Digital Tools, Analytics and Artificial Intelligence

Introduction

Project initiation increasingly occurs in a digital environment. A project charter may be drafted in a cloud workspace, assumptions may be traced through a requirements repository, stakeholders may be identified from organizational data, and portfolio decisions may be supported by dashboards and predictive models. These capabilities can improve speed, transparency and evidence quality. They can also create new risks: unreliable data, automation bias, privacy breaches, opaque recommendations, cybersecurity incidents and unjustified confidence in forecasts.

Digital tools do not replace initiation judgment. They extend the project manager’s ability to assemble, analyse and communicate information. Their value depends on the quality of the information architecture, governance arrangements and human decisions surrounding them. The central managerial question is therefore not whether an organization uses artificial intelligence (AI), but whether its digital and AI-enabled systems produce sufficiently valid, explainable, secure and decision-relevant evidence for authorization.

This chapter examines the use of project information systems, collaboration platforms, dashboards, portfolio analytics and AI during project initiation and chartering. It emphasizes responsible use: defining the decision before selecting the tool, validating outputs, protecting information, identifying bias and preserving accountable human oversight.

Learning objectives

By the end of this chapter, the reader should be able to:

1. Explain how project information systems and collaboration platforms support initiation and chartering.
2. Design useful initiation dashboards and distinguish descriptive, diagnostic, predictive and prescriptive analytics.
3. Apply portfolio analytics to prioritization, capacity and strategic alignment decisions.
4. Evaluate AI applications in forecasting, document drafting, risk discovery and stakeholder analysis.
5. Validate digital and AI outputs using appropriate evidence, performance measures and expert review.
6. Identify risks involving bias, privacy, cybersecurity, intellectual property, explainability and accountability.
7. Establish practical human-oversight and governance controls for AI-assisted project initiation.

1. The digital initiation environment

Project initiation converts an opportunity or problem into an authorized, governable proposition. Typical outputs include a business case, project charter, preliminary scope, stakeholder register, high-level schedule, risk summary, benefits rationale and decision record. Digital systems support these outputs through five connected capabilities:

• Capture: collecting proposals, requirements, assumptions, estimates and stakeholder information.
• Structure: organizing information using common fields, taxonomies, identifiers and metadata.
• Collaborate: enabling controlled participation, review, commenting and approval.
• Analyse: generating indicators, forecasts, comparisons and scenarios.
• Assure: preserving version history, access controls, audit trails and evidence of decisions.

A project information system (PIS) is the broader set of people, processes, data, software and infrastructure used to collect, process, store and communicate project information. It may include portfolio management software, enterprise resource planning systems, document management, scheduling tools, risk registers and reporting platforms. A collaboration platform is a more specific digital environment for communication and co-production, such as shared workspaces, discussion channels, document repositories, whiteboards and workflow approvals.

The distinction matters. A chat channel may support rapid discussion, but it is not automatically an authoritative record. A dashboard may display a risk score, but it does not explain whether the underlying data are current. A generative AI assistant may draft a charter, but the draft is not an approved charter until accountable people verify and authorize it.

Digital thread and single source of truth

A digital thread is a connected chain linking information across the project life cycle. For initiation, it might connect a strategic objective to an investment proposal, assumptions, benefits, requirements, risks, decision rights and the approved charter. The thread improves traceability: a reviewer can ask why a project was proposed, which evidence supports it and how its objectives will later be measured.

A single source of truth does not necessarily mean one physical database. It means that the organization has identified authoritative records and rules for resolving conflicting versions. For example, the approved budget may reside in a portfolio system, while working estimates reside in a planning workspace. The charter should link to the approved record rather than reproduce uncontrolled copies.

2. Tools for initiation and chartering

Project information systems

A PIS should support, at minimum, proposal intake, project identification, document control, schedule and cost estimation, risk and issue logging, stakeholder information, approvals, reporting and archival. During initiation, its most important contribution is disciplined information capture. Standard fields make proposals comparable, while mandatory evidence fields reduce unsupported claims.

Useful metadata include the proposal owner, sponsoring unit, strategic objective, expected benefits, affected users, estimate confidence, data classification, decision date and approval status. Controlled vocabularies—for example, standard risk categories and strategic themes—improve portfolio-level analysis. However, excessive standardization can suppress novel ideas. A good design combines required common fields with space for narrative explanation.

Collaboration platforms

Collaboration platforms support workshops, asynchronous review and distributed decision-making. Their advantages include visibility of comments, rapid access to current documents and participation across locations. Their weaknesses include information overload, fragmented conversations, accidental disclosure and the tendency to confuse activity with progress.

Effective initiation practice establishes a collaboration protocol:

1. Identify the authoritative location for each document and decision.
2. Apply naming, versioning and retention conventions.
3. Distinguish informal discussion from formal approval.
4. Record decisions with date, participants, rationale and owner.
5. Restrict sensitive material according to its classification.
6. Convert important chat or meeting outcomes into controlled records.

Dashboards

A dashboard is useful when it supports a decision rather than merely displaying attractive numbers. An initiation dashboard might show proposal volume, decision cycle time, estimated value, strategic alignment, resource demand, risk exposure, data completeness and approval status.

Dashboard measures should have clear definitions. For example, decision cycle time may be defined as:

where T_d is decision cycle time, t_a is the authorization timestamp and t_s is the submission timestamp. The timestamps must use the same time zone and event definitions. If proposals are paused while awaiting information, the organization must decide whether paused time is included. Without such definitions, comparisons across units are misleading.

Dashboards commonly support four analytical levels:

A dashboard should display data freshness, source, owner, definitions and uncertainty where relevant. A forecast shown as a single precise number can create false confidence; ranges, scenarios and confidence or prediction intervals are usually more informative.

3. Analytics for portfolio initiation

Projects are initiated within a portfolio, not in isolation. Portfolio analytics helps decision-makers compare proposals and examine the consequences of approving one project rather than another. It should inform, not mechanistically determine, investment decisions.

Strategic scoring

A simple weighted scoring model can structure discussion. For proposal j, let:

where S_j is the total score for proposal j, w_i is the weight assigned to criterion i, r_ij is proposal j’s rating on criterion i, and n is the number of criteria. Usually, weights are non-negative and sum to 1, and ratings use a defined common scale such as 0 to 5. The model is applicable only when criteria are sufficiently defined, ratings are comparable and the organization accepts the additive trade-off assumption.

For example, suppose strategic fit, benefits and feasibility receive weights 0.40, 0.35 and 0.25. A proposal rated 4, 5 and 3 respectively has:

S = (0.40 × 4) + (0.35 × 5) + (0.25 × 3) = 1.60 + 1.75 + 0.75 = 4.10

The result is a decision aid, not an objective truth. A high score may conceal unacceptable regulatory exposure or a critical dependency. Criteria should therefore include gates—such as legal feasibility, safety and minimum data quality—that cannot be compensated for by other high scores.

Value, cost and capacity

Portfolio analysis may compare expected benefits with estimated investment. A simplified net value measure is:

where N is estimated net value, B is the monetized or otherwise consistently valued benefit and C is cost. The expression is meaningful only when benefits and costs use comparable boundaries, time periods and valuation assumptions. For multi-period investments, discounted cash-flow methods may be appropriate, but initiation estimates often contain substantial uncertainty and should not be presented with unjustified precision.

Capacity analytics maps proposed demand against available people, funding, technology and organizational attention. A portfolio can be financially affordable yet operationally infeasible if several proposals require the same scarce specialist. Scenario analysis can test alternative sequences and reveal bottlenecks before authorization.

Uncertainty and sensitivity

A decision is more robust when its recommendation remains stable under reasonable changes in assumptions. Sensitivity analysis varies one or more inputs, such as benefit, cost, probability of success or resource availability. If a small change in an uncertain assumption reverses the ranking of proposals, the decision should be escalated for judgment rather than treated as a definitive analytical result.

4. Artificial intelligence in initiation

AI refers here to computational systems that perform tasks associated with human cognitive activity, including classification, prediction, language generation, pattern recognition and recommendation. Machine learning systems infer patterns from data; generative AI systems produce new text, images, code or other content in response to prompts and context.

The distinction between automation and augmentation is important. Automation executes a defined task with limited human intervention. Augmentation supports human analysis while leaving responsibility for interpretation and action with people. Initiation usually benefits more from augmentation because objectives, values, political considerations and ethical judgments cannot be inferred reliably from historical data alone.

Forecasting

AI and statistical models can forecast proposal volumes, likely approval times, cost ranges, demand, resource requirements or schedule risk. A forecasting model should be assessed against a baseline method, such as expert judgment, a moving average or a simple trend model. Greater complexity is justified only when it improves decision-relevant performance and remains sufficiently understandable.

Forecast error should be measured on data not used to train the model. A common measure is mean absolute error:

where m is the number of evaluation observations, y_t is the observed value at time t and ŷ_t is the forecast. MAE has the same units as the outcome, making it easy to interpret. It is not suitable for every purpose; for example, percentage-based measures can behave poorly when actual values are zero or close to zero. Evaluation should also consider calibration, bias, performance across groups and performance under changed conditions.

Forecasts are conditional statements, not promises. The charter should identify major assumptions, forecast horizon, data date, uncertainty range and owner responsible for monitoring forecast validity.

Document drafting

Generative AI can create first drafts of problem statements, charter sections, meeting summaries, assumptions, risk prompts and stakeholder communication. It is especially useful for transforming approved source material into alternative formats or identifying missing fields in a template.

Safe drafting practice requires:

• providing only authorized and appropriately classified information;
• using an enterprise-approved system with suitable retention and access controls;
• requiring citations or links to source records where possible;
• checking names, dates, figures, commitments and legal language;
• marking the output as draft until reviewed;
• retaining the final human-approved version and its approval history.

Generative systems can produce fluent but false content, sometimes called hallucination. Fluency is not evidence. A draft that invents a sponsor, benefit or dependency can contaminate the charter and create contractual or governance consequences.

Risk discovery

AI can examine proposals, lessons learned, incident databases, requirements and external information to suggest potential risks. It may identify linguistic indicators of ambiguity, dependencies between workstreams or recurring failure patterns. Risk discovery is a prompting and prioritization aid, not a substitute for risk workshops or domain expertise.

Each suggested risk should be tested using a structured formulation: cause, uncertain event and effect. The team should then assess probability, impact, proximity, owner, response options and evidence. A model may be particularly weak at detecting novel risks, low-frequency catastrophic events or risks absent from historical data.

Stakeholder analysis

AI can classify documents by stakeholder groups, extract named entities, identify communication themes and suggest likely interests or influence relationships. Network analysis may reveal that a person who is not formally listed as a sponsor is central to information flow. However, inferring stakeholder attitudes, competence, protected characteristics or trustworthiness from personal data can be intrusive and discriminatory.

Stakeholder analysis should therefore use legitimate, relevant information and provide a route for correction. The project manager must not treat an algorithmic label such as “low influence” as a definitive social judgment. Stakeholder engagement is relational and contextual; it requires direct interaction.

5. Validation, bias and human oversight

A validation chain

Responsible use involves more than checking the final output. Validation should occur at several layers:

1. Purpose validity: Is the tool answering a question that matters to the initiation decision?
2. Data validity: Are data relevant, accurate, current, representative and lawfully obtained?
3. Model validity: Does the method perform adequately against a suitable baseline?
4. Output validity: Are results coherent, traceable and consistent with known facts?
5. Decision validity: Is the resulting action appropriate given uncertainty, values and constraints?

A useful control is independent challenge. The person who benefits from rapid approval should not be the sole person validating the model that recommends approval.

Bias and fairness

Bias can enter through historical decisions, missing populations, measurement choices, labels, proxies and unequal data quality. For example, a model trained on previously approved projects may learn to favour departments that historically received more investment. This can reproduce past inequality rather than identify future value.

Fairness is context-dependent. A team should define the relevant harm, affected groups and acceptable trade-offs before selecting a metric. Possible checks include comparing error rates, false-positive rates, calibration or selection rates across groups. No single fairness metric resolves all ethical questions, and some metrics conflict mathematically when base rates differ. Quantitative checks must therefore be combined with contextual and stakeholder review.

Human-in-the-loop governance

Human oversight should be designed, not assumed. A practical arrangement assigns:

• a business owner, accountable for the decision and value;
• a data owner, accountable for data quality, access and retention;
• a model or tool owner, accountable for configuration, testing and monitoring;
• a reviewer, responsible for challenging outputs;
• an authorized decision-maker, responsible for approval or rejection.

High-impact decisions should require human approval, an explanation appropriate to the audience, an audit record and a means of appeal or correction. Human review is ineffective if reviewers lack time, expertise or authority to disagree. Organizations should monitor automation bias—the tendency to accept machine recommendations simply because they are machine-generated.

6. Security, privacy and regulatory responsibility

Initiation materials may contain personal data, commercially sensitive estimates, intellectual property, security information and strategic plans. Controls should include data classification, least-privilege access, multifactor authentication, encryption, secure integration, supplier due diligence, logging, retention limits and incident response.

Before entering information into an AI service, the team should establish where data are processed, whether prompts are retained or used for training, who can access outputs, how deletion works and what contractual protections apply. Confidential information should not be placed in an unapproved public tool.

Privacy principles commonly relevant to initiation include purpose limitation, data minimization, accuracy, storage limitation, transparency and protection against unauthorized processing. Applicable law differs by jurisdiction and sector; organizations must consult their legal, privacy and information-security functions. Regulatory compliance is necessary but not sufficient: an activity may be lawful yet ethically inappropriate or unacceptable to stakeholders.

7. A practical operating model for AI-assisted chartering

An effective workflow is as follows:

Define the decision. State what must be decided, by whom, by when and according to which criteria.

Map information. Identify authoritative sources, data classifications, owners, gaps and dependencies.

Select the least complex adequate method. Use a rule, template or simple analysis when it is sufficient. Do not introduce AI merely because it is available.

Configure controls. Establish permissions, approved tools, prompts or model parameters, retention rules and review responsibilities.

Generate or analyse. Produce a draft, forecast or risk list while recording the data date, method and relevant assumptions.

Validate. Check facts against primary sources, compare with a baseline, test edge cases and seek domain review.

Decide and document. Record the human rationale, uncertainty, dissenting views, conditions and approval authority.

Monitor. Track forecast error, data drift, unexpected outcomes, user complaints and changes in context. Retire or revalidate tools when their assumptions no longer hold.

A lightweight AI use record may include tool name and version, purpose, data sources, classification, prompt or configuration, output, reviewer, corrections, decision impact and retention date. This record creates proportionate traceability without requiring every informal spelling correction to be treated as a high-risk model deployment.

Illustrative case: Northstar Hospital scheduling modernization

Background

Northstar Hospital, a fictional regional hospital, proposed a project to modernize operating-theatre scheduling. The expected benefits were reduced waiting time, improved utilization and better coordination among surgeons, nurses and facilities staff. The initiation team used a portfolio platform, a collaboration workspace and an AI assistant supplied by the organization’s approved technology vendor.

Problem

The initial proposal claimed a 20% utilization improvement and identified surgeons and the theatre manager as the principal stakeholders. Historical data were incomplete: cancellations were recorded inconsistently, emergency cases were coded differently across departments and staff availability was stored in separate systems. The AI assistant generated a polished charter and predicted that the project had a high probability of delivering its target.

Analysis

The project manager created a data inventory and discovered that the utilization measure had three competing definitions. A baseline forecast using the previous year’s monthly median performed nearly as well as the more complex model on a held-out period. The AI-generated stakeholder list omitted recovery-room staff because their role was not explicit in the proposal documents. A workshop revealed that they controlled a critical downstream constraint.

The team also tested the weighted portfolio score. Strategic fit, patient benefit and feasibility were weighted 0.35, 0.40 and 0.25. Ratings were 5, 4 and 2, producing:

S = (0.35 × 5) + (0.40 × 4) + (0.25 × 2) = 1.75 + 1.60 + 0.50 = 3.85

Although the score was attractive, the feasibility rating triggered a governance gate because data quality was below the organization’s minimum threshold.

Approach and solution

The sponsor authorized a short discovery phase rather than the full implementation project. The charter was revised to define utilization, establish data-quality acceptance criteria, include recovery-room staff and state that the 20% benefit was a hypothesis rather than a commitment. AI was used to identify missing charter fields and summarize workshop themes, but all factual claims were checked against source records. Sensitive staff data were excluded from the AI workflow.

The dashboard displayed data completeness, forecast range, unresolved dependencies and decision status rather than only the predicted benefit. The steering group required a human sign-off for any recommendation affecting staffing or patient scheduling.

Outcome

The discovery phase showed that a smaller scheduling intervention could produce measurable benefits without the proposed system integration. The portfolio board approved a limited pilot with explicit evaluation criteria and a review after three months. The project avoided premature commitment to an unreliable benefit estimate and identified an operational stakeholder who would otherwise have been missed.

Learning points

The case illustrates four principles. First, a sophisticated model cannot compensate for poorly defined measures. Second, a simpler baseline is valuable for testing whether complexity adds value. Third, stakeholder analysis requires social and operational inquiry, not only document mining. Fourth, a digital charter is stronger when it records uncertainty, data limitations and conditions of approval.

Challenges and limitations

Digital initiation systems can produce the appearance of control while hiding weak assumptions. Common limitations include:

• Data fragmentation: information is distributed across incompatible systems.
• Data latency: dashboards report yesterday’s position while decisions concern tomorrow.
• Metric gaming: teams alter classifications to improve reported performance.
• Model drift: relationships change as strategy, technology or behaviour changes.
• False precision: numerical outputs are interpreted as facts rather than estimates.
• Black-box dependence: users cannot challenge a recommendation they do not understand.
• Digital exclusion: stakeholders without access or confidence in the platform are underrepresented.
• Vendor dependence: proprietary tools may limit portability, auditability and continuity.
• Information overload: more data can slow rather than improve decisions.

Managers should treat these as governance and organizational-design issues, not merely software defects.

Contemporary developments and best practices

Current practice is moving toward integrated portfolio platforms, natural-language interfaces, retrieval-augmented generation, process mining, digital twins and responsible-AI management systems. Retrieval-augmented systems can ground generated text in selected organizational sources, but they still require access control, source-quality checks and human review. Process mining can reveal how proposals actually move through approval workflows, including rework and hidden delays. Digital twins may support scenario analysis, but their conclusions depend on model boundaries and data quality.

Best practices for postgraduate project managers include:

1. Begin with the decision and the information need, not the technology.
2. Define measures, owners, data dates and thresholds before building dashboards.
3. Prefer transparent baselines and explainable models when performance is comparable.
4. Separate proposal generation from authorization.
5. Use gates for legal, ethical, safety, privacy and data-quality conditions.
6. Validate AI outputs against primary evidence and independent expert judgment.
7. Record uncertainty, dissent and rejected alternatives.
8. Design stakeholder inclusion beyond the people who are easiest to digitize.
9. Apply least privilege and approved-tool policies to all sensitive material.
10. Monitor outcomes after initiation; responsible use is a life-cycle obligation.

Chapter Summary

Digital tools strengthen project initiation when they create traceable, current and decision-relevant information. Project information systems organize proposals, records, risks, stakeholders and approvals; collaboration platforms enable distributed work; dashboards convert data into monitored indicators; and portfolio analytics supports strategic alignment, capacity analysis and prioritization.

Weighted scoring, benefit–cost comparisons and sensitivity analysis can structure portfolio decisions, but they depend on explicit criteria, comparable ratings and acknowledged assumptions. AI can assist with forecasting, document drafting, risk discovery and stakeholder analysis. Its outputs are conditional and fallible: they may contain fabricated content, reproduce historical bias, omit novel risks or expose sensitive information.

Responsible AI-assisted chartering requires purpose, data, model, output and decision validation; security and privacy controls; explainability proportional to impact; documented accountability; and meaningful human authority to challenge or reject recommendations. The most effective project manager is not the person who automates the most work, but the person who uses digital capability to improve judgment, transparency and responsible authorization.

Key Terms

Project information system: People, processes, data and technologies used to manage project information.

Collaboration platform: A digital environment supporting communication, co-authoring, review and workflow.

Digital thread: Traceable connection among information objects across a project life cycle.

Single source of truth: An agreed authoritative record or rule for resolving conflicting information.

Dashboard: A visual display of measures designed to support monitoring or decisions.

Descriptive analytics: Analysis of what has happened or is currently happening.

Diagnostic analytics: Analysis of why an observed condition occurred.

Predictive analytics: Analysis estimating what may happen in the future.

Prescriptive analytics: Analysis recommending possible actions under stated assumptions.

Weighted scoring model: A structured evaluation combining criterion ratings and assigned weights.

Portfolio analytics: Analysis used to compare, balance and sequence projects collectively.

Artificial intelligence: Computational capability performing tasks associated with human cognitive activity.

Generative AI: AI that produces new content such as text, code or images.

Automation bias: Excessive reliance on a machine recommendation because it is automated.

Model drift: Decline in model relevance as data, relationships or context change.

Data minimization: Limiting collected and processed data to what is necessary.

Explainability: The ability to communicate how an output was produced or why it is relevant.

Human oversight: Deliberate human authority to review, interpret, challenge and decide.

Audit trail: A record of actions, changes, sources and approvals.

Review Questions

Short Answer Questions

1. What is the difference between a project information system and a collaboration platform?
2. Why should a dashboard display data freshness and measure definitions?
3. State two limitations of AI-generated charter content.
4. What is automation bias?
5. Why is a baseline model useful when evaluating an AI forecast?

Long Answer Questions

1. Explain how a digital thread can improve traceability during project initiation and chartering.
2. Discuss the assumptions and limitations of a weighted scoring model for portfolio selection.
3. Explain how AI can support risk discovery while remaining inadequate as a substitute for expert risk assessment.
4. Analyse the security, privacy and governance controls required when using generative AI with initiation data.
5. Develop a human-oversight operating model for an AI-assisted project authorization process.

Analytical / Application Questions

1. An organization’s AI tool recommends approving a project because its predicted benefit is high. The underlying data exclude cancelled projects and contain inconsistent benefit definitions. Diagnose the validity problems and propose a validation plan.
2. Construct a weighted scoring calculation for three project proposals using strategic fit, benefit and feasibility. Then perform a sensitivity analysis showing how a change in weights could alter the ranking.
3. Design an initiation dashboard for a portfolio board. Specify at least six measures, their definitions, data owners, update frequency and one control against misleading interpretation.

Multiple Choice Questions with Answers

Question 1

Which capability most directly creates traceability between a strategic objective, assumptions, risks and an approved charter?

A. A digital thread

B. A chat channel

C. A presentation template

D. An unstructured data lake

Correct Answer: A. A digital thread

Question 2

Which type of analytics addresses the question “What action could be taken under current capacity constraints?”

A. Descriptive analytics

B. Diagnostic analytics

C. Predictive analytics

D. Prescriptive analytics

Correct Answer: D. Prescriptive analytics

Question 3

Using weights 0.40, 0.35 and 0.25 with ratings 4, 5 and 3, what is the weighted score?

A. 3.25

B. 3.85

C. 4.10

D. 4.50

Correct Answer: C. 4.10

Question 4

What is a principal limitation of a weighted scoring model?

A. It cannot compare any proposals

B. It assumes criteria and ratings are sufficiently comparable

C. It automatically removes strategic judgment

D. It requires no assumptions

Correct Answer: B. It assumes criteria and ratings are sufficiently comparable

Question 5

Which measure is expressed in the same units as the forecasted outcome?

A. Mean absolute error

B. A project code

C. A portfolio weight

D. A classification label

Correct Answer: A. Mean absolute error

Question 6

Why should an AI-generated charter remain a draft until reviewed?

A. AI outputs are always too short

B. Generative AI can produce fluent but false content

C. Dashboards cannot display documents

D. Human approval is unnecessary for low-risk projects

Correct Answer: B. Generative AI can produce fluent but false content

Question 7

Which stakeholder-analysis risk is specifically associated with relying only on historical documents?

A. The system may omit operational stakeholders not mentioned in those documents

B. The system will always identify too many stakeholders

C. The system cannot classify any text

D. The system automatically obtains informed consent

Correct Answer: A. The system may omit operational stakeholders not mentioned in those documents

Question 8

Which control best reflects the principle of least privilege?

A. Giving every participant access to all proposal data

B. Allowing access only to information necessary for an authorized role

C. Removing all audit logs

D. Publishing sensitive estimates in a public workspace

Correct Answer: B. Allowing access only to information necessary for an authorized role

Question 9

What does model drift mean?

A. A deliberate change in project scope

B. A decline in model relevance as conditions or data change

C. A duplicate version of a charter

D. A delay in a meeting

Correct Answer: B. A decline in model relevance as conditions or data change

Question 10

Which arrangement provides meaningful human oversight?

A. Accepting every AI recommendation automatically

B. Allowing reviewers to challenge and reject outputs before authorization

C. Hiding the tool’s use from stakeholders

D. Using a more complex model without validation

Correct Answer: B. Allowing reviewers to challenge and reject outputs before authorization