limits that must be explained. Delaying one feature may preserve the wider launch but require support training and changed marketing claims. Present these consequences together with the engineering evidence. Record the person accepting residual exposure and the monitoring or reversal condition. If no option fits the agreed risk tolerance, escalation should resolve the business commitment rather than simply asking the delivery team to work faster.
Connect service objectives to real prioritisation
A service objective should describe an aspect of performance that matters to its users and business commitments. It might concern successful completion, response time or the freshness of information used in a decision. Define how it is measured and what work takes priority when performance deteriorates.
Google’s SLO implementation guidance is an established reference for using reliability objectives in prioritisation. It does not mean every organisation needs a dedicated SRE department or should copy a numerical target from another service.
An objective has limited value if nobody can act on it. Agree who can reduce release scope, redirect capacity or escalate a dependency problem. Include the cost of collecting trustworthy evidence. A dashboard displaying an aspiration without an agreed response policy is a reporting mechanism; an operating model connects the evidence to authority and action.
Connect objectives to commercial promises with the correct granularity. A service-level target for successful ordering may matter more than a single availability percentage across unrelated endpoints. Review peak demand, critical customer cohorts and dependencies whose failure can prevent the user completing a task. Avoid treating an objective as an individual performance target: that can encourage exclusions and discourage honest reporting. The response policy should make it reasonable to surface a deteriorating service early. If teams expect punishment for disclosing uncertainty, formal ownership will not create dependable operational evidence.
Fund continuous ownership rather than an indefinite project
The service will continue to need change, support, supplier management and knowledge maintenance after its initial delivery. Make those obligations explicit in the investment model. Identify which costs support existing commitments and which fund new capability, while recognising that a single engineering task may contribute to both.
Keep a prioritised service backlog that includes business improvements, reliability work, technical debt and operational learning. Do not let only visible customer features compete for funding while essential maintenance accumulates without an owner.
Capacity planning should reflect actual dependencies. If the same specialists are assumed to deliver every new initiative and operate all existing services, the portfolio may contain more commitments than the organisation can support. Executives need a view of that contention before approving additional work. A clear operating model exposes this constraint instead of disguising it as a succession of isolated delivery delays.
Model retained ownership during procurement. Decide which capabilities must remain within the organisation to challenge proposals, accept work and operate the service across suppliers. That may include product ownership, architecture context, security decisions and access to delivery evidence. A supplier can provide specialist capacity while the client retains these accountabilities. Include transition time, documentation maintenance and knowledge transfer in the service budget. Funding only the initial build can force every later improvement through a new project approval, leaving no practical route to address routine reliability, dependency upgrades or lessons from customer use.
Worked economics: governance delay can exceed meeting cost
Consider an illustrative portfolio in which twelve cross-service decisions each month wait an additional five working days because the required reviewer is unclear. If each blocked decision leaves three people unable to progress for one hour per day, the direct interruption is 180 staff-hours per month: twelve × five × three. That is a deliberately narrow estimate; it excludes context switching, delayed revenue and the possibility that some work can proceed independently.
The response is not automatically to remove review. Classify the twelve decisions. Reversible local choices may be delegated. Decisions affecting shared customer identity, security or contractual service commitments may still need explicit authority. An agreed review rota and evidence template can reduce waiting without weakening the boundary. After introducing the change, measure actual decision age, avoidable rework and downstream incidents rather than simply counting fewer meetings.
The investment case should recognise when review prevents an expensive mistake. A fast decision based on missing evidence can create a longer delay later through migration or remediation. The useful objective is timely, informed commitment at the right level of authority. A credible operating-model review can show which decisions were accelerated, which were escalated for a good reason and whether teams retained a clear explanation of who accepted the consequences.
Review the model using outcomes and delivery evidence
Use a small set of measures to test whether the operating model works. Review decision latency, recurring dependency conflicts, unresolved exceptions and the stability of ownership. Pair those observations with business outcomes and service performance so that the model is not assessed only by how efficiently it holds meetings.
DORA’s April 2026 Quick Check update provides a contemporary reference for delivery throughput, stability and rework. These measures can reveal constraints in the change system. They should not be treated as a substitute for commercial benefit or a ranking of individual staff.
The practical output is a living decision system: an authority map, delegated limits, architectural records, service objectives, funding responsibilities and review triggers. Revisit it when the portfolio, organisational structure or risk profile changes. The best evidence of maturity is that the business can explain who decides, why they have that authority and what information allows the decision to improve.
Look for recurring decision patterns rather than only averages. A portfolio may show acceptable median decision time while cross-business disputes remain unresolved for weeks. Segment by decision class, dependency and impact. Review whether escalations repeatedly involve the same scarce person or unclear authority boundary. Combine quantitative observations with a small sample of decision records to see whether the evidence was adequate and the outcome understood. Measures should improve the operating system, not rank individual staff. Revisit the model after a major acquisition, supplier change or new product line because the earlier delegation boundaries may no longer match the business.
Questions this raises
Does an operating model require a large permanent team?
No. Several responsibilities can sit with one person or a mixed internal and external team. The important requirements are sufficient competence, clear authority, understood conflicts and continuity of ownership.
How does architecture governance avoid slowing delivery?
Delegate local decisions within agreed boundaries and focus wider review on cross-service or expensive-to-reverse choices. Short decision records make assumptions and consequences visible without requiring central approval of every implementation detail.
Who should own service performance after a project ends?
A named owner with authority to prioritise operational improvement and coordinate the relevant teams. The exact title can vary, but accountability, funding and the ability to act on performance evidence must continue beyond initial release.
Sources & further reading
- Government Functional Standard GovS 005: Digital 1 October 2026 (updated)
- Implementing SLOs 2018
- DORA Quick Check updates 22 April 2026
From thinking to doing