Issue deflection 1
52%
of customer issues resolved by self-service bots, rising to 96% on simple queries.
Independent AI / Technical Solutions · 2025 — present
Building AI solutions that actually work for your organization.
I design, build, and deploy AI systems for small businesses and nonprofits — chatbots, automations, integrations — with the full technical stack handled, so you don't need to hire a team to get started.
01The three ways
Three things I build, implement, and ship.
Chatbot & FAQ systems — built, integrated, and running on your stack
I design, build, and deploy custom chatbots trained on your content — for support, internal knowledge bases, and customer-facing FAQ. I handle the full stack: the architecture, the integration into your existing tools, the evaluation, and the handoff.
- Pulling in what you already know. Your existing documents, FAQs, and policies, turned into something a bot can actually query.
- Getting accurate answers. Making sure the bot doesn't hallucinate, and knows when to escalate.
- Testing before it goes live. Evaluation runs and quality checks before your customers see it.
- Knowing when to hand off to a human. Escalation paths built in, so nothing important falls through the cracks.
- Ongoing observability. Dashboards and logs so you can see what's being asked — and catch problems before they become habits.
What well-built can bots do
Independent
research
research
Response time 2
55%
average reduction in first-response time across CX teams using AI tooling.
CSAT lift 3
+40%
of mid-sized businesses report a 40%+ jump in CSAT within 3 months.
Resolution time 4
11min → 2
Klarna's AI assistant cut average resolution from 11 minutes to ~2.
◦ Sources
- 1YourGPT, AI Customer Service Statistics 2026 — self-service bots resolve 54% of issues, up to 96% on simple queries.
- 2Freshworks CX Benchmark Report 2025 — 55% reduction in average first-response time for CX teams.
- 3YourGPT 2026 — 70% of mid-sized businesses see 40%+ CSAT jump within 3 months of AI adoption.
- 4NexGen Cloud / Klarna case study — average resolution time dropped from 11 min to ~2 min.
Technical scoping — what to build, what it costs, and what's actually worth the effort.
A four-week technical scoping engagement for organizations ready to build something real. I get into your data, your tools and systems, your security posture, and your team's actual capacity — and deliver a technical build plan with a prioritized roadmap and a clear list of what to stop doing before you start. Works for small teams. You don't need a technical staff for this — that's kind of the point.
- Data. What you have, what's clean, and what an honest baseline looks like.
- Your tools and systems. What you're running on, what it can actually support, and what the integration complexity looks like.
- Org & risk. Who owns what when something the AI built goes wrong in front of a customer.
- Roadmap. Three things to build in 90 days. One to actually ship in the next few months.
Sample readiness scorecard
Data hygiene
7.2
Platform & infra
8.1
Observability
5.5
Risk & review
3.8
Talent & ownership
6.6
Composite5.9 / 10
Verdict: cautiously ready. Here's what to build first — and what to skip.
Nonprofit AI implementation — small budgets, one real pilot, no platform lock-in.
Nonprofits are being sold AI tools that require maintaining. I work with executive directors and program leads to scope, build, and run one actual pilot — designed for your staff capacity, your budget, and the people you serve — without the platform commitments that require a full-time engineer to keep running.
- Figuring out what's actually worth building. And which programs are quietly risky or not worth the engineering overhead.
- What you're being sold. Plain-language technical assessment of what's actually in the contract.
- Donor-ready documentation. A short written summary the board and funders can read and actually understand.
- One scoped pilot. Built, measured, and designed to be reversible. No platform commitments you'll be stuck maintaining.
Field note · Compass Pro Bono · 2024-present
In the 2024 engagement with a four-person volunteer team, the work was a Salesforce integration that eliminated four of every five manual data-entry steps for a Philadelphia non-profit. The 2025 engagement, as Deputy Project Leader for the Children's Scholarship Fund of Philadelphia, is a bigger challenge: scope, timelines, client communication, deliverables. The model or the integration is the cheap part. The honest conversation about what staff time gets reclaimed is the work.
Pro bono · Compass
— pk
02How an engagement runs
Short, written, and honest about uncertainty.
Intake
A technical discovery call
We talk through the problem, what you've tried, and what a realistic build looks like. If it's not the right fit, I'll say so and point you somewhere better.
Listen
Technical scoping
I dig into your existing stack, data, and tools. Two to five conversations with the people closest to the problem. I'm trying to understand how the systems actually work, not how they look on paper.
Draft
Build & iterate
You get working software early — not a polished document that misses the point. We test it against real questions, adjust, and keep going until it reflects what you actually need.
Hand off
Deploy + hand off
You get a running system you can actually use — along with documentation your team can maintain. One final walkthrough. After that, it's yours. No retainers, no ongoing dependency on me to keep it working.
◦ open for Q3 2026
Not sure what to build first? That's the most common starting point. Send me two paragraphs and I'll tell you what's actually buildable.
A short note is enough to start. Two paragraphs about the org, the problem you're trying to solve, and a rough timeline. Replies are direct and in plain language. I'll tell you what's buildable, what it'll take, and whether it's the right fit.
BasedPhiladelphia, PA
Formatbuilt, at your pace
Statusopen · Q3 2026
© 2026 Pratik Kubal · Independent AI / Technical Solutions