Clear the access before you fund the study

What you are buying when you fund a study

You approved an assessment to settle a specific question. Does this line need capital, or does it need to be sequenced differently. The deliverable you think you bought is a conclusion. What you actually bought is a set of observation opportunities, and every one of them depends on a permission or a window that somebody inside your plant controls. Nobody prices those. They do not appear in the scope, they do not appear in the schedule, and they are not anybody's job.

Here is what that looks like when it fails. In one engagement a measurement week lost 2 of its 5 days because plant maintenance was working on the installed sensors during the measurement window. No one had agreed who owned that window, so both answers were reasonable and the instruments sat dark. The same week, the field resource installed sensors, went to photograph the installation, and was told photography was not approved because the equipment was proprietary. He asked again the following day and got the same answer. A populated data room went unopened. The one person who could explain why the line behaved the way it did was in the building all week and was never interviewed, because no one had put the interview on a calendar. Not one of those was a method failure. The sensors were the right sensors, in the right place, for the right duration.

Access is a separate system from method

A study has two subsystems that fail independently. Method is what to measure, where to place the reading, how long to run it, and what statistic the result supports. Access is whether that measurement is physically and administratively possible inside the window you have. Method gets the attention because it is the part engineers and consultants are trained to argue about, and the part that survives into the statement of work. Access appears nowhere, because access is not made of technical choices. It is made of other people's standing decisions: a security policy on photography, a maintenance planner's week, an IT standard on guest devices, a supervisor's shift plan, a plant manager's working definition of proprietary. Each of those has a real owner inside your plant. None of those owners was in the room when the study was scoped.

That is why the failure is invisible from both directions. The team on site experiences a sequence of small, individually defensible refusals. Photography policies exist for real reasons. Maintenance has to maintain. IT has a standard because it has been burned before. The person issuing the refusal has no way to know what it cost, because the cost does not land on them and will not be legible for weeks. Meanwhile the team logs a partial week and keeps working, which means nobody in your organization ever reads a line that says two days of observation did not happen. The study arrives looking like a study.

The honest limit on this mechanism matters, because sometimes the access constraint is the right answer. A process may genuinely not be photographable. A line may genuinely not be instrumentable on the schedule you wanted. The claim is not that access should always be granted. The claim is that the decision belongs upstream of the fee and the calendar, because a constraint known in advance changes the method, while the identical constraint discovered on day 2 only removes evidence. Same constraint, two very different costs, and the only variable is when you learned it.

Where the cost lands

Start with the visible cost, which is the smaller one. You paid for days that produced nothing. But 2 of 5 days is not 40 percent of the conclusion. Measurement value is not linear in days, because a week is usually sized to catch specific events: a changeover, a shift handover, a demand peak, the one product that runs badly. Lose the wrong two days and you lose a whole class of evidence rather than two fifths of your confidence. What looks like it should be a partial refund is actually a missing category.

The cost that matters is downstream, in capital allocation. The study exists to support one decision, and that decision is usually invest or defer. A partial baseline does not announce itself as partial. It arrives as a number in a deck, with the same formatting as a complete one. From there, two things can happen and both are expensive. Either the organization funds equipment against a baseline that never observed the conditions actually binding the line, which is capital misallocation with a paper trail, or the people who were on the floor during that week know the data is thin, decline to defend it, and the decision quietly defers for another cycle while margin keeps leaking at the current rate. The second outcome is the common one and the harder one to see, because a deferral is never written down as a loss. It shows up as cost per unit that did not move and a capital request that gets reopened next year.

Then there is the reattempt. You cannot reconstruct an observation you did not make. No analyst recovers the changeover nobody watched, which means the correction is another mobilization, another week of plant disruption, another full fee, and a credibility discount that raises the bar the second study has to clear. Keep the states distinct here. Potential savings, an approved project, an implemented change, a verified result, and a sustained result are five different things. A capped study does not only slow the first of those. It weakens the baseline you would later need to prove the fourth, which is how a program ends up unable to demonstrate value it may well have delivered.

The access list, and who signs it

The fix is unglamorous and it is one page, produced before the fee is committed rather than after the readout. List every observation the design depends on. Next to each, write the access condition it requires and the name of the person who controls that condition.

Photography and video of the specific assets, named asset by asset, not a general permission. Network or cellular coverage adequate for the instruments, confirmed by a connected device during setup rather than by a policy statement in an email. The measurement window, written with an explicit exclusion that no maintenance work touches the instrumented assets while it runs, and the maintenance planner's name on that line. Interviews by role and by person, with accepted calendar invitations, because availability is not a booking. The data room, with one named person who confirms it is complete and a date by which the study team confirms it has actually been read. That last line is nearly free and routinely skipped, which is its own quiet waste: existing data costs nothing to collect twice and often answers part of the question before anyone mobilizes.

Two governance items make that list load bearing instead of decorative. First, ownership. Each line carries one name, and that name is a person inside the plant with authority over the constraint, not the sponsor who promised to go get it. A sponsor's promise is a request queued behind somebody else's priorities. Second, a written stop rule: if an access condition fails, the study team escalates inside that shift, and the sponsor decides the same day whether to clear it, re-scope the method against the new constraint, or stop and re-date the week. The rule exists to convert a silent loss into a decision while the week can still be saved. Without it, every incentive on site pushes the other way, because the team is already there, the clock is already running, and looking productive is cheaper than reporting a block.

Ownership of the list itself is the part that gets dodged. It is not the consultant and not the project manager, because neither one can compel your security, IT, or maintenance functions to move. The operations leader or plant manager who wants the answer owns the access list, and the first real test of a sponsor is whether they will put their name against the lines they do not personally control. If a condition cannot be cleared, learning that early is a gift, because it changes the design while changing the design is still cheap. You can move the instrument, lengthen the window, substitute a manual count, or narrow the question to one the plant will let you answer. Every one of those options is available at day 0 for almost no money. None of them is available at day 2.

The same page gives you the verification artifact nobody usually has. At the readout, every access line reads cleared or failed, with a date. That is the honest denominator under the conclusion, and it is what lets you decide how much weight the number can carry.

What a well run study week looks like

Every access line on the one page list carries a name and a date before the team arrives. Camera and network access are proven during setup by a test capture and a connected device, not by a policy email. The measurement window sits on the maintenance schedule as a no touch block on the named assets. Interviews hold accepted calendar invitations before day 1, including the operator or process lead who actually knows the line. Any failed condition is on the sponsor's desk within the shift it occurred, and the final readout states how many access lines cleared out of how many the design required.

The next decision

Pull the last assessment you paid for, whether it fed a capital request or a cost reduction program. Ask the team two questions: which observations the design called for, and which ones actually happened. If nobody can answer the second question, you are not looking at a baseline you can improve later. You are looking at an unknown denominator underneath a decision you have already made. Settle that before you fund the next study, and make the access list the first page of the next scope, with your name on the lines you control.