Administrative assistance
In short: Outsourcing administrative work is not merely a matter of handing it to somebody else: it first has to be made executable by somebody else. That means listing what recurs, writing down rules that until now stayed implicit, saying what may proceed without prior approval, and organising the movement of documents that almost always contain personal data. Whatever legally binds the company stays with the company, and irregular volume never repays the cost of the setup.
Administrative work is the part of the job nobody mentions while it is up to date, and the only one whose lateness is noticed at once. It bills to nobody, it fragments into short occurrences that arrive at the wrong moment, and it follows rules that the person doing it knows without ever having written them down. That last point is what makes outsourcing it less simple than it looks: handing these tasks to an outsider means making explicit what was not, saying what may proceed without approval, and organising the movement of documents that almost always contain personal data. This page describes what outsourced administrative assistance covers and where the line runs between carrying out and committing, how document flows and tracking tools are structured, what can be automated and what calls for judgement, how delegation is set up and then monitored, the mistakes that make it fail, and the situations where it is not justified — because the tasks concerned bind the company, because the volume does not repay the effort, or because what is really being asked for is something else entirely.
What outsourced administrative assistance covers
In short: Administration means every task needed to keep the activity in order and billed to nobody: incoming and outgoing documents, follow-up, preparing files for third parties, filing, retention. Outsourcing transfers execution and never responsibility, and it requires that what was implicit be made explicit first.
The word administration names no trade; it names a remainder — everything that has to be done for the activity to stay in order and that is billed to nobody. It includes handling incoming and outgoing documents: quotes, orders, invoices, receipts, contracts, correspondence; keeping up routine exchanges, appointments, replies, payment chasers; preparing the files that third parties need, whether an accounting professional, an insurer or any other body; filing and retaining whatever must remain findable; and the back room of every sale, from confirming an order through to following its settlement. These tasks have little in common in their content, but they share four properties that determine everything else: they recur, they obey rules nobody has written down, they tolerate lateness badly, and they arrive by interruption rather than at a chosen moment. It is that last property that makes them expensive: what weighs is not their duration, it is that they fragment the rest of the work.
Outsourcing does not change the nature of these tasks, it changes who carries them out — and the distinction between carrying out and deciding is the hinge of the whole subject. An outsider can prepare, draft, file, chase, enter, check, reconcile, keep up to date; the company signs, commits, pays, files declarations and arbitrates. This line does not move with the degree of trust granted: it holds because responsibility itself cannot be delegated. A document prepared by a third party remains a document of the company, issued in its name and enforceable against it; if the preparation is faulty, the company answers for what was issued. This has a practical implication whose weight is usually measured only after the fact: the broader the delegation, the more must have been written down about what may proceed alone and what must be escalated. Delegation without that written line does not produce fluidity, it produces a zone where nobody knows any more who decided.
Three neighbouring roles are regularly confused, and confusing them leads to asking one for what belongs to another. The qualified professional — in accounting, in law, in payroll — produces a regulated output for which they answer; what administrative assistance brings them is complete, filed and consistent material, which makes their work possible without standing in for it. The in-house assistant, next, absorbs the implicit: they know the habits, they see what is going on, they handle the unforeseen without being asked, and they can postpone writing the rules indefinitely since they carry them personally. The external contributor works in the opposite situation: they operate only on what has been made explicit. That difference is not a flaw in the arrangement, it is its condition of existence, and it explains why the work of clarification, optional in the first case, becomes a prerequisite in the second. It is often the occasion to discover that their rules existed nowhere.
Outsourcing takes three forms that do not call for the same things. Clearing an accumulated backlog is finite by construction: it bears on an identifiable stock, its completion can be defined in advance, and it often works as a revealer since it forces pending cases to be settled. Ongoing handling bears on a flow and presupposes stable rules; it is the most demanding form at the start and the most rewarding afterwards, because the cost of the setup is spread over repeated occurrences. Cover during an absence is the most fragile: it is poorly organised after the person concerned has gone, since what they knew is no longer transferable, and it works only if it was prepared while they were still there. Telling these three forms apart avoids a common mistake, which consists in treating a permanent flow with the improvisation acceptable for a one-off catch-up, or conversely in building a full arrangement for an operation that will not recur.
What remains is to say what this service is not. It is not a way of avoiding the decision about how the work should be organised: delegation does not exempt anyone from that decision, it makes it compulsory and makes it visible. It is not a layer of management: no external contributor holds the mandate to decide what binds the company. Nor is it a substitute for a qualified professional, whose involvement remains required wherever it already was. Finally, and this point deserves to be raised in the definition rather than at the end of the page, it is never a neutral activity from the standpoint of confidentiality. Whoever handles a company’s administration sees everything: the names of its customers, the nature of its commitments, its contractual documents, its dealings with suppliers, sometimes material concerning employed persons. That exposure is not a side effect to be managed later, it is a property of the arrangement from the outset, and it governs a large part of what follows.
Document flows, tracking tools, and what calls for judgement
In short: Every administrative task is a document or a piece of information moving through states: entry, qualification, handling, recording, filing, retention. Disorder almost always comes from having too many entry points. A stable rule can be automated; a case that fits no rule calls for judgement, and that is precisely how it is recognised.
Almost every administrative task comes down to an object — a document, or a piece of information — moving through a sequence of states: it arrives, it is qualified, it is handled, it is recorded, it is filed, it is retained for a period and then destroyed. Describing an administrative organisation therefore starts with knowing where things come in. That is often where the disorder sits: one invoice lands in a personal mailbox, another is handed over in person, a third is photographed and sent through a messaging app, a fourth waits in a shared space for somebody to notice it. As long as entry points multiply, no follow-up is possible, since there is no single place where what has arrived can be observed. The first useful move is therefore not to buy tools but to reduce: one entry point per kind of flow, known to senders and handlers alike, and an explicit rule about what happens to whatever arrives elsewhere. Without that, delegating means handing over work nobody can call complete.
Next comes the identity of documents, which decides whether they can be found at all. A naming convention has no aesthetic value: its only function is that two different people file the same item in the same place and find it again without talking to each other. It usually combines the nature of the document, the counterparty concerned and something that allows them to be ordered, and it is worth mainly its own consistency. Two further distinctions are regularly neglected. The first sets the original against its copies: a document exists in one authoritative instance, and every other version circulating in mailboxes, in download folders or as an attachment is a copy whose accumulation creates both confusion and exposure. The second sets the document against the information extracted from it: an amount, a due date, a reference copied into a tracking table then live lives of their own, and the gap between the two widens silently for as long as nobody has decided which one prevails.
A tracking tool, whatever it is, exists to answer three questions: what is in progress, with whom, and what is the next expected action. Anything that does not contribute is decoration. That presupposes that each tracked item carries a state chosen from a short and unambiguous list — received, to be qualified, awaiting a reply, to be approved, closed —, an identified owner, and a next action phrased as an action rather than as a wish. The form matters little: a shared table kept rigorously answers better than a specialised tool nobody updates. What counts is that a change of state be visible on both sides without having to ask for it, because that visibility is what replaces checking-in exchanges. One further requirement is specific to shared work: the trace of what was done, by whom and when. It looks bureaucratic while all goes well; it becomes the only way to reconstruct a situation the day an item is missing or a chaser goes out twice.
Part of this work runs without judgement, and that part deserves to be identified precisely rather than assumed. What can be automated is what obeys a stable rule applied to a regular input: issuing a recurring document from a template, extracting fields from items whose structure does not change, sorting by sender or by nature, triggering a chaser from a state that has not moved, carrying an entry from one tool into another. Three conditions separate useful automation from dangerous automation. The rule must be written before it is programmed, failing which nobody will know any more what the machine does. The inputs must genuinely be stable, because a rule applied to heterogeneous documents produces a regular error instead of an occasional one. And a checkpoint must remain, because automation does not reduce the probability of a mistake, it extends its reach: a wrong rule applies to every occurrence with the same steadiness as a right one, and without complaining.
The rest calls for judgement, and it is recognised by a single sign: the case fits no written rule. A handful of families recur. Qualifying an ambiguous item, where neither what it is nor what it attaches to is clear. The discrepancy, where a document received does not match what had been agreed — a quantity, a reference, an amount that do not line up — and where the decision is to accept, to contest or to ask. Anything that puts a relationship at stake: a chaser addressed to a customer in a delicate position, a reply to a complaint, a refusal to be worded. Anything that binds the company, however simple the gesture looks. And every new case, simply because it is new. In practice, a delegation arrangement is judged not by the refinement of its rules but by the quality of its exception path: what the person meeting an unforeseen case does, whom they turn to, how they obtain an answer, and what is then done with that answer — because a case settled without becoming a rule will come back identical.
How an administrative delegation is set up
In short: List the tasks from what actually arrives, write the rules as decisions rather than as gestures, open named and restricted access, run the arrangement in parallel with review during a first period, then hold checkpoints whose agenda is fixed.
The inventory is not a job description but a census of occurrences. For each recurring task, five elements are enough and none may be missing: what triggers it — a document arriving, a due date, a request —, the expected output, the counterparty concerned, what makes it possible to say it is finished, and the place where it leaves a trace. A sixth element deserves to be added where it exists: the consequence of lateness, since that is what ranks things. This inventory is badly built from memory. What people remember is the nominal case; what consumes the time is the variants. A record of what actually arrives across a full cycle of activity is therefore worth more than an interview, and it almost always produces two discoveries: tasks nobody had taken on, which move only when an incident brings them back to mind, and tasks carried out twice over by people unaware of each other.
Writing the rules means phrasing decisions, not gestures. A useful rule does not describe how to click, it answers the question “what do we do when…”. Three families cover the essentials. Routing rules say where what goes and who handles what. Approval rules say what may proceed without asking, what requires prior agreement, and from whom. Threshold rules say above which level — set by the company, whatever it may be — a case changes category and stops being treated as routine. A set of rules is never complete at the outset, and claiming otherwise is the surest way to make it unusable. What counts is that its holes be visible, that is, that there be a default rule, known and without exception: when no rule applies, you ask. The quality of an arrangement is measured less by the number of cases foreseen than by the reliability of that default behaviour.
Opening access is dealt with at the same moment rather than afterwards, because that is when the decisions are still reversible. Three principles apply. Access is named: one account per person, never a shared login, failing which nothing is traceable and nothing can be withdrawn cleanly when the day comes. Access is limited to what the task requires, and the scope a tool proposes by default is almost always broader than necessary. Document exchanges go through a channel that has been decided, is known, and is the only one, rather than through whatever is at hand at the moment somebody is in a rush. To these three principles is added a decision nobody wants to take at the start and everybody regrets not having taken at the end: what becomes of the documents, the working copies and the access when the engagement stops. Written at the beginning, that clause is a formality; raised at the end, it becomes a negotiation.
Handover works far better when it runs in parallel rather than by immediate substitution. During a first period, the contributor handles and the company reads everything back, not to supervise but to produce the missing material: every correction made must become a line of rule rather than a spoken remark, otherwise the same correction will come round again. It is at that moment, and not before, that the rule set is genuinely written, because real cases reveal choices nobody had ever phrased. Two simple indicators follow that period. The number of questions asked, which should be high at the start and then decline: if it does not decline, the answers are not being capitalised on. And the nature of those questions, which should shift from how the tools work towards the particular cases of the trade. A handover in which no questions are asked is not reassuring: it usually means unforeseen cases are being settled in silence.
Steady state rests on two appointments of different natures that are better not confused. The operational checkpoint bears on the state of the work: what is in progress, what is blocked and why, what is waiting on a decision from the company. Its agenda is fixed, which keeps it short. The periodic review bears on the arrangement itself: what exceptions have appeared, which rules have changed or should change, which access is no longer justified, which tasks have disappeared or been added without anyone deciding so. To that is added a criterion better verified regularly than on the day it is needed: reversibility. The company must be able to take the work back at any time, which presupposes that the rules are written down on its own side, that the documents live in its own space rather than in the contributor’s, and that access can be taken over without reconstruction. An arrangement whose recovery would require an investigation is not delegated, it is out of sight.
What makes an administrative delegation fail
In short: Delegating a task whose rules exist only as a habit; not saying what may proceed without approval; sending personal data through whatever channel is at hand and more of it than needed; not deciding who keeps what or what is deleted at the end; and letting knowledge concentrate again on a single person.
The first mistake is handing over a task whose rules exist nowhere but in the habits of whoever used to do it. The symptom appears quickly: every case produces an exchange, the attention spent answering exceeds what doing the work would have cost, and the company concludes that delegation does not work when what failed was the transfer. The symmetrical mistake exists and costs just as much: setting out to document everything before starting, producing an exhaustive manual written away from any real case, describing imagined situations and omitting the ones that turn up. The practicable path lies between the two. You write the frequent case first, the one covering the largest share of occurrences. You state the default rule for everything else. Then you let each real case add its line, provided somebody is charged with adding it — documentation that names nobody to keep it stops being accurate without warning, and inaccurate documentation is more dangerous than none.
The second mistake concerns approval, and it shows up in two opposite excesses. The first is submitting everything: nothing proceeds without agreement, the person who was meant to be relieved becomes a compulsory checkpoint, and the arrangement produces waiting instead of removing it. The second is checking nothing, and it is more insidious because it starts well. The contributor, wanting to be helpful, settles a case that had not been foreseen; nobody complains since the outcome suits; the scope widens without ever having been widened. The day a decision turns out badly, the question of who took it has no answer, and the discussion that follows damages the relationship far more than the mistake itself. Prevention is written and fits in few lines: what may be done without asking, stated positively; what may never be, stated negatively; and the person to turn to in case of doubt, designated by their function rather than left to be obvious.
The third mistake concerns personal data, and it is the most serious because its effects are not immediately visible. What circulates in an administrative delegation almost always contains some: customer names and contact details, addresses, billing information, bank details, contractual terms, exchanges relating to a dispute, sometimes material concerning employed persons. Two requirements follow and are regularly breached. The first is minimisation: you pass on what the task requires, and no more. Exporting a whole database because it was quicker to produce than an extract is a common breach, and it turns a bounded task into general exposure. The second is the channel: an attachment sent through a consumer service, a document dropped into a shared space whose link never expires, a screenshot forwarded through a personal messaging app are all copies that afterwards escape any control. Obligations exist regarding the protection of personal data; their content depends on the applicable framework and is established by a qualified professional, but none of them accommodates an improvised send.
The fourth mistake extends the previous one and concerns what is left behind. The mechanism is predictable: documents pile up wherever they passed through — in a mailbox, in a downloads folder, in the contributor’s own workspace — and they stay there after the end, without anybody having decided that they would. Four points are settled at the start rather than afterwards. Where the working copies live, and which of them are temporary by nature. Where the authoritative instance lives, which must sit in the company’s own space. What is deleted at the end, and by what means that is verified other than by a promise. And what must on no account be deleted, because retention obligations exist: their duration depends on the applicable framework and on the nature of the documents, it cannot be guessed at, and destroying an item that should have been kept is as much a fault as keeping indefinitely what should not have been. To this is added the classic residue: accounts opened for the engagement and never closed, which outlive it for a long while because closing them was nobody’s job.
The fifth mistake gathers three structural lapses that are often met together. The shared login first: a common mailbox, a password passing through several hands, and it becomes impossible either to know who acted or to withdraw one person’s access without disturbing everyone. Reconcentration next, which is the irony of this service: by dint of being effective, the contributor becomes the only person who knows how the company’s administration really works, and the company has simply moved its dependency rather than reduced it. The remedy is the same as at the start — rules written down on the client side, documents in its own space, reversibility verified now and then. The absence of any reference point last: without a minimal basis for comparison — what is late, what is waiting on a decision, what had to be redone — the discussion about the quality of the work reduces to impressions, and people end up judging the delegation by the amount of time consumed rather than by what is actually up to date. To all this must be added a rule of conduct binding on both sides: a document sent to the wrong recipient, an access left open, an item mislaid are reported immediately, because the delay in saying so always costs more than the incident itself.
When outsourced administrative assistance is not the right answer
In short: Whatever legally binds the company cannot be delegated this way; volume that is too small or too irregular never repays the cost of the setup; work that cannot be described cannot be transferred; some sensitive material should not leave the company; and a permanent need requiring a presence calls for hiring.
The first limit is sharp and cannot be worked around. Acts that bind the company are not delegated in this way: signing a contract, approving a payment, entering into a commitment, filing a declaration, settling a dispute, deciding anything concerning an employed person. An outside contributor may prepare each of these acts up to the last gesture; they cannot perform it in place of whoever answers for it. The confusion arises because tools do not draw that distinction: access to a payment space or to a signature service confers the technical ability to act, whereas responsibility does not follow the access. A company that delegates the ability without bounding its use has not unburdened itself, it has merely made its exposure harder to see. To this must be added regulated outputs, which belong to qualified professionals and for which administrative assistance prepares the material without ever standing in. If the reason to delegate is to no longer have to decide or sign, this service does not meet that need, and no arrangement will make it cover it.
The second limit is volume. The setup carries a cost that does not depend on the quantity of work handed over: you have to take inventory, write rules, open and configure access, read everything back during handover, hold checkpoints. That cost is paid in attention, and it is paid precisely by the person the delegation was meant to relieve; it is amortised only through repetition. Below a certain regularity, it is never amortised: explaining and checking costs more than doing. Two signs indicate the moment has not come. Tasks arise irregularly and differently each time, so that no rule settles. And nothing is genuinely late: the work fits into the gaps without producing any consequence. In that case, what helps lies elsewhere — reducing the number of entry points, scrapping a summary table nobody opens any more, no longer entering the same information in two places — and none of that requires an outside contributor.
The third limit concerns situations where the work does not lend itself to description. If every file calls for an arbitration, if the rules shift depending on who is involved, or if the person holding the task cannot state what they decide and on what criterion, delegation transfers only the typing: the judgement stays where it was, and routing the work through a third party adds a step without removing one. A neighbouring case is harder to hear. It happens that administrative disorder is not the cause but the symptom of a decision never taken: a pricing policy that was never settled, a disagreement with a customer nobody wants to resolve, a division of roles left vague between partners. Handing that disorder to somebody else produces a neatly filed version of the same ambiguity, and the filing sometimes makes the situation harder to unblock, because it stops being visible. The prerequisite here is not a delegation, it is a decision.
The fourth limit is the sensitivity of the material. Some files should not leave: items relating to a dispute in progress, documents concerning employed persons, anything covered by a professional secrecy attached to a particular profession, and more generally anything whose disclosure would cause serious harm to a third party rather than to the company itself. The question to ask is not whether trust exists, but whether the conditions for correct handling are met: a controlled channel, access limited to what is necessary, a basis for passing the material on, a written clause about retention and return, and the ability to establish afterwards what was consulted. Where those conditions cannot be met, the honest conclusion is not to pass the material on anyway while recommending caution: it is to keep that share of the work inside, even if only the rest is delegated. That split is often the right answer, and it is better decided at the outset than discovered after an incident.
The fifth limit is what an outside involvement cannot offer. Part of administrative work is inseparable from a presence: receiving visitors, handling physical items, taking in whatever arrives unannounced, being the point of contact for employed persons, or holding a recognised place in the organisation in order to get a department to respond. None of that transfers remotely, and pretending otherwise disappoints everyone. In the same way, where the need is permanent, substantial and presupposes continuous availability, the answer is a recruitment, and presenting outsourcing as an equivalent amounts to offering something other than what was asked for. There remains the most delicate case: the one where the real request is that somebody else carry the mental load of administration in the owner’s stead. No arrangement produces that relief in full, because what remains — deciding, arbitrating, signing, answering for what has been done — is precisely the share that cannot be delegated. Saying so in advance is better than letting it be discovered.
Frequently asked questions
What can actually be delegated in administration, and what cannot?
Everything that is execution can be delegated: preparing, drafting, filing, entering, reconciling, chasing, keeping a tracker up to date, assembling a file for a third party. What binds the company cannot — signing, approving a payment, entering into a commitment, filing a declaration, settling a dispute, deciding anything concerning an employed person. The line does not depend on the degree of trust but on the fact that responsibility stays with whoever answers for the act. One practical implication follows: granting technical access to a payment or signature tool does not move that line; it has to be set down in writing as approval rules, failing which it moves on its own.
Does administrative assistance replace an accounting or legal professional?
No, and the two roles complement each other rather than overlap. A qualified professional produces a regulated output for which they answer, and their involvement remains required wherever it already was. What administrative assistance brings sits upstream: complete, filed, consistent material available when they need it, which removes a large share of the back-and-forth and of the last-minute reconstruction. The distinction also applies to the questions asked: anything touching the tax, accounting or legal treatment of a situation, reporting obligations, or how long a given type of document must be kept is established by a qualified professional, whose answer varies with the applicable framework and with what the company actually does.
Does everything have to be written down before starting?
No, and trying to write everything in advance is a common way of failing: a manual drafted away from any real case describes imagined situations and omits the ones that turn up. The workable progression has three stages. You write the frequent case first, the one covering the largest share of occurrences. You then state a default rule without exception for everything else: when no rule applies, you ask. You finally let each real case add its line, naming somebody to add it. That last point is the one people forget: documentation nobody is responsible for stops being accurate without warning, and inaccurate documentation is more dangerous than none.
How do you decide what may be done without approval?
By writing it both ways rather than leaving it to judgement. The positive statement lists what may proceed without asking. The negative statement lists what may never proceed, whatever the circumstances. Between the two, threshold rules say above which level — set by the company, whatever it may be — a case changes category. Two excesses are to be avoided. Submitting everything turns the person who was meant to be relieved into a compulsory checkpoint and manufactures waiting. Checking nothing lets the scope widen without anybody having decided it, until the day a decision turns out badly and nobody knows any more who took it. The rule for doubt must designate someone by their function.
How should documents containing personal data be passed on?
Through a channel decided in advance, and by passing on as little as possible. Minimisation comes first: you send the extract the task requires, never a full export because it was quicker to produce. The channel comes next: an exchange space whose access is named and revocable is better than an attachment, than a sharing link with no expiry, or than a screenshot forwarded through a personal messaging app, because each of those forms creates a copy that afterwards escapes control. Two reflexes go with it: checking the recipient before sending, and reporting any misaddressing immediately. Obligations exist regarding the protection of personal data; their content depends on the applicable framework and is established by a qualified professional.
What access should be opened, and how should it be limited?
Named, limited and revocable access. Named: one account per person, never a shared login, without which nothing is traceable and withdrawing one person’s access disturbs everyone. Limited: the scope a tool proposes by default is almost always broader than the task requires, and it has to be restricted explicitly rather than hoped to go unused. Revocable: closure must be achievable from the company’s side, without depending on anybody’s cooperation. It is worth keeping the list of open access and rereading it at the checkpoints, because accounts created for a passing need are exactly the ones nobody thinks to close.
How long must a document be kept?
That depends on its nature and on the applicable framework, and this answer is no evasion: retention periods cannot be deduced from common sense and are established by a qualified professional. What does belong to the organisation is how the rule is made applicable. A period decided after the documents have piled up no longer applies, because nobody knows any more what attaches to what. Deciding upfront the category of each item, where it is kept and when it leaves makes it possible both to find what must be found and to dispose of what no longer has to be kept. The two mistakes are equally bad: destroying an item that should have been kept, and keeping indefinitely what should not have been.
What becomes of documents and access at the end of the engagement?
Whatever was provided for at the start, and nothing else — which is why that clause is written at the outset, when it is a mere formality, rather than at the end, when it becomes a negotiation. Four points belong in it. Return: what comes back to the company and in what usable form. Deletion: which working copies disappear, and how that is verified other than by a promise. Retention: what must not be deleted because an obligation stands in the way. And the closing of access, including whatever was opened along the way for a passing need. To this is added the question of taking the work back: the written rules and the documents must already sit with the company so that it can carry on without reconstruction.
How do you know whether the delegation is having an effect?
By choosing reference points before starting, failing which the discussion reduces to impressions. Depending on the case: what is late and since when, the number of files waiting on a decision from the company, the share of cases that had to be redone, the interval between a document arriving and its being recorded, and the ability to find an item without asking anybody. Two qualitative signals are also worth watching. The questions asked should decline and shift from how the tools work towards the particular cases of the trade. And the reappearance of parallel handling, done outside the arrangement because it was simpler, indicates that a real case is not covered by the rules.
From what volume onwards is setting this up justified?
There is no general threshold, but there is a test that can be put simply: does the same task recur often enough, and closely enough to itself, for a written rule to be used repeatedly? If so, the cost of the inventory, the rules and the handover is spread out and ends up repaying itself. If tasks arise irregularly and differently each time, no rule settles and explaining costs more than doing. A second sign counts just as much: is anything genuinely late, with consequences? Where the work fits into the gaps without producing anything unfortunate, it is often more useful to reduce entry points and to remove what serves nobody than to hand the whole thing to a third party.