MatrixBy Khaled Hawari

Recording What Each Manager Agreed To

Budget disagreements resurface in April, and the only useful defence is a record of what was actually agreed in October.

The record described here is a management artefact that the finance function happens to keep. That distinction is not pedantry, and it decides whether the thing works.

Here is the failure it prevents. A manager asks for two people. The budget comes back with one. Nobody writes down who decided that, or on what basis, or what would have to be true for the second one to be reconsidered. Six months later the department is behind, the manager says they asked for two, and the conversation that follows is about whose memory is better. Meanwhile the finance function, which merely typed the number, has spent those six months as the department’s antagonist, because in the absence of a recorded decision the person who reported the cut is assumed to have made it.

That assumption is usually unfair and it is not entirely wrong. When a submission comes back reduced and no owner is named, finance did make the call, by default, in the consolidation. The record fixes both problems at once: it forces someone with authority to actually take the decision, and it puts their name on it.

So: the owner decides, the manager signs, and finance holds the pen. Where a company cannot say which of those three did what, it does not have a budget process. It has a spreadsheet that got smaller. Holding the pen without holding the decision is one of the more useful things a fractional finance lead does, and it is harder than it sounds, because the quickest way to finish a consolidation is to make the call yourself.

What the record is not for

It is not a method for deciding where to cut. That question belongs to whoever runs the business, it depends on things a finance function does not know, and any general rule offered here would be advice about somebody else’s strategy dressed up as process. This piece takes the decisions as given and concerns itself only with recording them well enough that they survive the middle of the year.

The record

One row per request, not one row per department. A department asking for three things is three rows, because the three will be answered differently and a merged row hides that.

Field What goes in it The failure it prevents
Department and requester The person, not the function A decision attributed to a department nobody owns
The request What was asked for, in the requester’s own words, quantified the way they quantified it The request being restated by finance into something easier to decline
Requested effect The lines of the plan it would move, and when An argument about cost that turns out to be an argument about timing
Outcome One of the five below. One word, from a closed list. “Partly approved”, which means nothing in six months
What was approved The version that went into the plan, stated in full even where it is identical to the request A holder discovering the approved version differs from what they remember agreeing
Condition or trigger What would have to be true for a deferred item to be revisited, written as a test A deferral that quietly becomes a refusal because nobody defined the reopening
Review point The specific reforecast or close at which the trigger is checked, and by whom A trigger nobody ever looks at
Decided by A named person with the authority to decide it Finance being assumed to have decided
Date The date the decision was taken, not the date the record was typed A record backfilled after the argument started

Nine fields. It fits on one page turned sideways and it is the cheapest artefact in the whole budget cycle.

The five outcomes, and why one word is not enough

Most companies record two states, in and out. The interesting cases are all in between, and lumping them produces exactly the mid-year dispute the record exists to prevent.

Outcome What it means What the record must also carry
Approved The request went into the plan as asked Nothing extra. These rows are short.
Approved, changed The need was accepted, the shape was not. A permanent hire approved as a contract, a purchase approved as a lease, two roles approved as one plus overtime. The reason for the change, because the requester will need to explain it to their own team
Deferred to a date Accepted, and starting later than requested The month it now starts, in the plan, phased. A deferral that is not phased into the plan is a refusal with better manners.
Deferred to a trigger Accepted in principle, held until something observable happens The trigger, the measure it is read from, and who checks it
Declined Not happening this year One sentence of reasoning. This is the row people skip and it is the row that gets read.

The third and fourth are different animals and they get confused constantly. A date deferral is a commitment with a start month, and it belongs in the loaded budget. A trigger deferral is a contingency, it is not in the plan at all, and the company should know roughly what it would cost if every trigger fired in the same quarter. Add that total up once. It is frequently larger than anyone expects, and knowing it is the whole argument for writing triggers carefully.

Writing a trigger that will actually be checked

Most triggers fail on the same defect: they are written as sentiments rather than as tests. “Revisit if the team is struggling” cannot be checked, so it never is, and the deferral becomes a decline through inaction.

Three rules make a trigger real.

It reads from a number that already appears in the monthly pack. If checking the trigger requires somebody to build a report, the trigger will not be checked. Backlog, utilisation, headcount against plan, revenue against plan, a service level the operations team already tracks. Where the natural measure is not in the pack, either add it to the pack now or pick a different measure, and make that choice during the build rather than discovering it two quarters later.

It has a duration. A single month above a threshold is noise in a company this size. Two or three consecutive periods is a condition. Write the duration into the trigger or the first bad month will reopen the argument.

The threshold is set by the person who owns the measure, in the meeting, not by finance afterwards. The operations lead knows what backlog level is uncomfortable. Finance does not, and a threshold invented by finance is one the operations lead will disown at the moment it matters.

A worked page of the record

Illustrative rows, showing the shape rather than any company’s actual decisions. Note that none of them carries an amount: the amounts live in the plan, and duplicating them into the record creates a second version to keep in step.

Requester Request Outcome What was approved Trigger and review point
Operations lead Two field technicians, both starting in the first quarter Deferred to a date One in the first quarter, one in the third, phased into the plan accordingly None. This is a dated commitment, in the plan. Confirmed at the second quarter reforecast.
Operations lead Replace the vehicle that is out of warranty Approved, changed Lease rather than purchase, same month None. Cash effect re-phased in the capital plan.
Sales lead A second inside salesperson from the second quarter Deferred to a trigger Not in the plan Reopened if new bookings hold above the agreed run rate for two consecutive months. Checked by the sales lead at the second and third quarter reforecasts.
Sales lead Increase the trade show budget Declined Baseline retained None. Reasoning recorded: the two shows attended last year produced no traceable pipeline, and the sales lead agreed to reassess after the pipeline reporting is rebuilt.
Client services Move one contractor onto payroll Approved As requested, from the month stated None
Client services A software platform to replace three tools Deferred to a trigger Not in the plan Reopened once the client count passes the level at which the current tools stop working, as counted in the monthly pack. Checked by the client services lead at each quarterly reforecast.
Finance An additional part-time bookkeeper Declined Baseline retained None. Reasoning recorded: the close is being rebuilt and the workload will be reassessed once the new close calendar has run for a full quarter.

Two things are worth noticing about that page.

The finance row is on it. A finance function that records everyone else’s declined requests and quietly leaves its own out has told every other manager exactly how much the record is worth. Put your own refusals on the same page as theirs.

And the reasoning appears only on the declined rows and the changed ones. Approved rows do not need a justification and giving them one makes the page longer without making it more useful. This is the kind of asymmetry that keeps an artefact alive: the pages people actually maintain are the ones that only ask for information when it earns its place.

Circulation, and the signature that matters

Each manager sees their own rows and confirms them in writing before the budget locks. Not the whole record. Their own rows.

That single step does most of the work. A manager who has read and confirmed “one in the first quarter, one in the third” cannot credibly say later that they asked for two and were ignored, and more usefully, they will not want to, because they were part of a decision rather than the recipient of one. A manager who was never shown the final row has every reason to dispute it, and often a good one, because reductions do get applied in consolidation and not communicated back.

Whether the full record circulates to everyone is a judgement about the company. Some owners want every manager to see the whole page, on the grounds that a manager who can see what everyone else was declined stops treating their own decline as personal. Others find that it turns the next cycle into a negotiation between departments. Pick one deliberately and be consistent about it, because switching between the two years reads as concealment.

April

The conversation the record exists for goes like this. The department is behind, the manager raises the second technician, and instead of a memory contest there is a row on a page.

Read the row and one of three things is true. The trigger fired and nobody checked it, which is a process failure and the hire should proceed. The trigger did not fire, in which case the conversation moves to whether the trigger was the right measure, which is a much better argument to be having. Or the item was declined outright with reasoning, and the manager is asking for the decision to be revisited, which is a legitimate request that goes to the owner as a new request rather than as a grievance.

All three are short conversations. None of them is about who said what during the build. That is the entire return on nine fields.

What we do not do

We do not decide where a company cuts, and we say so plainly when asked to, because a finance function that quietly allocates reductions has taken a decision it will still be carrying eight months later when the department is short.

We do not record an outcome as deferred without a date or a trigger against it, because a deferral with neither is a decline that nobody had to defend.

And we do not let the record be written from memory after the meeting. It gets filled in the room, in the meeting where the decision is taken, or it gets filled in badly.

The engagement behind this document: The Reporting Pack We Deliver Every Month.

MoreOther working documents

If this keeps failing in the same place.

A document that has to be re-explained every period is a process problem rather than a documentation problem. That is the point at which handing the function over is cheaper than fixing it again.