Most conversations I have with principal investigators end at the same place. The PI already knows what the group needs, already knows what it costs, and thinks the price is reasonable. Then comes the sentence: "fine, but I have no idea how we would actually buy this." That is not a price objection. Nobody has ever explained to them which administrative route a service like this takes, and asking the finance office feels slightly embarrassing after fifteen years of only ever ordering reagents.
This is the explanation that is missing. It is not legal advice, and procurement rules differ by country, by funder and by institution, so treat what follows as the map you take to your own research office rather than a substitute for it. That is usually enough: the process is shorter than most PIs fear.
First: you are buying a service, not hiring a person
This is the misunderstanding that costs the most time. If you walk into the administration saying "I want to hire a statistician", the conversation goes straight to human resources: a post, a call, a selection panel, months. Rightly so, because that is what hiring someone means.
What most groups actually need is not a person on the payroll but a service: the analysis of a specific dataset, a methodological review before submission, a response to reviewers, a workshop. That is a procurement of services, and it runs through a completely different door, with a different procedure and, almost always, a different calendar. Weeks instead of months.
That said, some departments genuinely need the post. If your unit produces methodological work every week, handles clinical data that cannot leave the building and also has to teach methods, a staff methodologist beats any external arrangement. I go through that option and its real drawbacks in methodological support in a department: what works and what does not.
Every system has a direct-award threshold. Find yours.
Public procurement is built the same way almost everywhere: above a certain value you must run a competitive procedure, and below it the buyer can award the contract directly to a supplier of their choice with a much lighter file. The number and the paperwork differ, but the shape is the same, and the single most useful thing you can do this week is find out what your number is. Ask your research office or finance office this exact question: what is the value below which we can award a service contract directly, and what does the file need to contain?
One concrete example, so the idea is not abstract. In Spain, article 118 of Law 9/2017 on public sector contracts sets the threshold for a "minor contract" of services at 15,000 euros of estimated value, excluding tax. Below that, the award is direct: no tender, no bidding documents. The file still exists, but it is short — a report from the contracting body justifying the need, approval of the expenditure, and the invoice — and article 29.8 of the same law caps such contracts at one year with no extension, which is exactly why methodological support tends to be sold as a fixed annual package rather than an open-ended hourly arrangement.
Your country will have its own version of that number. What travels across systems is the logic: a fixed, defensible annual amount that fits under the direct-award threshold is far easier to buy than an open budget for hours. The annual bundles for research groups, research lines and departments on this site are sized for exactly that reason, and a pro forma invoice is available on request so your administrator can open the file without a meeting first.
What your finance office will actually ask for
The details vary; the list barely does. If you arrive with these, the file opens in an afternoon:
- A one-paragraph justification. What is being bought, why the group needs it, and why this supplier. No prose required: "statistical analysis and methodological review of the manuscripts of project X during 2027; this supplier is chosen for specialization in quantitative methods applied to health research."
- A quotation or pro forma invoice with the supplier's tax ID, description, amount, taxes and a validity date. This is the document that is missing most often and the easiest one to ask for.
- Supplier onboarding if the provider has never invoiced your institution before. It is a one-off form and then the supplier is in the system.
- A declaration that the supplier is eligible to contract with the public sector, and in many places evidence of being up to date with tax and social security obligations.
- An invoice at the end, in whatever electronic format your institution requires. Ask which one before the work starts; a supplier who cannot issue it in your format creates a problem at the worst possible moment.
One tax detail that surprises people who have never bought a service: if the supplier is an individual rather than a company, withholding may apply, and the amount leaving your budget line will not match the amount the supplier receives. It does not change the cost to the group, but it does change the numbers your administrator sees, and it is better said in advance than discovered as a discrepancy.
Which budget line pays
Almost always one of these four, and naming the right one is what unblocks the conversation fastest:
- The external services or technical assistance line of a grant. This is the natural route. It exists precisely for this: buying in specialized work the team cannot do at the quality or in the time it needs. Watch one real limit: many funders cap subcontracting as a percentage of the total award, and some require prior authorization. Check your own grant agreement before committing the amount, because that is the most common reason a perfectly legitimate cost gets rejected at reporting.
- Departmental or institutional research funds. Usually more flexible and with fewer eligibility strings, and the obvious route when the spend benefits several projects at once.
- The training budget — of the department, the doctoral school or the research office — when what you are buying is a workshop or a seminar. Different budget, different process, usually faster, and the reason so many institutions start there before committing to anything larger.
- Year-end underspend. Not elegant, but it is the reality of the last quarter everywhere. If your institution closes the financial year on a given date, a file opened two weeks before it will not make it. Ask your administrator for the exact cut-off; it is the single piece of information that most often decides whether this happens this year or next.
The mistake that gets the file sent back: splitting
Worth being blunt about, because it is the one thing that turns an ordinary purchase into an audit finding. Splitting a single need into two smaller contracts with the same supplier and the same object, purely to stay under a threshold, is exactly what procurement rules are written to catch, and control bodies detect it without effort. In the Spanish example above, the law explicitly requires the buyer to justify that the object of the contract has not been altered to avoid the thresholds.
If your real need exceeds the threshold, the answer is not to slice it. It is either to run the competitive procedure — longer, more paperwork, but it is the road — or to size the service honestly to what a direct award can cover, state in the justification what is left out, and buy the rest next year as a genuinely different object.
What to agree before you sign anything
Your finance office will not look at any of this, and it is what prevents the most trouble later:
- Scope in writing. How many manuscripts, how many sessions, what is in and what is out. "Statistical support" on its own is an inexhaustible source of misunderstanding.
- Turnaround. A service that replies within 24 working hours and one that replies in two weeks are not the same product, even at the same price.
- Authorship. This is the delicate one and it has to be settled first. If the person analyzing your data helps design the analysis, interprets it and drafts part of the manuscript, they may well meet the authorship criteria used by most journals in psychology and biomedicine — the ICMJE recommendations are the usual reference point. If the role is service work (executing a plan you defined, checking, formatting), acknowledgment is the correct place. Neither answer is wrong. What causes damage is discovering that you never discussed it once the manuscript is written.
- Confidentiality and personal data. If participant-level data is going to be shared, a confidentiality agreement is not enough on its own: you need a data processing agreement, and you should check what happens to the data when the work ends. In many cases identifiable data does not need to be shared at all, and that should be your default request.
- Reproducible syntax as a deliverable. Ask for the code, not just the results. If a reviewer asks you next year to rerun the model with one more covariate and you do not have the script, you bought an output instead of a capability.
Before you open a file, spend half an hour
A piece of advice that does not suit me commercially and that I will give anyway: before moving any paperwork, take the group's most advanced manuscript and put it through a pre-submission review. If the objections that come back are cosmetic — badly built tables, results without effect sizes, references that do not check out — that is an hour of work and you do not need to buy anything. If what comes back is substantive — the design does not support the question, the model is the wrong one, the sample will not carry the estimate — then you do have a problem money can solve, and you now know exactly how to describe it in the justification.
That is what the free AI paper reviewer is for: it returns the objections of two simulated reviewers and an estimated editorial decision. For the mechanical layer, the reference checker verifies your citations one by one against Crossref, PubMed, OpenAlex and Semantic Scholar. Neither replaces a methodologist, but both tell you in half an hour whether you need one.
A three-sentence email to your administrator
If you have read this far, this is the thing to write. Copy it, adapt it, send it:
Hi — I would like to buy an external service for statistical analysis and methodological review for the group, charged to [grant / departmental funds], for [X] per year plus tax. My understanding is that this amount falls under a direct award rather than a competitive procedure. Could you confirm what documentation you need from your side, and the latest date I can open the file in this financial year? The supplier's pro forma invoice is attached.
Two things make that email work. It already contains the amount and the pro forma, so your administrator does not need to ask you for anything before starting. And it asks about the deadline, which is the piece of information only they have and the one that actually decides the outcome.
When not to buy anything
To finish where I started, honestly. External support makes sense when the group has peaks of methodological work it cannot absorb, when the person currently doing all of it is the PI, or when there is a design decision beyond the team and getting it wrong costs a year of data collection. It makes no sense at all if what is missing is writing time, if the real problem is that nobody reads drafts before they are submitted — that is fixed by an internal rule, for free — or if you are about to start collecting data and the research question is still vague. In that last case what you need is not a statistician: it is a two-hour meeting.
And if what you want before deciding is a sense of what the current rate of rejections is costing you, I ran that calculation separately in what a rejection really costs at group scale.
References
- International Committee of Medical Journal Editors. (2025). Recommendations for the conduct, reporting, editing, and publication of scholarly work in medical journals. https://www.icmje.org/recommendations/
- Ley 9/2017, de 8 de noviembre, de Contratos del Sector Público [Spanish Public Sector Contracts Act]. Boletín Oficial del Estado, 272. https://www.boe.es/buscar/act.php?id=BOE-A-2017-12902 (articles 29.8 and 118, cited as one country's worked example)