Approval workflow design for organisers who launch a new edition monthly
Approval workflow design works best when the weight of an approval matches how reversible the decision is. A copy edit needs one peer. A show name change needs the brand owner. Adding approvers raises elapsed time without a matching drop in failures, so count approvers per brief and multiply by median response time.
A group with eleven shows launches something roughly every four weeks. Somebody counts the approval steps on one edition and gets to nineteen, spread across seven people. Nobody designed that. Each step was added after a specific incident, and each addition was reasonable at the time: the wrong logo went out, so brand now approves; a price was wrong on the site, so finance now approves; a speaker was misspelled, so the content lead now approves.
Approval workflow design is the discipline of undoing that accretion on purpose. The organising principle is simple to state and uncomfortable to apply: the weight of an approval should match how hard the decision is to reverse.
Sort every decision by reversibility
Take the thirteen sections of your brief and ask one question about each field. If this were wrong and we noticed on the day of launch, what would it cost to fix?
A typo in a session description costs nothing. Somebody edits the page and it is gone in four minutes.
A wrong rate on the exhibitor rate card costs a difficult conversation and possibly a discount you did not intend to give, because a published price that somebody acted on is partially binding whether or not it was a mistake.
A show name change costs everything downstream of it: domain, logo, signage, badge stock, existing contracts that reference the old name. That decision is effectively irreversible eight weeks out.
Those three sit at different points on a scale, and the approval they get should differ accordingly. Copy edits need one peer, and that peer can be whoever is nearest. Pricing needs the person who owns the number and nobody else. A name change needs the brand owner and it needs to happen before anything is drawn.
What most workflows do instead is apply the same chain to all three, which means the copy edit waits for the brand owner and the name change gets the same nod as the copy edit.
Does adding an approver reduce failures?
This is the question that decides the whole design, and there is published evidence on it.
The DORA 2019 State of DevOps report examined formal change management processes that require approval from an external body, such as a change advisory board or a senior manager, before a significant change can go ahead. Their finding, in their own words on the research site, is that "no evidence was found to support the hypothesis that a more formal, external review process was associated with lower change fail rates". The explanation they give is mechanical. Heavyweight approval slows the delivery process, which pushes teams towards larger batches released less frequently, and a bigger batch carries more risk when it lands.
The event version of that mechanic is familiar. When approvals are slow, teams stop shipping small corrections and start bundling everything into one big pre-launch push, and the big push is where things break.
That research is about software. The transfer is not automatic and I would not claim it is. What does transfer is the falsifiable part: if your organisation believes extra approvers reduce failures, that belief is testable against your own incident log, and the DORA result should at least make you check rather than assume.
What does peer review buy that a board does not?
The alternative that the same research recommends is peer review during the work, with automated checks where they exist.
Rigby and Bird looked at what peer review converges to across a set of large organisations in their 2013 paper at the FSE conference, covering projects at Microsoft, Google and AMD. Two of their findings matter for approval design. The median number of people who actually take part in a review is two, even at Microsoft where the median number invited was three or four. And when more reviewers were active, they found "only a minimal increase in the number of comments about the change", with no increase in the number of rounds.
More invited reviewers produced more names on the review and about the same amount of feedback. If that holds for briefs, and everything I have seen suggests it does, then the fifth approver on a section is contributing calendar days and very little content.
Their other convergent finding is about timing: review happens early, on small changes, close to when the work was done. An approval that arrives three weeks after the section was written is reviewing something the author has forgotten, and the cost of acting on it has gone up while the value has gone down.
The approval tax, calculated
Put a number on your chain before you argue about it. The calculation takes an hour with your workflow log.
For each section, count the approvers who sit on the critical path, meaning the ones whose approval must arrive before the section can move. Then take the median response time per approver, measured from the moment the section entered their queue.
Suppose a thirteen section brief averages 3.1 critical path approvers per section, with a median response of 1.8 days. Sections that are truly serial carry 3.1 times 1.8, which is 5.6 days of waiting each. Reading time, if you time it, is perhaps twelve minutes per approver, so 37 minutes of work per section against 5.6 days of elapsed time.
Across thirteen sections the reading total is 8.1 hours, which is one person-day of actual effort. The brief, meanwhile, is not waiting 5.6 days. It is waiting as long as its slowest chain, and if the same brand owner sits on nine of the thirteen chains, those chains are not independent and the queueing behind that one person governs the date. That is a different problem with its own arithmetic, and it belongs with the bottleneck analysis.
The 8.1 hours is the number to show anybody who defends the current chain on the grounds that approvals are cheap. They are cheap in effort and expensive in time, and the two are being confused.
Run one more cut on the same export. For each approver, count how many sections they touched and how many of those they changed anything on. An approver who has looked at 140 sections in a year and requested a revision on four of them is approving at a 97 per cent pass rate. That rate is either evidence that the upstream work is excellent or evidence that the approval is a formality, and you can tell which by sampling ten of the 136 passes and checking them properly. My expectation, which your own sample can refute, is that the clean rate on that sample will be well below 97 per cent, and that the approver will be glad to be taken off the chain.
A two tier design that survives a monthly cadence
The version that works for a team launching every four weeks has two tiers and one written rule for which tier applies.
Tier one is peer approval. One named peer, anybody with the relevant competence, approves and the section moves. This covers everything reversible: copy, session descriptions, most imagery, anything that can be edited after publication without contacting a third party.
Tier two is owner approval. Exactly one named owner, specified in advance, approves and nobody else is required. This covers anything binding or expensive to reverse: names, prices, contractual language, dates, capacity claims. The list of tier two fields should be short enough to fit on one screen, and if it grows past about fifteen fields somebody is being cautious rather than thoughtful.
The written rule is the reversibility test above, applied once per field when the template is designed rather than argued about per brief.
Everyone else who currently approves becomes a consultee, meaning they are notified and can comment, and the section moves without waiting for them. That is the change that saves the calendar, and it is the change people resist, because being removed from an approval chain reads as a loss of standing. Naming who holds each tier two field is worth doing properly, and an authority matrix is the instrument for it.
Where this stops
Two tiers is a simplification and there are cases it handles badly.
Regulated content is the clearest. If your show sells to a sector where claims carry legal exposure, a legal check is a tier two approval that cannot be collapsed into peer review, and the response time on it is outside your control. Design around it by getting the language reviewed once per year as a template rather than once per edition as a section.
The second gap is competence. Peer approval assumes the peer knows enough to catch what matters, and on a team of three where one person owns everything technical, there is no peer. In that situation a single approver is honest and a second approver is theatre, and it is better to say so than to invent a chain.
The third is that removing approvers redistributes accountability rather than removing it, and if nobody notices that shift, the first bad launch after the change gets blamed on the new workflow. Write down who is accountable for each tier two field before you remove anybody, which is also the argument for one named owner per section rather than a shared list.
Do this next week. Export the approval events for your last three editions, count distinct approvers per section, and calculate the median response time for each person. Sort the people by median response time and look at the top of that list. If one person accounts for more than a third of the total waiting across the portfolio, the design question is not how many approvers you have, and you should read the workflow data in your brief and publishing workspace with that person's queue in mind.
Questions people ask about approval workflow design
- How many approvers should a brief section have?
- One for reversible content and two at most for anything that binds outside the team. The count matters less than whether each approver can say what they are checking for. An approver who cannot answer that question is a signature rather than a control, and removing them costs nothing measurable.
- Does an approval board reduce the chance of a bad launch?
- The DORA 2019 State of DevOps report found no evidence that a more formal external review process was associated with lower change fail rates. Its explanation is that heavyweight approval slows delivery, which produces larger and less frequent releases, and larger releases carry more risk. The same mechanics apply to editions.
- What is the approval tax on a brief?
- Multiply the number of approvers on the critical path for a section by the median time each takes to respond. Three approvers in series at 1.8 days each is 5.4 days of waiting on one section, against perhaps forty minutes of reading. Do that across every section and the total is the tax.