Free Competency Framework Template | Role-Based Examples & Proficiency Levels
Free Competency Framework Template | Role-Based Examples & Proficiency Levels
Can you connect CV screening to softgarden? Yes, in three ways: softgarden's own AI-assisted applicant matching, softgarden's AI Screening Assistant, or an external screening module that docks onto softgarden's interfaces. All three are optional and in all three a person makes the final call. Which one fits depends on whether you need a ranking of applications or a documented, explainable assessment that your works council, your hiring managers and, if it comes to it, a court can follow.
This page compares the three routes, walks through how an external module connects to softgarden, works out the time saved on 500 applications, and sets out the legal position (German AGG, GDPR Art. 22, EU AI Act, works council) as of 18 October 2026. Where we cannot prove something, we say so. Sprad is one of the external vendors and therefore not neutral. softgarden facts come from softgarden's own documentation, Sprad facts from sprad.io.
What are the three ways to add AI screening to softgarden?
The differences are less about technology than about what you hold at the end: a marker, a criteria list, or a report with evidence.
| Route | What you get | Prerequisite | Limit |
|---|---|---|---|
| 1. AI-assisted applicant matching (softgarden) | Matching applications get a frame and icon in the lists; you can filter by match | CV parsing (Textkernel) active in the Marketplace and the feature switched on in settings; only applications received after activation | Highlights matches, the explanation sits in a hint on the icon. softgarden does not disclose in detail how the match is computed |
| 2. AI Screening Assistant (softgarden) | Criteria generated from the job description, each marked full, partial or no match, plus an overall score; bulk moves and email templates | Optional, off by default, activated in your account. softgarden does not publish terms | Assesses application documents, no conversation with applicants; according to softgarden the language model runs on EU servers |
| 3. External screening module (e.g. Sprad) | Assessment against your own criteria with evidence and a report; depending on the vendor also interview and candidate portal | Connection to softgarden through its interfaces, activation by softgarden support | How much is written back (fields, status) differs by vendor and project and has to be clarified up front |
If a prioritised list is enough and you do not want a second vendor, start with route 1 or 2. An external module pays off when you want to talk to applicants (chat, voice), build your own talent pool, or need every assessment backed by sources.
How does an external CV screening tool connect to softgarden?
softgarden documents third-party connections publicly. Partners can list themselves in the softgarden Marketplace, and the integration partner page (German) describes levels from an API connection to deep system integration. Two interfaces matter for screening vendors:
- The Assessment API: the vendor provides four endpoints (list tests, invite a candidate, fetch the result, delete candidate data) that softgarden calls. Access runs on a bearer token that the customer enters in their softgarden settings, and setup goes through softgarden support.
- The Apply API: it lets third parties create applications in softgarden. It is not for reading candidates and not for building your own application forms (that is what the Frontend API is for).
What we did not find in softgarden's public documentation (as of 18 October 2026) is an open interface that writes scores into arbitrary candidate fields, or webhooks. Anyone who promises that should be able to show it on your tenant. Another screening vendor publicly documents a pattern that is common in practice: moving a candidate into a defined pipeline phase starts the export, softgarden support enables the vendor, and the job description may have to be supplied separately.
What Sprad documents: on its screening page Sprad lists softgarden among the ATS it works with and says the finished report lands in your existing system or on your own webhook. Which fields or statuses are written in your softgarden tenant is not stated. Clarify that in the setup call before you start, and get the answer in writing.
How does the screening work, step by step?
At Sprad the flow is not "CV in, score out" but a conversation with evidence. This is how Sprad describes it on its product page; the details depend on your per-role configuration.
- An application arrives in softgarden. Depending on setup, a phase change or an interface call starts the flow.
- Invitation. Applicants get a personal link, no password or signup, by email, web, WhatsApp or phone call.
- Chat instead of a form. The building blocks (basics, questions with follow-ups, documents, interview, booking) can be required, optional or off per role.
- Anonymisation before scoring. A fixed rule removes name, email, phone, age and salutations; a second model pass removes origin, gender, family status, health and appearance (per Sprad's AI ethics page).
- Scoring against your criteria. Every criterion is scored three times independently and the median counts. Each score points to the quote it came from.
- Report to the ATS. One scorecard per person with criteria, quotes and transcript.
- A person decides. According to Sprad nothing is rejected automatically; rejections are drafted with reasons and sent by a person.
What gets written back, and how transparent are the criteria?
Transparency here means you set the criteria and every score traces back to a statement. A scorecard typically contains:
| Component | Content | Why it matters |
|---|---|---|
| Criterion | Set by you per role, e.g. process ownership, tooling depth, language level | A clear yardstick the works council can review |
| Rating | Level per criterion, median of three runs | One outlier does not change the result |
| Evidence | Verbatim quote from CV or conversation, with its source | Traceable for hiring managers and applicants |
| Open points | Criteria without enough evidence are flagged as an open question, not guessed | Targeted interview questions |
In practice, which part appears as a field, tag or note in softgarden and which only in the report is a configuration question. Test with real applications whether your recruiters see the result where they work.
How much time does it save on 500 applications, and what does it cost?
Worked example with marked assumptions, not measured: one role, 500 applications. Assumption manual: 6 minutes per CV including a note. Assumption with screening: the recruiter reviews the top 125 applications (25%) from the scorecard at 4 minutes each and pulls 40 spot checks from the other 375 at 2 minutes each, so that no misjudgement goes unnoticed.
| Minutes per CV (manual) | Manual effort | Effort with screening | Saving |
|---|---|---|---|
| 3 | 25.0 h | 9.7 h | 15.3 h |
| 6 (assumption) | 50.0 h | 9.7 h | 40.3 h |
| 10 | 83.3 h | 9.7 h | 73.6 h |
With screening: 125 x 4 min + 40 x 2 min = 580 min = 9.7 h. The saving is mostly a question of your current minutes per CV. At an assumed fully loaded cost of 45 EUR per hour, the middle row is worth about 1,800 EUR.
Sprad cost per the pricing page (net, as of 6 October 2026): CV screening, talent agent and chat interviews cost 0.49 EUR per application on the flat rate, so 245 EUR for 500 applications. If you only want the CV assessed, pay by credits: 2 credits per assessment, which at the 0.10 EUR list price per credit is 100 EUR for 500. Voice and video interviews cost extra. softgarden does not publish prices for its modules; you have to ask.
Is AI screening allowed under the AGG and GDPR?
Yes, if a person genuinely decides and you can avoid and evidence the absence of discrimination. Three points apply today, independent of the AI Act:
- AGG (German General Equal Treatment Act): the prohibition of discrimination (section 1) covers applicants, and under section 22 you must prove there was no discrimination once indications are shown. So record criteria, evidence and decisions. Removing names, gender or origin before scoring lowers the risk but does not eliminate it. Sprad does not claim bias-freedom and reports for its own test set that 97.6% of the detectable gender signal was removed, the remainder not proven to be zero.
- GDPR Art. 22: people have the right not to be subject to solely automated decisions with significant effect (Regulation (EU) 2016/679). Under the CJEU's SCHUFA ruling (C-634/21, 7 December 2023) a score can itself be a decision if the recipient relies on it decisively. Rubber-stamping scores is therefore not enough. The review has to be real.
- Information: applicants must be told that and how AI is used. Say so in the invitation and the privacy notice.
Does the EU AI Act cover CV screening, and from when?
Yes, as a high-risk system. Annex III point 4(a) lists AI systems used to analyse and filter job applications and to evaluate candidates. After the Digital Omnibus the Annex III obligations apply only from 2 December 2027 instead of 2 August 2026, according to Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026. If you still read sources saying "from 2 August 2026", they show the old position.
For deployers (employers) the obligations include human oversight, use as instructed by the provider, keeping logs, and informing workers' representatives and affected people. The details are in the AI Act (Regulation (EU) 2024/1689). Also ask your vendor how it will meet its own duties until then. This page is not legal advice.
Does the works council have to co-determine?
Usually yes, and involving it early saves months. Two provisions of the German Works Constitution Act matter:
- Section 95 BetrVG: selection guidelines for hiring require the works council's consent, and since the reform this expressly applies where AI is used to draw them up (section 95(2a)). From 500 employees the works council can demand such guidelines.
- Section 87(1) no. 6 BetrVG: co-determination for technical systems that can monitor employees' behaviour or performance. That concerns your recruiters more than applicants, where the system logs their use.
A document with five answers helps: which data is processed, which criteria apply, what is written where, who decides, how long it is stored. Sprad lists works council documentation as part of its offer. Bring that document to the first meeting with the works council.
What mistakes happen most often at the start?
- Criteria copied from the job ad. Separate must-haves from nice-to-haves, or whoever uses the most keywords wins.
- No parallel test. Re-score 50 to 100 closed applications and compare with the original selection.
- Result sits in the wrong tool. If recruiters have to hunt for the report, they will not use it.
- Works council too late. Concept first, then pilot, not the other way round.
- Skipping spot checks. Check monthly among the low-scored whether good people slip through.
When does an external module fit, and when is softgarden alone enough?
An external module fits if you stay with softgarden, get many applications per role, want conversations with applicants and must document every assessment. It fits less if you hire rarely, a simple marker is enough, or you do not want another processor. In that case check softgarden's own features first. For the difference between genuine assessment and a mere score display, read Candidate Matching Software 2026. How Sprad turns every applicant into a talent pool is shown on the Talent Base page.
FAQ
Does this replace softgarden?
No. softgarden stays your applicant management system; per Sprad the module runs on top and nothing is migrated.
Where do recruiters see the results?
Per Sprad in your existing ATS or via your own webhook. Which fields are used on your tenant you clarify before starting.
Is the screening automatically free of discrimination?
No. Removing identity signals lowers the risk, a residual risk remains. Document criteria and review results regularly.
May AI reject applicants?
Not solely automated if the rejection has significant effect (GDPR Art. 22). A person must review and decide.
What does screening cost at Sprad?
0.49 EUR per application on the flat rate (screening, talent agent, chat interviews) or 2 credits per assessment. As of 6 October 2026, net, see pricing.
What about AI-generated applications?
Flawless documents say little. Evidence-based methods therefore weigh statements in the conversation above CV wording and flag anything unclear as an open question.
Jürgen Ulbrich
Jürgen Ulbrich has more than a decade of experience in developing and leading high-performing teams and companies. As an expert in employee referral programs as well as feedback and performance processes, Jürgen has helped over 100 organizations optimize their talent acquisition and development strategies.
