The Unwritten Enterprise · Part II

On Friday afternoon, a learner named Nora finishes the practical assessment in a spreadsheet analysis course. Her instructor watches her reconcile two sets of figures and tells her she passed. That evening, the course portal still says “assessment pending.”

The instructor uploaded the signed assessment packet, but the page with Nora’s score is missing. Learner support has an email from Nora. Her employer needs evidence of course completion by Monday to reimburse the fee. The instructor can confirm what she saw. The records office has to stand behind any certificate it issues, including months later when someone asks how the result was established. Learner support owes Nora an answer before the records office reopens.

The training provider is fictional. Its problem is ordinary. Who can record that Nora passed? What can learner support tell her while the record is incomplete? Who can issue the certificate, and what evidence must that person have? The answers may be in a policy, a software permission, or the phone number of a colleague who still picks up on Friday evenings. Taken together, they describe how the provider actually operates.

In Business Architecture Has a Power Problem, I argued that architects need to understand who can shape a decision and who can make it stick. The next question is what happens when those arrangements become ordinary. Over time, an enterprise settles who gets discretion, whose evidence counts, which obligations come first, and where the cost lands when two reasonable commitments collide. That settlement is as much a part of the operating model as the capabilities, roles, and systems that carry it out.

The deal under the diagram

I use settlement in a fairly plain sense. It is an arrangement that lets people with different responsibilities keep working together. It can be a negotiated agreement, an executive decision, or an accommodation nobody remembers making. It may have been imposed by a stronger party. It may survive because reopening it would cost everyone more than putting up with it. The word implies enough stability for people to plan their work around it. It carries no presumption of consent or fairness.

There is a reason that stability matters. Richard Cyert and James March, in A Behavioral Theory of the Firm, treated conflicting demands as an ordinary feature of organizational decision-making. An enterprise can live with objectives it has never fully reconciled because it attends to them in different places or at different times. One team tightens a control. Another handles the cases the control slows down. Leadership accepts the arrangement as long as both reported outcomes stay within bounds. The day a new system forces both sides to follow the same rule, the unresolved conflict becomes visible.

That distinction helps explain why an operating-model workshop can seem so productive on a Tuesday. Everyone supports prompt answers, reliable records, timely certificates, and an appropriate amount of local discretion. The agreement is sincere. Each person can still imagine a different answer to the first difficult case. Once the team specifies the evidence required, the person who can authorize an exception, and the time by which someone owes the learner an answer, the choices get harder. People can see which promise they will have to keep with someone else’s permission.

When a result goes missing at this provider, the program coordinator usually calls the instructor and checks the roster. Learner support trusts the coordinator’s word because it has learned what she checks before she calls back. The records office keeps the certificate on hold until it has the assessment record. A records supervisor occasionally accepts a signed replacement from the instructor, then saves it in a shared folder where the next reviewer can find it. Nobody designed this whole arrangement. It works tolerably well because enough people know their part and remember whom to call.

Elinor Ostrom called attention to the working rules people invoke to explain and justify their actions to one another. Some are written. Others are worked out through repeated dealings. An architect can learn more about those rules by asking a records supervisor to walk through three actual exceptions than by asking whether the team follows policy. Both accounts matter. The policy establishes what the organization has authorized. The cases show what evidence people relied on and what they did when the policy left them with a learner waiting on the phone.

What a single status would settle

Suppose the provider replaces its old tools with a common platform. The proposed design records the assessment as passed only when a reviewer accepts the complete, signed assessment record. That result then appears in the learner’s portal and satisfies one of the conditions for issuing the certificate. The attraction is obvious. One verified result, one record, fewer contradictory answers.

The rule also gives one piece of evidence three jobs. It establishes the provider’s official record of the assessment, governs what learner support may say, and clears the assessment requirement for certificate issuance. If the packet is incomplete, all three wait. The instructor can confirm what happened in the room, but her account has no standing in the system until someone recovers the missing page. The records office gains a safeguard. Learner support inherits the wait. Nora is left to wonder whether she misunderstood her instructor.

It is easy to caricature either side. The instructor can be right about Nora’s performance and have little idea what it takes to reconstruct a decision after an instructor has left. A records reviewer can be right about the evidence required and have no idea that Nora stands to lose her reimbursement. Both forms of ignorance are predictable. The work puts different consequences in front of different people. Their disagreement is one of the few places the enterprise gets to see both at once.

This is where I would start with the smallest useful piece of architecture. The question is narrow enough to work through in a room with actual cases. What can the provider tell a learner when the instructor has confirmed a pass and the official record is still incomplete?

The four core business architecture domains each clarify a different part of the answer.

  • Capabilities. Assessing learner performance, communicating results, and authorizing certificate issuance are distinct abilities. Linking them may be appropriate. A shared status field gives us little basis for deciding how tightly to link them.
  • Value streams. A learner who undertakes training needs to know what they can do and where they stand. A learner requesting a certificate needs a durable record they can present to someone else. Feedback can be valuable while the provider is still checking eligibility to issue that record.
  • Organization. The instructor assesses the work. The program office can verify the source of a reported result. Learner support can communicate what the provider knows. The records office can authorize the certificate. Someone still needs the authority and capacity to resolve a case when the ordinary path fails.
  • Information. The assessment result, the signed record that substantiates it, the status shown to the learner, and certificate eligibility have distinct meanings. A correction to any one of them needs a clear effect on the others. Otherwise the software will supply a rule by default, and the business will discover it when a case breaks.

It takes very little modeling to expose the decision. Take one learner’s case, put the evidence and timestamps in order, and show which role may make which assertion at each point. Then ask the people who will work the exception to try it. If someone has to make an off-the-record phone call before the case can move, ask what that call supplies. It might be missing evidence, missing permission, or simply an answer the designated owner cannot give in time. Each has a different remedy.

A workable bargain

One option is to let a program lead record the instructor’s confirmation while the signed assessment record is pending. The entry preserves the instructor’s identity, the time of confirmation, and what the instructor actually confirmed. Learner support can then tell Nora that her instructor confirmed a passing result and that the records office is still reviewing her file. The certificate stays on hold until the records office receives the missing evidence or an authorized reviewer accepts a documented replacement under the program’s rules. The program office owns recovery. Cases still open after an agreed interval go to someone who can decide what happens next.

That is one plausible arrangement. It has costs. A program lead now has to make and record a judgment. Learner support needs language that makes clear what the instructor confirmed and what the records office still has to decide. The records office needs enough reviewers to handle the exceptions it retains. The provider has to correct the status and contact the learner if a confirmation turns out to be wrong. Each obligation requires capacity, and capacity costs money. If nobody is willing to allocate it, the proposed discretion exists mostly on paper.

A strict common gate could be the sounder design for programs that require formal review before any result is communicated. It might also be the simpler default if records arrive promptly and the few remaining cases can be resolved before learners feel the delay. A separate confirmation path might make sense where assessment records arrive unevenly, instructors can verify results promptly, and learners face deadlines of their own. The cases should tell us which arrangement is warranted. A diagram that looks clean under either design has little to say about the choice.

The difficult work is making each obligation reciprocal. If program staff lose the discretion to communicate a result directly, they need a dependable route to a decision, especially when courses run outside office hours. If the records office accepts replacement evidence, it needs the right to reject weak submissions and a record that will survive staff turnover. If learner support can communicate an instructor’s confirmation, it needs to know who must correct that communication if the evidence changes. Put each promise beside the authority that makes it possible. An escalation to an unattended queue tells the program team everything it needs to know about whose problem this really is.

Performance measures belong in the bargain too. A program office rated on how quickly it clears pending results has an incentive to record confirmations before the evidence is sound. A records office measured on errors in issued certificates can do well by letting difficult files sit. A support desk measured on time to close a ticket may hand off a case it could have resolved. A useful review follows the same cases across the handoffs. How long did the learner wait for an accurate answer? How often did the provider have to correct one? How long did eligible learners wait for certificates? Who spent time chasing the evidence? A gain that depends on handing uncounted work to another team deserves a second look.

What the workaround knows

The temptation in a redesign is to sweep away the old phone calls and side files. Sometimes that is exactly right. A personal contact can turn into privileged access. Learners who know whom to call get exceptions their classmates never hear about. The shared folder can become an unofficial record that nobody is obliged to maintain. An inherited workaround earns an explanation before it earns a place in the target design.

Michael Lipsky’s work on discretion in public services helps explain why. He studied people who apply public policy case by case under limited time, imperfect information, and competing demands. His setting was government, and the duties those workers owed citizens differ from a training provider’s obligations to learners. The comparison is useful because both settings give frontline workers judgment over facts that arrive late, ambiguous, or incomplete. A policy that removes one outlet for that judgment may simply move it to a place that is harder to see.

Suppose the new platform can show only a verified result or a pending one, and the records office has no way to hear urgent cases. Learner support still owes Nora a reply. A staff member might forward a private message from the instructor. Another might attach the incomplete packet again under the wrong document type, hoping to get it in front of a reviewer. The dashboard can show a tidy backlog either way. You would have to trace what Nora was told and what the reviewer actually received to understand what the rule produced.

The opposite mistake is to preserve discretion wherever it has helped someone. Suppose coordinators get quick replies from instructors they know, while evening staff covering unfamiliar courses can rarely reach anyone. Or suppose a coordinator has been treating a passing result on the practical assessment as proof that a learner met every course requirement. Those are reasons to define when someone can use the exception, give staff access to the same evidence, and specify exactly what each status means. The local practice tells you where the design needs judgment. It also tells you where the judgment needs a boundary.

Who can reopen the settlement?

An operating model has to survive changes in people, volume, program requirements, and technology. The person who holds the arrangement together today may take another job. A program may introduce a second reviewer for certain assessments. A better submission process may make incomplete records rare enough that a staffed exception desk is wasteful. And a rule that works on twenty cases a week may fail when an entire term ends and two hundred files reach the records office by lunch.

The initial decision should name who can reopen it and what evidence would justify doing so. For this provider, I would pilot the narrower arrangement in one program whose rules allow it and look at ordinary cases, contested results, and the work that lands outside regular hours. If provisional messages regularly need correction, tighten the evidence requirement. If credible confirmations are available but nobody can review them before learners need an answer, revisit who is authorized or when that authority is available. If cases cluster around a defect in how assessment records are submitted, fix that process and see how much exception handling is still necessary. Those are different responses to different findings.

I would keep a short decision record tied to the part of the operating model it governs. Give it the original problem, the options considered, the rationale for the evidence threshold, the delegation and its limits, and the review trigger. A successor who finds that record can tell whether the rule was meant to protect the integrity of the credential or simply to compensate for a limitation in an old system. That is a better starting point for changing the model than asking a room of people why the business has always done it this way.

There is a limit to what an architect can settle. We can show where the choices sit, identify the obligations a proposal leaves uncovered, and give the decision-makers a case they can argue about in concrete terms. The people who bear the risk and have the authority to commit the enterprise still have to make the bargain. If they agree to faster answers for learners and decline to staff the work that makes those answers possible, record that choice plainly. It will otherwise be made again, a little differently, by each person covering the desk.

At the end of the next course, another assessment record will be incomplete. The instructor can tell you what she saw. The records office can tell you what it needs. A good operating model gives both of them somewhere legitimate to put what they know, and gives the person answering the learner’s email enough authority to say where things stand. The learner should be able to get that answer without knowing anybody’s personal phone number.

Sources and further reading

  • Richard M. Cyert and James G. March, A Behavioral Theory of the Firm. Originally published 1963. Second edition, 1992. Wiley
  • Elinor Ostrom, Understanding Institutional Diversity. Princeton University Press, 2005. See chapter 1 on the origins of rules and working rules. Princeton University Press chapter excerpt
  • Michael Lipsky, Street-Level Bureaucracy. Originally published 1980. 30th anniversary expanded edition, Russell Sage Foundation, 2010. Russell Sage Foundation