How to Choose the Right IT Outsourcing Company: 12 Questions to Ask in 2026/27

Where this sits: Start at IT Outsourcing. Diligence pages: Data Security & Compliance (DPA, transfer, Article 27) and Guarantees (baseline, cadence, remedy). Lateral: Questions to Ask an Outsourcing Company.

Who is the legal employer — and why does it matter?

Ask who carries employment liability, payroll and attrition — if you do, it is not outsourcing. In ZI Model 01, every deployed person is full-time on ZI’s payroll, working exclusively for you. ZI carries recruitment, infrastructure, supervision, QA, daily reporting and replacement; you never become the employer. One managed service fee. Agencies and EORs fail this test: agencies place and leave; EORs provide payroll rails while you manage everything else — “they make it legal; we make it work.” A staffing agency that cannot show supervision and attrition cover in writing is selling a placement, not a workforce. Expect the contracting line to name ZeusInfinity Services (trading as ZI Workforce) — the registered entity — not a brand with no registration. That clarity closes the largest trust gap before it opens.

How do you hire — and what is the hit rate?

Demand the hiring mechanism, not the deployment promise. 10–14 days from brief sign-off is now category-normal (Hyvid, Acelerar, parallelstaff publish similar windows); lead with the six-stage process. Ask: sourcing channels, work-sample tasks, assessed share, supervisor sign-off, probation and attrition history — and the stated accuracy (ZI: 70–80%). A vendor that hides stages behind speed is optimising for placement, not fit. Beware “dedicated VA team” language in comparison tables — it self-identifies as the category ZI says it is not. The right answer sounds like a system: pipeline → screening → skills task → stakeholder review → offer → supervised QA and daily reporting from day one.

How will I see the work every day?

If visibility is a manager, it will not survive scale — ask for a system. ZI’s 3-Layer Transparency Framework is the cost control and the trust control: Layer 1 productivity monitoring (active/idle, app use, screen snapshots), Layer 2 daily reporting (start/end logs), Layer 3 supervisor oversight (internal QA and output review). Ask for a live demo of the daily log and the QA review, not a screenshot. First-party proof that a system matters: ZI internal daily reporting time fell from 60 minutes per person to under 10 minutes via a structured prompt sequence — no tool built — while verification tooling held bounce below 2% with zero subscription cost (ZI Operations, 2026). Those numbers exist because the work was measured and the bottleneck removed with the smallest thing that works.

How do you measure improvement — and who decides success?

Measurement must have a baseline, a method and a cadence — otherwise “AI-enabled” is vocabulary. ZI’s baseline order of preference: (1) client work logs/hours from your PM tool, (2) if logs are absent, live measurement in opening weeks, or high-level estimates stated as estimates — never repackaged as measurement. Method: list every manual operational activity → client stakeholders prioritise the 2–3 bottlenecks that unblock the most downstream work → research per item with a tech lead (automate / upskill / prompt-system / small utility / execution fix) → build smallest thing that works, prompts first. Review at 90 days, then quarterly, each quarter resetting to the improved position — the first 25% is the easiest 25%. Ask who prioritises: if the vendor does not require stakeholder participation in discovery and implementation, it is a handoff, not a partnership — and that partnership requirement is stated openly as non-negotiable.

On why 25%+ is framed as a target with terms: the UK productivity data on structured management and digital adoption shows modest but consistent gains accrue when measurement is systematic —ONS Business Insights and Management Expectations data and McKinsey’s state of AI 2024 both converge on the same lesson: systematic implementation beats experimentation (ONS BICS 2024–25, ONS; McKinsey, 13 Mar 2024, The state of AI). Without a measured baseline, no vendor can responsibly quote a guarantee.

What is the guarantee — with remedy?

A commitment without a remedy is a target — ask for the remedy clause. ZI commits to a documented 90-day outcome review and quarterly cadence; the remedy for missing the 25%+ is discussed upfront (options presented: continue at no fee, service credit, penalty-free exit, or a combination) and written into terms — see Guarantees. That page is gated on a founder decision precisely because the guarantee is the one page type no competitor in any studied category publishes. If a vendor cannot show baseline source, measurement method, review cadence, exclusions (department-head access, scope change, volume vs efficiency) and a remedy, treat any percentage as a headline, not a commitment. This is also where cost language discipline matters: the evaluative saving is 30–40% lower all-in (≈65% of 1.55× UK true cost); “up to 80%” belongs only in attention-stage ads, never in a guarantee.

What are the data security and GDPR controls?

Data never leaving your control is the claim — paperwork and routing must prove it. ZI’s posture: NDA + DPA signed before day one; professionals work inside your systems; data never leaves your control — the same model trusted by accounting/legal firms for 15+ years. Ask a prospective vendor for: named legal entity and contracting party, processing locations and sub-processors, UK GDPR transfer mechanism post-Schrems II, Article 27 UK representative, ICO registration, retention and deletion policy, access controls and audit trail — all dated. The ZI public site is explicit that the registered entity is ZeusInfinity Services (ZI Workforce is a trading name, to be registered as volume grows) and that the Data Security page carries the full mechanism — if that page is thin, pause. A good signal is a vendor that volunteers its limits (healthcare/legal flags, the need for controls by message two) rather than claiming blanket coverage.

What else belongs in the 12 (IP, tooling, attrition, refs, phase)?

Seven more questions turn a pitch into diligence:

1. Who owns IP? Model 01 workforce tooling is ZI’s IP, delivered as a service (benefit without build cost); Model 02/03 builds are client’s IP entirely. Clients’ data is always client’s. Get it in one sentence.
2. What is the tooling philosophy? Prompts first; small utilities where they pay — not SaaS resale. Ask for one owned build and its operating cost.
3. What happens on attrition? Replacement guarantee — ZI covers re-recruitment at no fee, no vacancy tax. Ask for time-to-replace data.
4. Can I speak to a client in my shape? One named client is not proof for every vertical — ZI labels cross-domain proof or illustrative scenarios plainly; ask how thin proof is handled.
5. Why does this role sit offshore? The vendor should name tasks and systems (Jira, GitHub, Figma, Xero/Shopify, GA4) and the bottleneck this role removes by month three — not generic “efficiency.”
6. Is this phased or aggressive? Phased (one/two areas) is default; Aggressive parallels workstreams at higher velocity but same process. Delivery mode should be explicit.
7. What does “what disappears” look like? The vendor should describe what you stop doing — reporting drag, chase lists, backlog — at founder level, not just at operator level.

How do the answers compare across vendor types?

Use the table as your scoring sheet. Score 0–2 per row; any vendor scoring <10 is labour arbitrage, not an AI-enabled workforce partner — the interview is over regardless of rate.