Three projects.
Same 15:1 guarantee.
Three industries.
Each engagement below started with a 20-minute Clarity Call. The numbers are audited against a baseline both parties signed at the start of the Readiness Sprint. Names anonymised on request - references on file, available after the call.
Extended write-up: E-commerce sales automation case study.
The problem.
Each routine tax memo took an associate 3-4.5 hours to draft, plus partner review. With only 2 partners and 10 staff, the firm was hitting a hard capacity ceiling. Lead times to clients were stretching to 8-9 working days; competitors were quoting 3-4.
The Readiness Sprint identified memo drafting as the highest-payback workflow: high volume, structured inputs (client data + jurisdiction), and a partner review loop already in place to catch errors. Under the firm’s EU AI risk assessment, memo drafting was classified as limited risk — internal advisory tool, partner-supervised.
What we built.
A retrieval-augmented drafting system over the firm's own memo archive (10+ years of internal memos, 2,500+ documents), running on EU-hosted open-weight models. Associates input client facts via a structured form; the system produces a first draft citing internal precedents. Partner reviews and approves before client release.
The numbers.
Cancelled hires: 1 associate (€75K/yr loaded). Lead time to clients: 8-9 days → 3-4 days. Partner review time unchanged - the system doesn't replace judgment, it removes the 3-hour blank-page problem.
“3.5 hours per tax memo became 35 minutes. We avoided the extra associate hire we'd already budgeted for. Quality went up - reviewers see citation-backed drafts now, not someone's 11pm prose.”
The problem.
The agency served 14 mid-market clients on monthly retainers. Each got a monthly performance report across 7 ad platforms plus on-site analytics. Account managers spent the last week of every month assembling decks; the second week of the next month went into defending why the report was late.
Margin per client was eroding - each new tier-2 client needed the same 18 hours of report assembly. The partner's hypothesis: AI could do 80% of it. We sharpened that into a clear rule: "AI does the data assembly and first commentary; humans keep the strategic recommendations."
What we built.
A reporting platform that connects to all 7 ad APIs, normalises metrics, generates a baseline narrative ("CTR fell 12% week-over-week, driven by audience X on platform Y"), and produces an editable slide deck. Account managers spend their time on strategic interpretation and client conversation - not data assembly.
The numbers.
Reports now ship on the contractual date 100% of the time (was 23%). Account managers report higher job satisfaction in the post-rollout survey - they spend more time on the strategy work they were hired for.
“We won two new accounts because we could now serve them at the same margin we used to serve one. The math is brutal - and obvious. Apexa shipped on the date the SoW said. Both of them.”
The problem.
The largest assembly line had 14% unplanned downtime. The key customer had begun applying late‑delivery penalties. The PLC data from the line was being logged to a server in the maintenance office but never analysed; maintenance was strictly reactive.
A large vendor had quoted the company €380K for a predictive‑maintenance programme. The maintenance lead suspected they could start smaller, using the data they already had. The Clarity Call confirmed it - the signals were there, the failure modes were well-understood by veterans, the missing piece was a model that learned the early signatures.
What we built.
An anomaly-detection model trained on 18 months of PLC time-series data, running on a small on-prem server (no cloud, no vendor portal). Maintenance leads get a mobile alert 12-48 hours before a predicted failure, with a recommended part to inspect.
The numbers.
Models, code, and infrastructure all live on the client's premises. Apexa has no ongoing data access. The maintenance team has been trained to retrain the model on new data every quarter.
“Predictive maintenance models we own, running on our PLC data, on a server in our basement. No vendor portal. No subscription. No surprises.”
Yours could be the fourth.
Same 20-minute call all three of these clients started with.
Book your Clarity Call →