Lünendonk's study IT-Sourcing-Trends 2026/2027, published on 30 September 2026, asked more than 150 IT and sourcing managers at mid-sized and large companies how AI changes the way they buy IT services. Seventy percent of them are already negotiating discounts on time-and-material or FTE-based pricing as their use of AI grows. The study covers IT service providers, not freelancers. My view is that the same question will reach every hourly line on a purchase order, contractors included, and that the hourly rate survives while the hours get questioned.
I work as an independent firewall and network-security contractor in German enterprises, so I read the study from the seat that bills by the hour. This post does not repeat the rate data from what twenty years of experience is worth per hour, or the consultancy figures from the German consulting market is sorting, not shrinking. It looks at the billing model itself, which is what the buyers in this study now want to change.
What did the Lünendonk IT-Sourcing study find?
The Lünendonk IT-Sourcing study found that AI is changing the contract terms for IT services, not just how the work gets done. According to Lünendonk's release, the survey covered more than 150 IT and sourcing decision-makers from mid-sized and large companies across industries, plus an analysis of 36 IT sourcing consultancies. Ganowski also summarised the results in a guest article for IT-BUSINESS, and GIT SICHERHEIT reported on the study.
| Finding (share of respondents) | Value |
|---|---|
| Expect concrete proposals from their IT providers on how generative AI improves efficiency and quality | 90% |
| Gradually adjusting contracts with IT providers to regulate the use of generative AI explicitly | 80% |
| Already negotiating discounts on time-and-material or FTE-based pricing as AI use increases | 70% |
| Expect more outcome-based models | 77% |
| Expect time and material to lose importance as a billing model | 61% |
| Expect managed services to grow in importance | 95% |
| Expect more AI-supported, highly automated delivery models | 87% |
Source for every row: Lünendonk, IT-Sourcing-Trends 2026/2027, release of 30 September 2026.
Tobias Ganowski, Senior Consultant at Lünendonk, put it this way in the release: “KI verändert nicht nur die Produktivität der Leistungserbringung, sondern auch die über Jahrzehnte etablierten wirtschaftlichen Mechanismen im IT-Sourcing.” In English: AI changes the economic mechanics of IT sourcing that have held for decades, not only the productivity of the delivery. He adds that the open question is how customers and providers turn efficiency gains, risks and investments into fair contract and pricing models.
Does this apply to freelancers at all?
Not directly. The respondents describe their relationships with IT service providers and managed-service partners. Nobody in the study was asked about freelance contractors. I still think the finding reaches us, for a plain reason that is my reading rather than the study's: the procurement team that negotiates an AI discount with a systems integrator uses the same purchase-order template for a firewall contractor. Once a buyer has asked one supplier “how much of this did a tool do?”, the question is on the checklist for every hourly line.
The rate data already hinted at this. The freelancer rate surveys I looked at in September showed stable hourly rates alongside a growing share of freelancers with low annual revenue. Lünendonk now shows the mechanism from the buyer's side. In my view, buyers will not cut the rate first. They will ask why a task still takes the hours it took before.
Will AI lower the hourly rate itself?
The Lünendonk study does not say that hourly rates fall. It says buyers negotiate discounts on time-and-material and FTE-based models when AI is in use, and that 61% expect time and material to lose importance as a billing model. Those are two different pressures. A discount on an hourly model is a correction inside the old logic. A shift to outcome-based models, which 77% expect, replaces the logic. My reading is that the second one matters more for contractors, because a rate negotiation is something we already know how to have. A conversation about what a finished piece of firewall work is worth, with no hours attached, is new for most of us, and for most purchasing departments too.
Where does AI actually cut hours in firewall and security work?
In my experience, AI cuts hours in firewall work where the job is reading and writing, and does almost nothing where the job is getting a decision. Rule-base analysis, documentation and the first draft of a change request now go much faster with a model doing the reading. The slow parts of a firewall project were never those. They are the conversations with application owners and the sign-off that someone accepts the risk, which I described in why freelance projects stall on the client side.
| Task in a firewall project | Effect of AI (my assessment) | Why |
|---|---|---|
| Analysing a rule base for shadowed, unused or overly broad rules | Large cut | Pattern work on structured config, easy to check afterwards |
| Documenting flows and writing change requests | Large cut | Drafting text from known facts |
| Translating configuration between vendors during a migration | Partial cut | Fast first draft, but every object and NAT rule still needs review |
| Confirming with application owners which flows they still need | No real cut | The bottleneck is the owner's time, not the analysis |
| Accepting the risk of removing a flow | None | A person with authority has to sign |
| Cutover night and rollback decision | None | Someone has to be accountable at 2 a.m. |
For contractors who bill analysis time, that is uncomfortable. If a buyer pays by the hour for work a model now drafts in minutes, the 70% in Lünendonk's study are right to ask for a discount. The hours that remain are the ones with accountability attached, and those are hard to put on a timesheet.
What does a security contractor sell when hours stop being the unit?
A security contractor sells accountability for a defined result once hours stop being the unit. Lünendonk's own advice to providers points the same way: standardise services, scale what can be automated and align responsibility with measurable results. For a firewall contractor, I would translate that into work packages with a definition of done, for example:
- A recertified rule base for one segment, with every remaining rule mapped to an owner and an evidence file an auditor can read.
- A migrated policy set, tested against agreed traffic, with a documented rollback and a fixed cutover window.
- A decision pack: the rules proposed for removal, the risk of each, and a signature line for the client's risk owner.
Priced per package, the efficiency gain from AI stays with whoever does the work, and the client pays for a result it can check. In my view that is a better deal for both sides than an hourly rate with a negotiated AI discount, which charges the client for hours nobody needed and still leaves the contractor arguing about each one. The same logic applies to AI use inside the client: the people closest to the work often adopt it faster than the governance around them, as I wrote in shadow AI: employees hiding their AI use.
Does an outcome contract change the sham self-employment risk?
An outcome contract changes the legal framing, though not automatically the risk. German law describes employment through two indicators. § 7(1) SGB IV names work under instructions and integration into the instructing party's organisation as signs of employment. § 1(1) AÜG uses the same two tests for temporary agency work: a worker is placed with a client if integrated into its work organisation and subject to its instructions. A contract for work under § 631 BGB, by contrast, obliges the contractor to produce a promised result, which may also be a result achieved through a service.
My reading, and not legal advice: a package with a defined result fits the contract-for-work logic better than an open-ended hourly engagement where the client hands out tasks each morning. But the paper alone does not decide. If the contractor still sits in the team's daily stand-up and takes instructions on how to do the work, the label on the contract will not help much. The reform debate I covered in Statusfeststellung reform: legal certainty at 16.74% of the fee runs in parallel. Clients pushing for outcome contracts for cost reasons may find they also get a cleaner separation from the contractor, provided they actually work that way.
What should a firewall contractor do now?
A firewall contractor should change how the work is defined before a buyer forces it through a discount. Three steps follow from the study, as I read it:
- Use AI openly for the reading and drafting work, and say so in the offer. 90% of buyers in the study expect providers to show how generative AI improves the service. Hiding it invites the discount conversation later.
- Offer at least one package with a definition of done next to the hourly option, so the client can compare.
- Keep the accountable steps visible: risk acceptance, cutover, rollback. That is the work the client cannot buy from a tool.
The governance pattern is the same one that decides whether AI projects work at all, covered in why enterprise AI pilots fail on governance: the constraint is usually ownership. Lünendonk's 70% describe buyers who have understood that AI changes what an hour is worth. In my view, contractors who sell accountability for results will keep their rate, and those who keep selling hours will spend more of each engagement defending them.