pixel · of · software

Case Study · Pro Bono

Krakow Education Foundation: AI Strategy Consultation Pro Bono

Pro bono AI strategy consultation for a Krakow-based education foundation: scoped use case, vendor recommendation, governance plan in 6 weeks.

Client

Krakow education foundation (anonymized at client's request)

Foundation · Education non-profit

Team size

8 staff, 25 active volunteers

6 week engagement

Volunteer screening time per applicant

4 hours → 1.5 hours (projected after build)

Context#

Q4 2025 cohort of our Pro Bono dla Polskiego Biznesu program. The applicant: a Krakow-based education foundation running a tutoring program that matches volunteer university students with secondary-school students from underserved backgrounds.

Their challenge: volunteer screening takes ~4 hours per applicant (background check verification, motivation assessment, scheduling matching, training enrollment). The foundation’s two paid coordinators handle ~40 volunteer applications per quarter. They were turning away applicants because they couldn’t process them fast enough — limiting program reach.

They asked: could AI help, and if so, where specifically?

Engagement scope#

The pro bono engagement was a 6-week AI strategy consultation, not a build. Specifically:

The pro bono program does not include build work. We’re explicit about this in the application guidelines: consulting that de-risks an investment decision is a different deliverable than implementation.

Diagnosis#

We ran a 4-week diagnostic. Key questions:

  1. Where exactly does the 4 hours go? Time-tracking with the two coordinators showed:

    • 1.5 hours: reading and synthesizing application essays + cross-referencing with motivation interview
    • 1.0 hour: scheduling matching (volunteer availability vs. student availability)
    • 1.0 hour: background check verification (with external service, much of it wait time)
    • 0.5 hours: training enrollment + record-keeping

  2. Which steps are AI-tractable? We assessed each step:

    • Essay synthesis: High AI-tractability. Structured summary plus risk-flag identification is exactly what LLMs do well with proper grounding.
    • Scheduling matching: High AI-tractability, but solvable equally well by simpler scheduling tools.
    • Background check verification: No AI value. This is wait-time on an external service.
    • Training enrollment: Low AI value. Mostly data entry; better solved with form automation.

  3. Where would AI break governance? The foundation handles minor data (under-18 students). EU AI Act and Polish data protection rules apply. We identified three guardrails any AI workflow would need: no minor data in prompts, all outputs requiring human approval before action, audit log of every AI-touch on a volunteer record.

Recommendation#

Our written report (delivered in week 5) recommended:

Scope AI to one workflow only — application essay synthesis#

Building an AI assistant that:

Projected effect: 1.5 hours of essay synthesis becomes ~30 minutes of human review. 2.5 hours saved per applicant. ~100 hours saved per quarter at current volume.

Solve scheduling with non-AI tools#

Calendly + a simple matching script. Estimated build: 1-2 weeks of volunteer-engineer time (not paid engineering). Cost: zero incremental on tools they already use.

Don’t AI-touch background checks or training enrollment#

Wait time and data entry respectively. AI adds nothing. Investing AI budget here would be performative.

Vendor and architecture sketch#

We recommended Anthropic Claude Sonnet 4.6 as the LLM, primarily for: stronger refusal behavior on ambiguous applications (foundation needs the assistant to escalate gracefully), better structured-output reliability for the citation-grounded summary, and clearer EU AI Act alignment documentation.

Architecture (sketch):

Estimated build cost (volunteer-engineer time, not paid): 40-60 hours.

Governance plan#

We delivered a 6-page governance plan covering:

What happens next#

The foundation has an internal volunteer engineer who is starting the build in May 2026. We’ve offered (free of charge) two follow-up touchpoints during the build: an architecture review at the 50% mark and a governance review before go-live. This is consistent with our pro bono program — consulting de-risks the decision; implementation is the foundation’s work.

Why we publish this case study#

Two reasons:

  1. Other Polish non-profits with similar challenges should know this is a tractable path. The recipe (essay synthesis with strict grounding and human approval) generalizes to many non-profit screening workflows.
  2. Showing what pro bono actually delivers keeps the program honest. Our pro bono is a 6-week strategy engagement, not a free build. The foundation knows; future applicants should know.

This case study is published with the foundation’s written consent. The foundation is anonymized at their request — they cited concerns about being approached by other consultants offering paid alternatives.

Apply for the next pro bono cycle#

Q3 2026 nabór opens July 1. Polish foundations, NGOs, social cooperatives and qualifying education organizations: Pro Bono dla Polskiego Biznesu.