Field Notes from Quality & Delivery Transformations

Blog

  • Field Note: What Belongs in a Product Reference Model?

    If quality is assessed, something must provide the basis for judgement. We need to know what the product is intended to achieve, who it is intended to serve and what “good enough” means in its context.

    Requirements provide part of that reference. But they rarely provide the whole of it. Requirements tend to describe selected behaviour or changes that engineering should implement. User stories, acceptance criteria and specifications help translate product decisions into development work.

    A product reference model would need to preserve a wider perspective. It should help answer not only:

    Did we build what was specified?

    But also:

    Are we building the right product, with the right qualities, for the right purpose?

    More than requirements

    The model would probably begin with the product’s purpose. Why does the product exist? What problem should it solve? What business or customer outcome is expected?

    It would also need to identify the customers, users and other stakeholders whose needs and constraints matter.

    From there, it should describe the capabilities the product must provide and the quality characteristics that are important in its context. For one product, performance and scalability may be decisive. For another, usability, security, auditability, accessibility, data quality or ease of integration may matter more.

    The model should also include what is uncertain. Product development is based on assumptions:

    • that a customer need exists;
    • that a proposed capability will address it;
    • that users will adopt the solution;
    • that the product can operate under expected conditions;
    • that the investment will create value.

    These assumptions are part of the product understanding and should not disappear when work enters the backlog.

    A possible structure

    A product reference model might therefore contain:

    • business intent and strategic context;
    • customers, users and stakeholders;
    • intended outcomes and value;
    • required product capabilities;
    • relevant quality characteristics;
    • constraints and obligations;
    • assumptions, risks and uncertainties;
    • dependencies and relationships with the wider product or ecosystem;
    • operational conditions;
    • evidence required to judge whether expectations have been met;
    • learning from actual use.

    This should not necessarily become one large document. Much of it may already exist across roadmaps, business cases, customer research, architecture, requirements, quality scenarios and operational objectives.

    The problem is often not that the information is completely absent. It is that it is fragmented, implicit or disconnected.

    A living model

    The product reference model cannot remain fixed. Business priorities change. New customer groups are targeted. Products scale. Regulation develops. Operational evidence challenges earlier assumptions.

    The model must therefore evolve as the product and its context evolve. It should guide development before release and support judgement after release.

    That makes it more than an assessment artefact. It becomes a shared model for creating, evaluating and learning about the product.

    We do not yet know exactly what form such a model should take. But the need is becoming clearer:

    Requirements describe parts of what should be built. A product reference model preserves the wider understanding needed to build, assess and evolve the right product.

    The product reference model is not another source alongside the existing sources. It is the structure that makes those sources function as one coherent model of the product.

  • Field Note: What Is the Business Value of Quality?

    The business case for quality is often reduced to defects. Better quality means fewer defects. Fewer defects mean less rework, fewer incidents and lower support costs.

    That is true, but incomplete. A product can contain few defects and still solve the wrong problem. It can function correctly while being difficult to use, expensive to operate or impossible to scale. It can be delivered quickly and reliably while creating little customer or business value. If quality is broader than testing and defect prevention, its business value must also be broader.

    DORA measures an important part of the story

    DORA metrics help organisations understand how effectively engineering delivers software changes. They provide evidence about speed, stability, recovery and rework within the delivery system. This matters. An organisation that cannot change its product quickly and safely will struggle to respond to customers, risks and opportunities.

    But delivery performance is not the same as product quality. DORA can show how well we deliver changes. It does not tell us whether we selected the right changes, whether the product is good enough for its context or whether it achieves the intended outcome.

    The broader question is:

    How effectively can we turn business intent into a valuable, fit-for-purpose product—and continue improving it sustainably?

    Quality reduces the cost of being wrong

    When product intent, stakeholder needs and assumptions remain unclear, engineering may deliver exactly what was requested without creating meaningful value. The result may be:

    • rarely used functionality;
    • solutions to misunderstood problems;
    • unnecessary complexity;
    • rework caused by late clarification;
    • delayed learning about customers.

    A broader quality approach makes product reasoning explicit and challenges assumptions earlier. Its value is not only fewer implementation defects. It is less wasted product investment.

    Quality reduces the cost of failure and operation

    Some weaknesses become incidents, security exposure, data errors or customer complaints. Others become permanent operating costs:

    • recurring support contacts;
    • manual workarounds;
    • unstable services;
    • repeated defect correction;
    • difficult recovery;
    • complicated maintenance.

    These costs are often distributed across engineering, operations, support and the business. They may never appear together as the cost of poor product quality. A lifecycle perspective connects them to the product decisions that created them. Its value includes lower risk, lower cost to serve and fewer expensive surprises.

    Quality protects the ability to change

    Quality also affects how easily the product can be improved tomorrow. Weak maintainability, poor testability, unclear product intent and accumulated technical debt make every future change slower and riskier. The organisation then spends more effort preserving existing behaviour and less effort creating new value.

    Quality engineering protects the ability to change through:

    • shorter lead times;
    • safer releases;
    • less rework;
    • lower development cost;
    • less dependence on individuals;
    • greater responsiveness to markets and customers.

    This is where DORA becomes part of the wider quality story. It helps show whether engineering can change the product quickly and sustainably.

    Quality increases the likelihood of useful outcomes

    Quality is not only about reducing cost and risk. A reliable, understandable, responsive and context-appropriate product is more likely to achieve:

    • adoption;
    • customer satisfaction;
    • trust;
    • retention;
    • productivity improvements;
    • business benefits;
    • access to more demanding markets.

    The causal chain is rarely simple, but it can be made visible:

    better product understanding

    → better engineering decisions

    → stronger product behaviour

    → better customer experience

    → improved business outcomes

    The aim is not to claim perfect attribution. It is to make the relationship explicit enough to support better decisions.

    Quality also has a cost

    A broader quality approach requires investment in:

    • understanding stakeholders and outcomes;
    • defining relevant quality expectations;
    • improving testability and observability;
    • collecting engineering and operational evidence;
    • interpreting that evidence;
    • involving specialists;
    • improving products and ways of working.

    There is also a risk of bureaucracy. If quality modelling becomes a documentation exercise, it can slow delivery without improving judgement. The depth of assessment should therefore reflect the product’s value, risk, complexity and cost of failure. The goal is not maximum quality activity. It is sufficient quality understanding to make responsible business decisions.

    A broader business case

    The business value of quality can be understood through four effects:

    • reducing the cost of being wrong;
    • reducing the cost of failure and operation;
    • reducing the cost of future change;
    • increasing the likelihood of valuable outcomes.

    The cost lies in the modelling, measurement, assessment and improvement needed to achieve those effects.

    This should not be compressed into one universal quality ROI figure.

    Each product needs an explicit quality argument:

    Which business risks and outcomes matter?

    Which product and engineering qualities influence them?

    What evidence shows whether they are improving?

    Is the expected value worth the required investment?

    DORA provides an important view of engineering delivery.

    A broader product-quality model connects that view to customer value, operational cost, business risk and product outcomes. That may be the real business case for quality:

    Quality reduces the cost of being wrong, the cost of failure and the cost of future change—while increasing the likelihood that the product achieves its intended purpose.

  • Field Note: An Assessment Is a Snapshot. Improvement Requires Reassessment.

    I have usually thought of an assessment as a one-time intervention. We define a reference model, collect evidence, judge the current state and identify improvements. The result is a snapshot. That is useful for diagnosis.

    But it is not enough when the purpose is improvement. Once change begins, the subject of the assessment starts to move. Practices change. Responsibilities shift. New evidence appears. Some improvements work, others do not. New weaknesses become visible. The original assessment gradually becomes a description of a state that no longer exists.

    An assessment may be a snapshot. Assessment should be a recurring capability.

    The first assessment establishes a baseline

    A thorough initial assessment remains valuable. It clarifies:

    • what is being assessed;
    • what good looks like;
    • the current state;
    • important strengths, weaknesses and risks;
    • priorities for improvement.

    Without that baseline, improvement can become a collection of disconnected actions. But the baseline is only the beginning.

    Improvement creates new evidence

    Every improvement action is also a hypothesis. We expect that changing a process, role, tool, structure or behaviour will lead to a better result.

    That expectation should be tested. Did the change actually happen? Did behaviour improve, or only documentation? Did the intended result appear? Did the improvement create new problems elsewhere?

    Without reassessment, these questions are often answered through impressions or completion reports.

    But completing an improvement is not the same as achieving improvement. Reassessment brings the organisation back to evidence.

    Reassessment does not mean repeating everything

    Continuous assessment does not require a full formal assessment every sprint or every month. That would quickly become bureaucratic. Reassessment can happen at different depths. A lightweight review may examine recent evidence and a few critical indicators. A periodic deeper assessment may revisit the complete reference model. A major change, failure or shift in context may trigger an earlier reassessment.

    The principle is simple:

    Reassess often enough to learn whether improvement is actually happening.

    The reference model may also change

    Reassessment is not merely updating a score against a fixed model. As understanding improves, the definition of good may also evolve. Some expectations may prove irrelevant. New capabilities may become important. The organisation may discover that its original target state was incomplete or misguided.

    Reassessment should therefore ask two questions:

    Are we moving closer to what we defined as good?

    And:

    Is our definition of good still the right one?

    From report to improvement loop

    A static assessment often ends with findings and recommendations.

    A continuous assessment process closes the loop:

    reference model

    → evidence

    → judgement

    → decision

    → improvement

    → new evidence

    → reassessment

    The assessment should therefore identify not only what needs to improve, but also:

    • what evidence will show progress;
    • when reassessment will occur;
    • what should trigger an earlier review;
    • which assumptions remain uncertain.

    Assessment then becomes part of the improvement system rather than an event before improvement begins.

    The real purpose of reassessment

    The purpose is not to prove that the plan was followed. It is to discover whether the plan was right. Reassessment helps the organisation detect ineffective improvements early, adapt priorities, expose unintended consequences and remain focused on outcomes rather than activity.

    An assessment tells us where we are. Reassessment tells us whether we are actually moving—and whether we are still moving in the right direction.

  • Field Note: Quality Is the Integrity of the Chain

    We usually discuss software quality somewhere in the middle of the product lifecycle. Requirements have been defined. Work has entered the backlog. Engineering has started. Testing, Quality Assurance and Quality Engineering help ensure that the software is built and delivered well.

    But the product’s quality story began earlier. It began when the organisation formed an intention: solve a problem, serve a customer, enter a market, reduce a cost or create a business outcome.

    And the story does not end when the product is released. It ends—if it ends at all—when we can understand whether the product produces the outcome for which it was created.

    This suggests a broader definition:

    Quality is the integrity of the chain from business intent to business outcome.

    The chain

    The chain may be expressed simply:

    Business intent

    → product decisions

    → engineering delivery

    → operational reality

    → business outcome

    Each link translates something into something else.

    Business ambition is translated into a product direction. Product direction is translated into capabilities, priorities and quality expectations. These are translated into engineering decisions and working software. The software enters an operational environment where customers, users, integrations, data and real workloads affect its behaviour. That behaviour contributes—or fails to contribute—to the intended outcome. Quality depends on the integrity of all these translations.

    How the chain breaks

    A product may fail because the original intent was unclear. It may fail because the organisation misunderstood the customer need or selected the wrong route to its ambition. It may fail because the right intent was translated into the wrong product. It may fail because engineering realised the product poorly. It may function as designed but prove unreliable, difficult to use, expensive to operate or impossible to scale. It may operate successfully and still fail to produce adoption, efficiency, customer value or commercial results.

    These failures look different, but they share something important:

    Somewhere between intent and outcome, the chain lost integrity.

    This expands the meaning of a quality failure. A defect is a quality failure. But so is building a technically excellent capability that customers do not need. So is improving performance when the real obstacle to enterprise adoption is permissions, auditability or integration. So is meeting every acceptance criterion while failing to create the intended outcome.

    Different disciplines protect different parts

    Testing, Quality Control, Quality Assurance and Quality Engineering remain essential. But they do not protect the complete chain. Testing guards against not knowing how the product behaves. Quality Control guards against accepting outputs that do not conform to defined expectations. Quality Assurance guards against relying on processes and controls that cannot consistently produce dependable results. Quality Engineering guards against treating quality as something that can be inspected into the product at the end. Each protects an important part of the chain.

    Product-quality assessment asks a more complete question:

    Does the available evidence support the judgement that the product is right, good enough for its intended context and capable of producing its intended outcome?

    It therefore guards against false confidence. Tests may pass. Requirements may be met. Delivery metrics may be healthy. The system may operate as designed. The organisation may still be wrong about the product.

    The role of Product Ownership

    Someone must preserve continuity of intent as the product moves through the chain. That does not mean one person owns every quality decision. But Product Ownership has a central responsibility for maintaining coherence between:

    • the business ambition;
    • customer and stakeholder needs;
    • product decisions;
    • quality expectations;
    • engineering work;
    • operational evidence;
    • product learning.

    Product Ownership should not merely feed work into engineering. It should help ensure that what emerges from engineering can still be traced to the purpose that justified the investment—and that evidence from operation changes future product decisions.

    The role of governance

    Product-quality assessment creates a basis for judgement. Governance determines what happens next. Leadership must decide:

    • which outcomes and risks matter;
    • what evidence is sufficient;
    • which trade-offs are acceptable;
    • who may accept remaining risk;
    • whether to release, invest, redirect or stop.

    This is why the complete view of quality cannot be delegated to testing or contained within delivery. It reaches into product strategy, investment and accountability.

    The economic meaning of integrity

    When the chain lacks integrity, the organisation pays in several ways. It pays the cost of being wrong when it invests in the wrong capability, assumption or market response. It pays the cost of failure through defects, incidents, disruption and lost trust. It pays the cost of operation through support demand, workarounds and recurring interventions. It pays the cost of future change when complexity, weak maintainability and unclear intent make every next improvement slower and more expensive.

    When the chain retains its integrity, the organisation increases the likelihood that engineering effort becomes product value and that product value becomes business outcome. That may be the most complete business definition of quality:

    Quality is the integrity of the chain from business intent to business outcome.

    Everything else helps us understand where that chain can break, what evidence reveals its condition and how the organisation should respond.

  • Field Note: Product Quality Can Only Be Fully Understood in Operation

    Product quality must be assessed before release. Testing, reviews and demonstrations help us decide whether the product is ready.

    But that judgement is always provisional. Before release, we assess the product under selected conditions. After release, reality adds evidence we could not fully create in advance.

    Pre-release evidence is necessary, but incomplete

    Testing can show whether expected behaviour is present, whether important risks have been explored and whether quality characteristics such as performance, security or resilience meet defined expectations.

    But testing always uses models. We select scenarios, create representative data, simulate load and define expected outcomes.

    All of this is essential. But it cannot reproduce the full complexity of real use. Actual users behave differently from test users. Real data contains unexpected patterns. Dependencies fail in unfamiliar combinations. Support demand reveals confusion that no acceptance test exposed. A product may pass every planned assessment and still teach us something important after release.

    Operation produces a different kind of evidence

    Once the product is in use, new evidence appears:

    • incidents and escaped defects;
    • helpdesk calls and customer complaints;
    • performance under real load;
    • usage and abandonment patterns;
    • manual workarounds;
    • operational effort;
    • customer and business outcomes.

    This evidence does not merely tell us whether engineering made mistakes. It tells us how the product behaves in its actual context. A repeated support question may reveal weak usability. A production incident may expose an architectural assumption that was never explicit. Low adoption may show that a correctly implemented feature does not solve the intended problem. High operational effort may reveal maintainability or supportability problems. Reality makes the product more understandable.

    The team’s responsibility continues after release

    In many agile and DevOps environments, the development team does not hand the product over and disappear. The same people who design, build and test the system also help operate, support, maintain and improve it. That creates a direct quality-learning loop. Incidents reveal weaknesses in design, testing or operational preparation. Difficult fixes expose maintainability problems. Repeated manual interventions show where operability or automation is insufficient. Production behaviour tests assumptions made during engineering.

    This evidence should matter deeply to the team. It is not merely reporting about a product they once delivered. It is feedback on the quality of their own engineering decisions. The team that creates the product should also learn from how that product behaves in reality. Done may end the implementation cycle. It should not end the team’s interest in quality.

    Monitoring is not yet assessment

    Production systems generate dashboards, logs, alerts and incident records. But observations do not interpret themselves. A green dashboard does not prove that the product is good. A metric may remain within its threshold while users still struggle.

    Operational evidence must be interpreted against:

    • intended outcomes;
    • stakeholder needs;
    • quality expectations;
    • acceptable risks;
    • operating context.

    Monitoring produces evidence. Assessment turns that evidence into a quality judgement.

    The reference model must survive release

    A product-quality reference model cannot be used only before delivery. It must remain relevant in operation. The expectations defined earlier should help us interpret what happens in reality.

    At the same time, operational evidence should be able to change the reference model. An incident may reveal a missing quality characteristic. Customer behavior may challenge an assumption. A support pattern may expose a need that was never understood. A change in scale may make yesterday’s acceptable quality level inadequate today.

    The reference model must therefore be living. It provides the basis for assessment, but it also evolves through assessment.

    Quality learning follows the lifecycle

    The Definition of Done marks an important engineering boundary. Release marks another. Neither marks the end of quality. Before release, we ask whether the available evidence gives us enough confidence to proceed. After release, we ask what reality reveals about the product in actual use. These are different stages of the same quality-learning process.

    Product quality must be assessed before release, but it can only be fully understood in operation. A mature team prepares the product carefully before release—and remains willing to learn what reality reveals afterwards.