The discovery-led playbook

Monday arrives, a client is unhappy and someone asks when the job ad will be ready. You agree that the role needs to be clear, that candidates deserve an honest account of it and that you should examine relevant work. What happens first? What must be settled before you speak to people, and what can you only learn once you do?

This chapter follows one hire from the first request to the offer and handover. The agency, candidates, answers and later events are all invented for teaching. They are not a disguised client case or evidence that the method works. The example shows why the search might wait, how to distinguish a sound response from confident overreach, and what support makes a learning gap acceptable. Chapter 11 follows the hire into the job.

The four stages are Discover, Design, Source, Match. You may have to go back when a candidate exposes a requirement you haven't thought through or a work sample shows that you designed the wrong assessment. Record what changed and why. The printed toolkit contains the forms and exercise used here, so you can run it without buying a service.

Discover the problem before opening the role

Our example is a 15-person agency. Client work is arriving late, hand-offs keep slipping and the founder is drawn into delivery decisions every day. The first request is for a senior operations manager who can take the problem away.

The delivery lead sees it differently. They believe the team has enough senior judgement. It lacks a reliable way to get approvals, keep dates current and make priority changes visible. Another senior person could add a layer of argument without reducing the delays.

The founder's title cannot settle this, and a vote wouldn't explain what is happening. They take four recent jobs that were late and reconstruct the hand-offs: the promised date, work still needed, who had the information and who could approve the next step.

One job sat waiting for the founder's approval even though the delivery lead believed it had been delegated. On another, a specialist received a new priority directly from the founder while the account manager continued telling the client the old date would hold. The founder has described the organisation they intended to create. The team has described how decisions have been made. They need an explicit agreement about how those decisions will work next.

The delivery lead proposes a two-week trial before recruiting. The founder agrees to delegate ordinary priority conflicts to the lead, while retaining decisions that change a contract or price. They create one shared delivery list, name an approval owner and deputy, and remove a duplicate status report.

If this removes the problem and existing staff can carry the remaining work, there is no vacancy. Temporary work might justify short-term support. If the agency cannot fund a role and its training, it may have to reduce commitments or wait. A full inbox alone doesn't tell them which choice to make.

Approvals improve during the trial. Coordination still pulls two account managers away from client work, and neither has spare capacity. The delivery lead examines what remains, rather than treating every hour previously spent on confusion as a future job. Dependencies still need checking, the delivery list needs maintaining and agreed actions need follow-up.

That supports a client-delivery coordinator reporting to the delivery lead. The agency has reduced part of the problem itself and identified the capacity still needed. It no longer has a reason for the original senior operations manager request.

In this example, the founder agrees to keep the approval route that made the trial possible. Without that agreement, the delivery lead would need to recommend a different role or a pause and explain why. The person running the search might not be able to stop it. They would still have to correct any claim about authority that the founder had not granted. An unresolved disagreement inside the business must not become a settled promise to a candidate.

Design the job people will actually do

The Role Decision Sheet in the toolkit records why the job exists, what it owns and what evidence will support the hiring decision. Start with the business problem the agency has just clarified.

The coordinator's purpose is to make the state of client delivery visible early enough for people with authority to act. They keep the delivery list current, check dependencies, confirm owners and dates, and flag risks. Their responsibility does not extend to every result that could affect a deadline.

The delivery lead sets priorities and allocates capacity. Account managers own client promises and updates. Creative specialists own work quality. The founder decides contract and price exceptions. The coordinator can organise a hand-off and follow up agreed actions, but cannot promise a client a new deadline or overrule the delivery lead.

Writing this down exposes a practical question: what happens when the delivery lead is unavailable? The sheet names a deputy. Without one, “take ownership” would leave the coordinator responsible for progress without a way to get a decision.

The agency needs evidence that someone can track several dependencies, distinguish urgent work from important work, write a clear update and escalate at the right time. They need basic competence with shared task records. A particular software brand is unnecessary if the person can learn it with the support available.

Agency experience could reduce the learning required, so it remains useful. The delivery lead must decide which knowledge is needed on arrival and which can be taught, by whom and within what realistic period. Calling something trainable does not supply the time or the teacher.

The sheet also describes the harder conditions. Client requests can change. The coordinator will chase colleagues who are busy or senior to them. The team needs an agreed daily overlap window. Routine follow-up fills part of the week, alongside occasional pressure. None of that needs dressing up as an adventure.

The founder, delivery lead and account managers check the sheet against the incidents they examined. If one still expects the coordinator to make a decision another plans to retain, the design is unfinished. The person accountable must make that decision and leave its consequences visible, even if the wording isn't everyone's preference.

Source widely, then narrow for a reason

The agency publishes a plain account of the work, pay range, working pattern, decision rights and selection steps. It invites applications and uses relevant referrals and direct outreach to reach people who may miss the advertisement.

Applying does not make someone less capable, any more than having to be approached makes them stronger. Employment status tells the agency about availability. It cannot substitute for evidence of the work.

The first screen is small. Candidates provide an existing example or a short description of coordinating work with competing deadlines. They are told not to share confidential client material. A description is acceptable because not everyone has permission to take work away from an employer. The agency also checks essential conditions, including the overlap window and pay range.

Before reading applications, the delivery lead records what will move someone forward: an intelligible example of handling dependencies, a credible account of their own contribution and no unresolved conflict with an essential condition. A brief answer may provide that evidence. A polished essay may not.

If applications arrive in large numbers, the agency reviews batches against the same requirements, records its reasons and checks uncertain exclusions. It seeks a short clarification where missing information could change the decision. A large applicant pool doesn't justify handing everyone an unpaid assignment to reduce the employer's reading.

The shortlist must allow the agency to examine plausible options and give each person the attention promised. No fixed number can guarantee that. If nobody supplies the necessary evidence, the agency reviews the search or requirements. It cannot quietly lower an essential requirement to meet a deadline.

Referrals go through the same process. A trusted colleague's recommendation can help find someone, but doesn't establish how they will handle this work.

Match with questions and a fair piece of work

The interview begins with the role, including what the employer hasn't made easy. Candidates need that account to ask useful questions and decide whether to continue. These conversations also let the agency learn their individual priorities.

The delivery lead uses the same core questions for each shortlisted person: a hand-off that went wrong, competing deadlines, an unclear decision boundary and a time they questioned a manager's request. Follow-ups seek the candidate's actions, information available at the time and what happened afterwards.

The best answer isn't necessarily the story closest to the interviewer's career. Someone from another industry may have faced the same underlying problem. Someone with little experience may need a relevant scenario to show their reasoning. Past examples and hypothetical situations can both help if the assessor is clear about what each establishes. Neither gives direct access to a person's character.

Candidates can test the employer's account too. Who changes deadlines? What happens when the founder bypasses the delivery lead? Which tasks are repetitive? What support will help them learn the agency? An unanswered question belongs in the decision record.

Next comes a short simulation, the Firefighter Morning, using a fictional problem that resembles the role. No live client output is requested or used. The complete brief and two invented response extracts are in the toolkit.

Both candidates have 30 minutes to prepare and 20 minutes to discuss their response. This states the intended demand, rather than rewarding whoever finishes fastest. The agency offers accessible formats and adjustments, supplies equivalent tools and explains what information is available. Candidates may ask clarifying questions. An answer that changes a material fact goes to everyone.

Normal tools and optional AI are allowed. Candidates explain what they used and remain responsible for checking the result, without surrendering personal accounts or private notes. A different role requiring unaided writing would need a different, clearly stated rule. Changing the standard after seeing a polished answer would undermine the assessment. Any substantial further assignment would require payment, scope and use to be agreed beforehand.

The criteria are visible before the exercise: establish the immediate risk, order actions using the facts, respect decision rights, communicate clearly and identify uncertainty. A delivery lead who understands the work assesses the responses. Another reviewer can challenge that assessment but cannot replace it with an impression of confidence.

Here is the problem. A client deliverable is due at noon. At 9 am, a reviewer finds an error that may affect the recommendation. A specialist needs 30 minutes to assess it. The account manager is on a call until 9.30 am, the delivery lead is available, and another client expects a routine update at 10 am.

Candidates receive the role boundaries and are asked for a first-hour plan, a brief internal update and one uncertainty or action they would defer. They are coordinating a response, not supplying the technical answer to the client's problem.

Candidate A has less agency experience. Their answer first checks the error with the specialist and alerts the delivery lead. It keeps the account manager informed without inventing a new client promise. A also leaves a route for the routine update, rather than forgetting it when something more dramatic appears.

Candidate B has more agency experience and knows the language of client delivery. Their initial plan immediately promises the client that the work will arrive at noon, then asks the team to make it happen. The answer sounds decisive. Capacity and the error are still being checked, however, and the coordinator does not own the promise.

The assessor asks both what they would change if the specialist said the error needed another two hours. A adjusts the internal escalation and asks the account manager to agree an updated client message. B still assumes the coordinator can commit the team without the delivery lead's decision.

That gives the concern about B a job-related basis. After clarification, the response still crosses a consequential authority boundary.

A has shown a plausible approach to a short, stated problem. The agency hasn't observed weeks of follow-up, all the technical knowledge involved or a difficult day. The interview, relevant prior work and later checks still matter. This is evidence about part of the job, not proof of the whole person.

Decide on the evidence, including what is missing

The decision record puts requirements beside evidence. It doesn't rank enthusiasm or turn an impression into a decimal score.

A has stronger evidence of ordering actions and communicating within the role's authority. B's agency knowledge could shorten the learning period, but the response leaves a concern about commitments outside the role's authority. A will need help learning the agency's dependencies. The choice involves those particular strengths and gaps, not a verdict that one person is generally good and the other bad.

Reviewers make initial notes separately, then compare them. A disagreement sends them back to the answer that produced the judgement. “I liked them” is not something another reviewer can inspect. “They changed the deadline without checking who could approve it” is.

They decide to offer the role to A, provided the working pattern and support can be agreed. A needs to learn how the agency moves work from client approval through specialist production and review. For the first two weeks, the delivery lead sets aside half an hour each morning to work through A's delivery list, check dependencies and explain unfamiliar hand-offs. This teaching time replaces part of the lead's normal internal reporting work. Without that capacity, the reason for accepting the gap disappears. The agency would need to pause, change the role or keep searching.

This is the useful part of a candidate package: what the role needs, what was observed, where accounts disagree, what remains uncertain and why the recommendation follows. A resume can sit beside it as background without carrying a decision it cannot explain.

The record must also allow a no. If both candidates lack evidence of an essential requirement, choosing the less worrying one won't supply it. A hiring deadline may justify a different staffing plan. It cannot turn missing evidence into positive evidence.

Make the offer honest enough to decline

A wants a dependable daily overlap window and clarity about what happens outside it. The employer wants coverage for delivery decisions. They agree a window and an emergency rota, rather than allowing “flexibility” to become permanent availability.

The agency has published a pay range and agrees terms within it. A can ask for more, and the agency can state what it can afford without turning the negotiation into a judgement of A's values.

The delivery lead explains the training, deputy arrangement and work A can own. There is no guaranteed promotion to operations manager, and routine chasing will remain. A job can be worth choosing without providing everything someone might eventually want.

The employer can advocate for the work and explain why it thinks A could do it well. A can negotiate, seek information or decline. If essential hours cannot be reconciled, a reasoned no helps both sides. Agreement reached by concealing the hard part is a debt someone will collect after the start.

Hand over what you learned, then check what happens

The delivery lead receives the role record, selection evidence, known uncertainties and offer commitments. In the first week, A gets a bounded delivery list and guided review of its dependencies, beginning useful work while learning the agency.

They plan check-ins for the end of week one and days 30, 45 and 90. These dates belong to this example. For another role, the manager would need occasions that reveal enough real work early enough to act.

The handover gives the manager a record to test against the work. Chapter 11 returns to this hire at day 30, when the first difficulties need a response.

Use the depth the decision needs

An ordinary role with known work may need a few focused conversations, recent examples and a clear decision sheet. A senior role crossing several functions may need much more investigation. Regulated or safety-critical work also needs the relevant specialist standards. Match the time spent to the uncertainty and consequences.

Elapsed search time is different from time spent understanding a role, and neither becomes valuable because it is long. Ask whether another round of work is likely to change a consequential decision and whether an important uncertainty has been left without an owner.

You now have a sequence of decisions against which to judge what AI might do: the information needed, errors that would matter and responsibility someone must retain.