When Systems Cannot Produce the Answer
Organisations are increasingly required to answer for themselves, and their systems cannot produce the answer.
A sustainability officer at a listed manufacturer opens a reporting template. One line asks for water consumption at a named plant, for the financial year, in kilolitres, assured.
The plant is metered. So are the other eleven. Some meters serve a single line, others a whole block. One was replaced in March, and its predecessor's readings sit in a register the maintenance team retired. Readings live in logbooks at the gate, in a utilities spreadsheet, in the maintenance system, and in tanker delivery notes filed by accounts. There was a fortnight in August when a line was down, and consumption fell, which the shift supervisor remembers and no record explains.
Every number needed is somewhere in the company. Producing one defensible figure means knowing which meter covers what, which readings can be trusted, how to treat the fortnight, and who can confirm any of it. That knowledge is distributed across a dozen people and written down nowhere.
This is the ordinary shape of a problem that is arriving more often every year.
Four rooms, one structure
A regulator asks a listed company for a disclosure that a third party will assure, on a metric the company has measured in several places and reconciled in none.
A retail buyer asks a supplier to demonstrate how chemical storage is handled at every site, against a code revised in March, with documentary evidence per requirement.
A customer asks for a quotation on a configuration close to two quoted last year. The substitutions accepted then, and the rates applied, sit in the estimator's judgement.
A clinical protocol asks what a field worker concluded during a household visit, in a structure a trial can audit, rather than the free text she typed on her phone.
Four sectors, four counterparties, four different consequences for getting it wrong. Underneath, they are the same problem. Somebody outside the organisation wants a structured, defensible response, in a fixed format, on a fixed cycle. The organisation holds the raw material and cannot assemble it, because assembling it requires knowledge about the material that the systems do not carry.
What has changed is frequency. These requests used to arrive occasionally and were absorbed by somebody senior working late. They now arrive continuously, from regulators, from customers enforcing supplier codes, from auditors applying assurance standards, and from buyers who have professionalised their own procurement. Organisations are being asked to account for themselves at a cadence their predecessors never faced.
Where the answer actually lives
Enterprise systems were built to record events. A purchase order was raised. A batch was released. A meter was read. Two decades of implementation made those records reliable and auditable, which was a real achievement and still pays for itself.
What no system holds is the map. Which meter serves which line. Which register superseded which. Why the March replacement matters and how to bridge it. Which of two conflicting readings the utilities engineer would stand behind, and why.
The same gap appears in a different guise on the commercial side. A system knows a material was accepted. The reasoning that connected the input to the outcome was never captured, because no field existed to hold it and nobody was asked. It was accepted because the supplier had cleared an equivalent specification the previous quarter and the incoming inspector recognised the lot number. That sits in one man's head and leaves when he does.
So an organisation facing an external request is rarely missing data. It is missing the interpretive layer that turns scattered records into one figure somebody will sign. That layer is held by people, in fragments, and has never been asked for in a form anyone could reuse.
The assumption the category makes
Most software built for compliance and reporting begins from a different premise: that the answer sits in a system, and the work is connecting to it.
That premise produces connectors, warehouses, integration layers, and dashboards aggregating what has already been captured. Where it holds, the tools do their job well, and a fair amount of reporting genuinely is an aggregation exercise.
It holds for part of what is now being asked. For the rest, connecting to the systems returns twelve meter readings and no basis for choosing among them. The pipe delivers everything except the thing that would let you use it.
Where that happens, the problem changes character. If the interpretive layer exists only in people, then obtaining it is an act of elicitation, and elicitation is a different discipline from integration.
From integration to interview
Getting a reliable account out of somebody who holds it is harder than querying a database, which either has the row or does not.
Consider what a capable interviewer does that a form cannot. She sequences her questions, because the answer to the third changes what the fifth should be. She recognises when a reply is a working estimate rather than a measured fact, and marks it as one, because "roughly forty" carries an assumption that matters more than the number. She notices when somebody is describing the official process instead of the one they follow, which is the most common failure in any operational conversation and is almost never deliberate. She probes a vague answer rather than recording it, and knows which vagueness is worth pursuing and which is honest uncertainty that should be captured as uncertainty.
She also knows when to stop. Pushing a maintenance supervisor toward a figure he cannot support yields a number that will fail assurance and costs the relationship that made the conversation possible in the first place. Recognising which questions cannot be closed, and routing those to whoever is accountable for signing, is part of doing the work properly.
None of this is a technical capability. It is a communication discipline with a century of practice behind it in survey research, clinical history-taking and investigative work, and it has a literature. Building a system that conducts such an interview means encoding that discipline: the sequencing, the probing, the separation of a stated fact from an inferred one, the judgement to escalate.
It also means going to where people work. A shift supervisor will not open a portal to answer forty questions, and any design that requires him to will collect blank fields. The conversation has to happen on the channel he already uses, in the language he thinks in, in the minutes he has between other things. That constraint shapes the architecture more than any reporting requirement does.
This is the harder half of the problem, and it calls for a skill the category has not been hiring for.
What the account has to carry
A figure assembled this way will be examined by somebody who was not there when it was put together. That imposes requirements beyond arithmetic.
It needs its basis travelling with it: what it was derived from, which assumptions were applied, who supplied each component, on what date, and what was excluded. An assured disclosure is worth what its trail is worth, and building the trail afterwards is close to impossible.
The trail has to accumulate while the work happens. Anything reconstructed at the end is a reconstruction, and a reviewer can tell. This is the same pressure that has moved governance ahead of capability in enterprise buying decisions, and it applies to reporting with particular force.
And the questions that could not be closed have to reach a person, marked open, with the reason attached. A system that fills a gap with a plausible number has not finished the job. It has moved the problem to the point where an auditor will find it.
There is a return here that organisations rarely anticipate. Doing this properly the first time yields structured operational knowledge the company never previously held: which meter covers what, how the estimator substitutes, which requirements bind which site, where every piece of evidence sits. It was assembled to satisfy an outside party. Its more durable value is internal, and it compounds on the next cycle, when the map already exists.
A category, not a vertical
It is tempting to file all of this under compliance, since that is where obligations are densest and deadlines hardest.
The same structure turns up well outside it. A quotation resting on judgement exercised before and recorded nowhere is this problem in commercial dress. So is an underwriting file, a tender response, a clinical record, a supplier qualification pack. In each case somebody outside wants a structured, defensible account, and the systems hold events.
The category is the production of institutional answers. Compliance sits inside it and is currently the loudest room, which is why most of the building has happened there. Naming the whole subject after it would be naming a category after the first place anyone noticed it.
What we are still working out
How much of the interview can be conducted well without a person present is genuinely unsettled. We are confident about the mechanical parts: sequencing, probing a thin reply, separating a stated fact from an inferred one, escalating what cannot be closed. We are less sure about the judgement an experienced auditor brings to a conversation that is going wrong in a way nobody has yet articulated. Our current answer keeps a person in the path at that point, and we hold it as a working position.
We are also watching the obligations themselves. Much of the present wave assumes a human respondent somewhere in the chain. If assurance standards start demanding evidence that no person mediated, which is not far-fetched, some of what we have built will need rethinking. Better to say that now than to find out during an audit.
In closing
The useful question inside your own organisation is narrower than which reporting platform to buy. Of the accounts you are now obliged to give, how many can be assembled from records alone, and how many need somebody to explain what the records mean?
For the second group, the work is reconstruction rather than connection: conducted through conversation, assembled with its basis attached, and handed to whoever has to sign. More organisations face that work every year, and most are doing it by hand, late, and with the people they can least afford to tie up.
—
Mitochondria is an agentic AI product company based in Amsterdam and Pune. ISO 27001:2022 certified. Designed to fall within the limited and minimal risk tiers of the EU AI Act, with controls aligned to the GDPR, UK GDPR and India's DPDP Act.