Loretta Osei-Bonsu had forty-eight hours. The Illinois Department of Employment Security's new AI-assisted claims triage system was six weeks from launch when a state senator's office called. They had heard about the project from a union steward, not from her team. The union was threatening to file an unfair labor practice charge. And somewhere in a FOIA (Freedom of Information Act) request stack, a journalist was about to get hold of an early project memo that used the phrase "reduce headcount." Loretta, the department's Chief Information Officer, had the technology mostly right. She had stakeholder communication almost entirely wrong.

Why communication is the project

A government AI initiative lives or dies in the space between what the system does and what stakeholders believe it does. You can deploy a model that reduces claims processing time by 40 percent and still lose the project to a press story about a single erroneous denial. Public-sector leaders often treat communication as a finishing step, the press release after the contract is signed, and that instinct will cost you the initiative. In government, communication is not downstream of the work. It is the work, and the people who did not hear about it from you will hear about it from someone who wants it to fail.

Federal AI initiatives rarely fail because the model is wrong. They fail because stakeholders were not aligned when the crisis arrived. The Department of Defense Project Maven contract with Google was technically sound and operationally valuable, and it collapsed in 2018 because the company had not engaged its own engineering workforce as a stakeholder and had not helped defense public affairs prepare the narrative. The IRS Direct File pilot succeeded in 2024 in part because Treasury, OMB and the IRS Taxpayer Advocate were aligned on scope, phased expectations and talking points months before launch. The difference was not the quality of the technology.

Effective federal stakeholder communication serves five purposes that compound over the life of a program. It builds shared understanding of scope, so that OMB, the CIO Council and the agency AI governance board describe the initiative the same way in writing. It surfaces statutory risks early, including Privacy Act System of Records Notice requirements, Paperwork Reduction Act triggers and Section 508 accessibility obligations, so they enter the schedule rather than arriving as launch-blocking surprises. It develops internal advocates, particularly among career senior executives whose continuity across administrations is what actually sustains multi-year AI programs.

The remaining two purposes matter most when things go wrong. Communication manages expectations, so that neither a Cabinet Secretary nor a subcommittee chair believes the AI system will do something it cannot do, which is the failure mode that destroyed trust in early predictive-policing deployments at state and local level. And it builds durable accountability by creating a written record that the agency engaged the right parties at the right time with the right information, which is exactly what GAO, an Inspector General and a congressional committee will look for in any post-incident review. Note what that record is and is not: it evidences that you engaged, not that anyone agreed.

Every communication a federal AI leader authors must survive three tests. The FOIA test asks whether you would be comfortable if this email appeared on the front page of a national newspaper. The IG test asks whether the agency Inspector General would find this communication accurate, timely and within your authority. The oversight test asks whether a House Oversight or Senate HSGAC staffer reviewing the record would conclude that the agency acted responsibly. Those three tests are the working discipline of the rest of this lesson, and a document that fails any one of them should not leave your outbox regardless of how well it reads.

Stakeholders who hold statutory power

Stakeholders are the people and institutions with formal or informal interest in or influence over your AI initiative, and in government service they arrive with authority that private-sector stakeholders rarely possess. The OMB desk officer assigned to your agency under Circular A-11 has de facto authority to slow your budget cycle. The agency Inspector General, operating under the Inspector General Act of 1978, has subpoena authority over your records. The House and Senate committees of jurisdiction can compel testimony, affect apportionments through the appropriations process, and direct GAO to audit your program under 31 USC 712.

Two of these actors deserve particular attention because programs underestimate them most often. The union local recognized under chapter 71 of title 5 has a statutory right to bargain over changes in working conditions, and OPM and the Federal Labor Relations Authority have interpreted that to include AI deployment which materially changes how work is performed. The Senior Agency Official for Privacy, under OMB M-17-12, must approve a Privacy Impact Assessment before a rights-impacting AI system enters production. Treat either of those as a post-launch courtesy notification rather than a pre-launch stakeholder and your initiative will be stopped, slowed or publicly embarrassed.

Why does this matter more in government than in industry? Multiple audiences must be served simultaneously rather than sequentially. Competing interests must be balanced without violating the Antideficiency Act, the Hatch Act or the Anti-Lobbying Act. Every communication is potentially FOIA-releasable and discoverable. Skepticism is high because many stakeholders lived through failed modernization programs such as FBI Sentinel, USDA MIDAS, Air Force ECSS and DoD DIMHRS. And risk aversion is structurally encoded in Federal Acquisition Regulation Part 39, the FISMA authorization process and the Privacy Act. Those are not bugs. They are features of a republic that funds its government through annual appropriation and holds it accountable through Inspectors General and legislative oversight. Design your stakeholder plan for that environment rather than against it.

Mapping your stakeholder universe

Government AI projects touch more stakeholders than most private-sector deployments. Map them into four rings before you write a line of code, and use named roles rather than generic archetypes, because a plan that says "engage leadership" produces nothing while a plan that names the Deputy Secretary's staff secretary produces a meeting. The interest-by-influence matrix is the standard tool; what makes the government version different is that several of the boxes are filled by statute rather than by your judgment about who cares.

Ring 1: authorizers and overseers

• Legislators and their staff. They control appropriations and can hold hearings. Brief them before they read about the project in a trade publication. At federal level this means the committees of jurisdiction, including House Oversight, Senate HSGAC and the relevant appropriations subcommittees.
• Oversight bodies. Inspector General offices, peer IGs through CIGIE, government accountability offices and audit committees. They are not opponents. Pull them in early; they will find out anyway and prefer not to be surprised.
• Budget offices. Your OMB (Office of Management and Budget) counterpart approves spending and will need to explain cost projections to leadership.
• Agency leadership. The agency head, the Deputy Secretary or Chief Operating Officer, the Chief AI Officer role established by Executive Order 14110, and the Performance Improvement Officer under the GPRA Modernization Act. Their interest is mission, political exposure and statutory compliance; their cadence is a quarterly executive briefing plus an annual strategic review.

Ring 2: affected workers and their representatives

• Unions and labor councils. Many government AI projects that change how employees work trigger formal duty to bargain obligations, with bargaining rights under 5 USC 7106 and 7114 attaching before any material change in working conditions. Recognized federal unions commonly include AFGE, NTEU, NFFE and IFPTE. Know your collective bargaining agreements first. Loretta's union contract required 90-day written notification before implementing any system that changed workflow for covered employees. Her team had missed that clause.
• Frontline staff and program owners. Operational owners at the component or bureau level see what the system actually does. Their interest is throughput, quality and job impact, and their cadence is monthly operational review plus ad-hoc incident notification. Skip them and you lose your best source of ground-truth feedback and your credibility with the union at the same time.

Ring 3: external advocates and the public

• Equity and advocacy groups. Civil rights, disability access, privacy and economic justice organizations, including groups such as EPIC and the ACLU, will scrutinize automated decisions. Engage them in the design phase, not the crisis phase.
• The general public and the press. Public comment periods are often required for rule changes that affect benefits, and a 30-day period that receives 2,000 submissions is a documented record that you asked. Public channels also include the Federal Register, agency public affairs, the public AI use-case inventory and congressional constituent services.
• State, local and regulated partners. Where your system touches their operations, they are stakeholders with their own oversight and their own press.

Ring 4: internal partners

• Legal, privacy and compliance. General Counsel, the Senior Agency Official for Privacy, the Equal Employment Opportunity office and the ethics office under 5 CFR 2635. Their interest is Privacy Act compliance, civil rights compliance under Title VI and Section 504, and alignment with the applicable OMB minimum practices. Their cadence is pre-decisional consultation before every major design choice, not review after the design is fixed.
• IT leadership. The agency CIO under Clinger-Cohen and FITARA, the CISO under FISMA and the Chief Data Officer established under the Evidence Act. Their interest is authorization readiness, FedRAMP boundary, data provenance and continuous monitoring; their cadence is a standing architecture review board and a quarterly risk reporting cycle.
• Communications and public affairs. Bring them in at month one, not month eleven, and coordinate with them before any press release or public inventory posting.

Tailoring messages by audience

The same fact means different things to different listeners. A claims accuracy rate of 94 percent sounds excellent to a technologist and alarming to a legislator focused on the 6 percent. Message tailoring is not spin; it is translating one underlying use case into the register each audience needs in order to do their own job. The registers below are distinct enough that you should write them separately rather than producing one deck and varying the emphasis, which is how a document ends up serving nobody.

Managing expectations across a multi-year initiative

Illinois's claims triage system ran across three fiscal years and two gubernatorial administrations. That span is not unusual for government AI work, and it is why expectation management is a discipline rather than a tone of voice. The discipline is stating in writing what the system will do and what it will not do, on what schedule, at what confidence level, with what human oversight, and at what cost. The largest failure mode in public-sector AI is implicit over-promise. The largest success factor is explicit phased commitment with a named decision point at the end of each phase.

A credible phased statement has a shape. Phase one establishes the foundations: data governance, the Privacy Impact Assessment, the authorization package, the union bargaining and the training curriculum. Phase two runs a pilot within a single bureau or office with full human review of every output, measured against a documented baseline. Phase three expands to production with risk-tiered human review and continuous monitoring, and with explicit stop criteria tied to fairness, accuracy and privacy metrics. Each phase ends at a pre-committed decision point where the AI governance board reviews the evidence and decides to proceed, revise or stop, and that decision is minuted and available to the IG and GAO on request.

The single most useful sentence in a phased commitment is a NOT-DOING statement. "This system will not make final eligibility determinations in phase two" is worth more than three paragraphs of aspiration, because it is falsifiable and because it survives the departure of everyone who wrote it. Over-promising is what killed public confidence in programs such as IRS CADE 2 and what shaped the launch-readiness failures at Healthcare.gov, and in both cases the written record of what had been promised was the instrument of accountability afterwards. Write the limits down while they are still limits and not excuses.

Build a communication calendar at the project's start. Loretta's team eventually adopted quarterly stakeholder briefings: a two-page written summary to all Ring 1 and Ring 2 stakeholders, followed by optional thirty-minute calls. Summaries covered project status, budget to date, milestone changes and risks on the horizon, and every summary went through legal review because every summary was a potential FOIA document. Set specific, bounded commitments. "We will complete disparity testing by March 15 and share results with the Equity Advisory Panel by March 22" is trackable. "We are committed to equity" is not.

When timelines shift, and they will, communicate the change before stakeholders discover it independently. A six-week delay announced proactively reads as responsible management; the same delay discovered by a journalist reads as concealment. Be careful about what the cadence itself buys you, though. A briefing calendar produces a record that you informed people on a schedule. It does not produce agreement, and a stakeholder who was fully briefed and still objects is not evidence that the communication failed. Sometimes it is evidence that the objection is real and that the program has a decision to make rather than a message to refine.

Executive briefing techniques

An agency head or elected official gets roughly ninety seconds of genuine attention before forming their response, and a formal briefing slot is typically fifteen minutes. Both numbers are true at once, which is why structure matters more than content density. The first ninety seconds decide whether the remaining thirteen and a half minutes are a conversation or a formality. Structure every briefing so the decision is available immediately and the supporting material is available if wanted.

1. Lead with the decision or the ask. "We need your approval to proceed to Phase 2 by July 1." Not background. Not history.
2. State the one risk they need to know. "The union has raised concerns about the notification timeline. Here is our response and what we need from your office."
3. Give them the headline. "If this goes well, 340,000 Illinoisans will receive claims decisions in four days instead of seventeen."
4. Put detail in a one-pager. Attach it. Do not read it aloud.
5. End with a clear next step and owner. "We need a response by Thursday. Maria on your staff has the full briefing document."

The fifteen-minute version, engineered for a Deputy Secretary, Under Secretary, agency COO or component head, allocates its time deliberately. Minutes zero to two cover strategic alignment: which objective in the agency strategic plan this advances, which entry it is in the AI use-case inventory, and how it is risk-classified under applicable OMB policy. Minutes two to five cover progress against the phased plan: milestones completed, milestones missed and why, on or off schedule against the approved baseline. Minutes five to eight cover risk: the top three residual risks from the AI risk register, the mitigations in place, and the items requiring a decision from this executive.

Minutes eight to eleven cover value: measured benefit to date, forecast benefit through end of phase, and indirect benefits to workforce and public trust. Minutes eleven to thirteen cover resources: the budget line, personnel authorization, data-sharing agreement and policy clearance needed to proceed. Minutes thirteen to fifteen are the ask, stated as questions the executive can answer. Do you approve proceeding to phase two, do you direct us to accelerate or slow, and do you commit the resources listed. A good briefing answers the executive's three unspoken questions before they are asked: is this legal, is this working, and is this worth the political and fiscal cost.

The mechanics around the briefing matter as much as the fifteen minutes. A one-page read-ahead circulates 48 hours in advance through the front office, a single decision memorandum with named signatories accompanies the briefing, and minutes are filed to the governance record within 5 business days. For committee hearings, prepare three answer lengths for every key question: one sentence, thirty seconds, and two minutes. Elected officials follow their own interests rather than your outline, and the ability to answer at the length actually asked for is what separates a credible witness from a defensive one.

The annual cadence

A twelve-month cadence synchronizes internal and external review so that no audience is ever learning something for the first time in public. Weekly, the program team holds an operational review. Monthly, the AI governance board reviews risk, metrics and incidents. Quarterly, the Chief AI Officer or equivalent briefs the agency head, the CIO Council representative and the budget office. Semi-annually, the program reports to GAO, the IG and the congressional committees of jurisdiction through the regular budget, oversight and inventory channels rather than through ad-hoc contact.

Annually, the program participates in the AI use-case inventory refresh and the strategic plan update under GPRA Modernization. Union consultation runs on the schedule negotiated in the collective bargaining agreement, which is typically monthly for active projects with material workforce impact. Public communication runs through the agency public affairs office and is coordinated for any newsworthy milestone. The value of writing this down as a calendar rather than a policy is that a calendar has dates, and a date that passes without the meeting happening is visible to everyone, including the auditor who eventually asks for the minutes.

What goes wrong when communication fails

Government AI projects fail publicly in recognizable patterns, and the recognizability is the useful part: each of these has a specific, cheap preventive step that somebody chose not to take. Read the list as a set of preventable events rather than as cautionary tales, and check your own project against each one before you finish the section.

• The union surprise. A labor organization learns about workflow changes from a staff member rather than from management. They file a charge, the project freezes during arbitration, and the delay costs $800,000 in carrying costs. Fix: labor counsel reviews the project plan at month one and confirms what the collective bargaining agreement requires.
• The legislative ambush. A committee chair who was never briefed calls a hearing. The agency head testifies with incomplete information. The IG opens a review. Fix: quarterly briefings to key legislative offices, including the quarters when there is nothing dramatic to report.
• The FOIA document crisis. An internal email saying "this will basically replace the case managers" is released under a records request and becomes a headline. Fix: write every email as though a journalist will read it, because one might.
• The equity story. A news outlet shows that the AI denies claims at higher rates for claimants from majority-Black zip codes. The agency had the data and had not published it. Fix: publish disparity analyses proactively, with context and a remediation plan, before someone else publishes them without either.
• The staff morale collapse. Frontline employees who were never consulted start giving negative interviews to local reporters. They know the system's failure modes better than the project team does. Fix: a formal staff feedback channel during development, with documented responses to what they raise.

FOIA implications of written communications

The Freedom of Information Act and its state equivalents mean virtually every written communication by a government agency is a potential public record. Treat this as a discipline rather than a threat. When a team writes as though every document will eventually be public, internal reasoning improves, embarrassing disclosures become rarer, and the agency builds a record that holds up in an oversight hearing. The discipline also has legal edges worth knowing: the Hatch Act restricts political advocacy by career staff, the Antideficiency Act restricts implied commitments of future funding, and the Anti-Lobbying Act restricts grassroots campaigns aimed at pressuring Congress.

Four practices reduce FOIA exposure without reducing transparency. Set a records retention policy at project launch, so you know what gets kept, for how long and where. Have legal counsel review any document describing the system's intended effect on employment or benefits eligibility before it circulates internally. Keep project communications on agency systems, because when decisions migrate to personal email or messaging apps the agency loses the audit trail and gains a records problem. And proactively release key documents, including contracts, disparity analyses and governance frameworks, on the agency website, which reduces reactive FOIA volume and demonstrates good faith at the same time.

Anti-Patterns

• Over-promising in writing. A program director commits on paper to benefits the model cannot deliver at its current maturity, usually to secure funding or political cover. The commitment outlives the person who made it and becomes the standard the program is judged against. Counter it with explicit NOT-DOING statements, phased commitments tied to measured baselines, and confidence levels stated alongside every projected benefit.
• Explaining the model but not the work. The program communicates thoroughly about the technology and barely at all about how jobs change, leaving the frontline workforce ambushed by a redesign nobody described to them. Pair every technical rollout with an explicit account of the work redesign, a competency framework update, and a training plan sponsored at the human capital level rather than the project level.
• Treating concerns as obstacles. Legitimate questions from unions, privacy advocates or the public get handled as resistance to be managed rather than as inputs to be evaluated. Document every concern, the response, and the owner, and publish the log where appropriate. Then watch for the second-order failure below.
• Letting the engagement log become the engagement. A concern register that records every objection and every response is a genuine control, and it is also easy to turn into a paper exercise where logging replaces deciding. A concern that has been documented, responded to and never acted on is not a resolved concern; it is a documented one. Review the log for items where the response was acknowledgment rather than change, and treat that pattern as the finding.
• Preaching to believers. The program engages friendly audiences with enthusiasm and neglects skeptics on appropriations staff, authorizing staff or in the IG's office, because those meetings are unpleasant. Schedule outreach to skeptical audiences with the same rigor and the same calendar discipline as the friendly ones, and go before you need something.
• FOIA blindness. Internal communications are drafted assuming a privacy that FOIA does not grant, and the assumption holds right up until a records request lands. Write every email, slide and memo to the standard of the newspaper front page, and route anything describing employment or eligibility effects through counsel first.
• Mistaking silence for consent. A stakeholder who received the quarterly summary and did not respond gets recorded as aligned, and the program proceeds on that basis until the objection arrives at the worst possible moment. Briefing someone establishes that they were informed, not that they agree. Where alignment actually matters to the plan, ask for it explicitly and record the answer, including when the answer is no.
• Treating the communication plan as the trust. A well-run cadence, a tailored message set and a complete stakeholder map improve your odds substantially and guarantee nothing. Trust is built by what the system does to people and by whether the agency tells the truth when it goes wrong. A communication plan that is used to manage perception of a badly performing system will fail faster than no plan at all, because it adds a documented record of what you said while it was happening.

Practice Prompts

• Build the map with names. Take a live AI initiative and construct the stakeholder map using named roles rather than offices: who specifically is the budget desk officer, the IG contact, the union local president, the privacy official, the committee staffer. Place each on an interest-by-influence matrix. The boxes you cannot fill with a name are your first week of work.
• Write the same thing nine ways. Take one use case and draft the opening paragraph for each register in the tailoring table. Then hand each draft to someone who actually occupies that role and ask what question it leaves unanswered. Most programs discover their compliance and workforce registers are the weakest.
• Draft the NOT-DOING statement. For your initiative, write three sentences describing precisely what the system will not do in its current phase, in language a subcommittee staffer could quote. Then check whether anything you have already said publicly contradicts them, and fix the contradiction now rather than at the hearing.
• Build the fifteen-minute briefing. Prepare an executive briefing to the minute allocation in this lesson, with a one-page read-ahead and a decision memorandum naming signatories. Rehearse it against the three unspoken questions: is this legal, is this working, is this worth the cost. Time yourself; the section that overruns is the section you are least sure of.
• Run the three tests on real documents. Pull five recent internal emails or slides from your project and apply the FOIA test, the IG test and the oversight test to each. Note which ones you would not want released and why. That "why" is usually the thing the program has not yet decided to say out loud.
• Build the twelve-month calendar. Lay out the weekly, monthly, quarterly, semi-annual and annual touchpoints on a single page with owners and dates. Then identify which of those meetings has never actually happened, and schedule it.

Reflection

• Which stakeholder on your current initiative has statutory power you have been treating as a courtesy notification?
• If a records request landed tomorrow on all project communications from the last quarter, which document would you least want released, and what does that tell you?
• Have you written down what your system will not do, or only what it will do?
• Which of your stakeholders have you briefed but never actually asked for a position, and what would you do if their answer were no?
• When a stakeholder objected most recently, did the program change anything, or did it only record the objection?
• Who on your skeptical audience list have you not spoken to this year, and what is the honest reason?

Glossary

• Stakeholder map. An interest-by-influence matrix naming the statutory and operational stakeholders in a specific initiative, using roles rather than archetypes.
• Message framework. The translation of a single use case into distinct registers for leadership, budget, IT, compliance, program staff, labor and the public.
• Executive briefing. A structured fifteen-minute briefing for a senior executive, with a one-page read-ahead circulated in advance and a decision memorandum with named signatories.
• Expectation management. Explicit phased commitments with NOT-DOING statements, measured baselines and pre-committed decision points at the end of each phase.
• NOT-DOING statement. A written, falsifiable statement of what the system will not do in the current phase.
• Change communication. Coordinated communication about work redesign and workforce implications, negotiated with recognized unions rather than announced to them.
• Duty to bargain. The statutory obligation to negotiate with a recognized union over changes in working conditions before implementing them.
• FOIA test, IG test and oversight test. The three standards every internal communication in government service must pass: front-page comfort, IG accuracy and authority, and oversight-staff judgment of responsible conduct.

Related Lessons

• Communicating AI Projects to Leadership goes deeper into the executive briefing itself and the decision memorandum that accompanies it.
• Communicating AI Success Stories covers proactive public narrative once a program is delivering.
• Change Management for AI Adoption covers the work redesign and workforce transition that thin change communication leaves out.
• Transparency: Citizens' Right to Know covers proactive disclosure obligations and what the public is entitled to see.
• AI for Constituent Services and Public Engagement covers the public-facing side of the same relationship.
• Developing an Organizational AI Strategy covers the strategic artifact your stakeholder plan should be attached to.
• AI Strategy for Different Government Contexts covers how these obligations differ at federal, state and local level.
• Data Governance for AI covers the privacy and records obligations that surface earliest in stakeholder engagement.

Closing

Loretta's forty-eight hours were spent doing in a panic what should have been done in month one: calling the union, calling the senator's staff, and getting ahead of a memo she wished had never been written. The project survived, six weeks late, with a bargaining process that could have run in parallel with development instead of in place of it. Nothing about the technology changed. What changed was who knew what, when, and from whom, which turned out to be the variable that determined whether the department got to keep the system it had built.

Think of stakeholder communication as load-bearing infrastructure. Every briefing and memo is a beam, and skipping enough of them collapses the structure, not because the AI failed but because trust did. The work is unglamorous and calendar-shaped: name your stakeholders as people, write in the register each one needs, say out loud what you will not do, brief before you are asked, and keep a record that would survive a records request, an IG, and a committee staffer reading it two years from now. None of that guarantees a stakeholder will agree with you. All of it means that when they disagree, the disagreement happens in a meeting you scheduled rather than in a hearing you did not.

Key Takeaways

• Communication is load-bearing, not decorative. Skipping briefings creates structural failure points that no technical fix repairs later, and the failures show up as legal, budgetary and press events rather than as technology problems.
• Several of your stakeholders hold statutory power. Budget desk officers, Inspectors General, committees of jurisdiction, recognized unions and privacy officials have authority granted by law, not by your project charter. Treat any of them as a post-launch notification and expect to be stopped.
• Map all four rings before the project starts. Authorizers and overseers, workers and unions, external advocates and the public, and internal partners each face different consequences when they feel bypassed. Use named roles, not offices.
• Check the collective bargaining agreement on day one. Loretta's agreement required 90 days' written notice before any system that changed covered employees' workflow. Agreements differ, so read yours rather than assuming a standard period.
• Tailor every message to the listener's actual concern. A 94 percent accuracy rate means something different to a legislator, a union steward, a privacy officer and a civil rights organization. Write the registers separately.
• Say what the system will not do. Explicit NOT-DOING statements and phased commitments with pre-committed decision points are the antidote to the implicit over-promise that has ended more government technology programs than any technical failure.
• Structure the briefing to the minute. Alignment, progress, risk, value, resources, ask. A one-page read-ahead 48 hours ahead, a decision memorandum with named signatories, and minutes filed within 5 business days.
• Write every internal document as a future public record. Apply the FOIA, IG and oversight tests before sending, keep communications on agency systems, and release governance documents and disparity analyses proactively rather than reactively.
• A briefing cadence produces a record, not agreement. Being informed is not the same as being aligned. Ask explicitly where alignment matters, log the answer including when it is no, and do not let a concern register substitute for acting on a concern.

Frequently Asked Questions

How early is too early to brief a legislative office or a union? There is effectively no such thing as too early, but there is such a thing as briefing without content. The useful threshold is the point where you can state the problem you are solving, the scope you are considering, and what you do not yet know. Briefing at that stage costs you nothing and buys you the ability to say later that the office has been engaged from the beginning, which is the sentence that defuses a hearing.

What if legal counsel wants to slow down every external communication? That tension is structural and mostly healthy, since counsel is protecting against the FOIA and litigation exposure you would otherwise absorb personally. The practical fix is to move review upstream: agree a standing template for the quarterly summary, get it approved once, and reserve individual review for documents that describe employment effects or eligibility decisions. Reviewing a format is fast. Reviewing every document is what creates the delay everyone then works around.

Our agency is a small local one with no IG and no union. Does this still apply? The named federal actors change but the categories do not. A local agency still has an authorizing body, an auditor of some kind, affected staff, advocacy organizations, a records law and a press. Run the same four rings against your own institutional landscape, and pay particular attention to the records law, because state and local public records statutes are often broader in scope than the federal one.

What do we do when a stakeholder simply will not be satisfied? Distinguish between an unmet informational need and a genuine disagreement about values or priorities. The first is a communication problem and more information helps. The second is not, and continuing to send briefings in the hope of conversion is the polite version of ignoring them. Record the disagreement accurately, make sure the decision-maker sees it as a disagreement rather than as a message that has not landed yet, and be prepared to proceed over an objection you have documented honestly.

How much of this record actually gets read by oversight? Usually none of it, right up until something goes wrong, at which point all of it. That asymmetry is the whole argument for the discipline. The cost of maintaining a clean engagement record is a few hours a quarter. The cost of reconstructing one under an IG review, from memory and partial email threads, is measured in months and in the credibility of everyone involved.