Your IT provider hasn’t answered in a week. A critical delivery is running late. How do you avoid these situations? And if they do happen, what do your contracts say when the service falls short?
One answer is to put a few essential commitments in writing, in a service level agreement, or SLA: response times, expected quality, availability, consequences when they are not met. We help you draft these contracts with the help of AI: a process within reach of any SME, to limit unpleasant surprises.
What is an SLA?
A Service Level Agreement (SLA) answers, in writing, two questions nobody asks at signing time:
- What exactly am I buying?
- What happens if I don’t get it?
An SLA typically covers response times, availability, expected quality levels and what the supplier owes when it fails to meet them.
Three situations where an SLA is worth it for an SME
- IT, where “we’re looking into it” can mean two hours or two weeks.
- Recurring services, where quality degrades slowly and quietly: cleaning, maintenance, logistics.
- Any supplier you could not replace in less than a month.
If you tick one of those boxes without an SLA, you have an operational problem that simply hasn’t surfaced yet.
That leaves the objection we hear most often: “We don’t have time to draft an SLA for every supplier.” It is a fair one. AI now makes it possible to answer it, provided you know where its role stops. Here is how we went about it on a real assignment.
The starting point: a chocolate maker without a safety net
During one of our consulting assignments, our procurement consultant sat down with the purchasing director of a well-known chocolate company. The finding was clear: no robust SLA with several key suppliers.
The issue went well beyond compliance. With no service level in writing, there was no objective basis for discussing a late delivery, a non-conforming batch or quality slowly slipping. Every incident was replayed from scratch, verbally, case by case.
The drafting time estimated internally, legal back-and-forth included: several weeks. So nobody had ever taken it on.
Step 1 — A first version drafted by AI
Our consultant used a generative AI tool to produce the first draft of the SLA. He built a structured prompt from our template, where many settle for a “write me an SLA” that never produces anything usable.
In practice, the AI was instructed to:
- take on a specific role: expert in procurement contracts;
- draft an SLA built around measurable KPIs, in this case On-Time Delivery (OTD) and On-Quality Delivery (OQD);
- include penalty clauses in the event of non-compliance;
- cover the essential legal requirements, the environmental, social and governance (ESG) commitments and the dispute resolution procedures.
The difference between an unusable output and a workable draft comes down to four things: a role, a scope, named indicators and explicit constraints. Ask for an SLA without naming OTD and OQD and you will get a text about “service quality”, which is to say nothing enforceable.
Step 2 — Bringing the stakeholders into the loop
An SLA drafted by procurement alone will be an SLA that production cannot meet and that legal will refuse to approve. So we brought quality, marketing, production and legal around the draft.
A series of structured meetings, with a clear mandate for each team: read the text, flag what is missing and what will be impossible to hold on the ground.
An example. The quality department asked for a clause on recurring quality failures. That left “recurring” to be defined, a word nobody had ever put a number on. The answer came during the session: more than 10% of deliveries. A threshold like that is settled among people who know the reality of the flows, and it is what turns an intention into an enforceable clause.
Each piece of feedback was then fed into a new prompt and the document rewritten in real time, during the meeting. Participants saw their contribution appear in the text before the end of the session, which removed the three weeks of email ping-pong that usually weigh this kind of exercise down.
Step 3 — Where we keep control
The AI produced the text; it arbitrated nothing. Four safeguards we apply systematically:
- Legal validation stays human. AI produces an advanced draft. A lawyer must review it before signature, in particular on penalties and termination clauses.
- Numerical thresholds are decided internally. 10%, 48 hours, 99.5% availability: those figures commit your operations. A model you don’t give them to will invent plausible ones.
- Mind the data you enter. Supplier names, prices, volumes: check what your tool does with your data before pasting a contract into it. A consumer account does not offer the same guarantees as a corporate environment.
- Every legal reference cited must be verified. A generative model produces regulatory references that are credible, and sometimes wrong.
The result
A complete SLA template, covering all the issues identified, in a few days instead of several weeks.
The time saved comes less from how fast the AI writes than from the disappearance of intermediate versions. No more “I’ll send that back to you next week”, no more v3_final_REAL.docx. The draft existed from day one and everyone worked on it at the same time.
Above all, the AI acted as a support for internal collaboration, by giving each team a concrete text to challenge from the very first meeting.
What this changes for your procurement processes
Combining an AI-supported process with genuine stakeholder involvement produces three measurable effects:
- Accuracy and compliance: essential clauses are no longer forgotten because whoever was drafting was in a hurry.
- Shorter lead times: days rather than weeks, on a deliverable that is usually postponed indefinitely.
- Stronger collaboration: internal expertise arrives before signature rather than after the incident.
And back to the objection we started with: you don’t need an SLA for every supplier. Start with the ones you couldn’t replace within a month. As a rule, they fit on a single page.
Get our prompts
If you want to see exactly how we went about it, we are sharing all our prompts, the initial version as well as the updates that came out of stakeholder feedback, along with our prompt structure template.
From theory to practice
A prompt template gives you a starting point. Deploying it on your own situations and connecting it to your teams’ needs takes something else.
Our AI training courses for procurement cover that part: building prompts that produce usable results, running a draft-review-improve cycle and bringing in the input of internal teams so the process reflects your organisation’s expertise.
