Portfolios tell you less than you think
Every agency portfolio shows the best work, presented at its most flattering, often without saying what the agency actually did. The site you're admiring may have been designed elsewhere, or built by a team that has since left, or shipped in a state substantially different from the case study screenshots.
More useful: ask to see a project that went badly and what they did about it. Every firm with real history has one. The ones who answer candidly are telling you how they'll behave when your project hits trouble — which it will, because they all do.
Also ask what they turned down recently and why. A partner willing to decline work is a partner who'll tell you when your idea needs changing. One who has never turned anything down is optimising for booking, not for outcomes.
Questions that actually predict the outcome
'What will you need from us, and when?' A good answer is specific and slightly uncomfortable — content by week two, a decision-maker available for weekly reviews, access to your systems by a named date. Vague answers here mean the delay will later be framed as your fault.
'What happens when we disagree about a design decision?' You're listening for a process, not for deference. A partner who always defers to the client produces incoherent work; one who never does is unpleasant to work with.
'Who exactly will be working on this?' Agencies commonly sell with senior people and deliver with junior ones. Ask for names, ask to meet them, and ask what proportion of their time you're getting.
'What does your handover include?' Documentation, credentials, training, a support window and a repository you own. If any of these are extra, you want to know before signing rather than after.
Warning signs worth walking away from
Guaranteed rankings. Nobody controls Google's index. Anyone promising page one for competitive terms in a fixed timeframe is either targeting keywords with no volume or misrepresenting what they can deliver.
A firm price without discovery. A precise quote off a two-paragraph brief means assumptions have been made that you'll discover as change requests, or the scope has been quietly trimmed to fit.
No questions about your business. If the entire conversation is about features and none of it is about how you make money, who your customers are, or what happens if this fails, you're buying deliverables rather than outcomes.
Ownership ambiguity. Any arrangement where leaving means losing your site, your domain or your data is a commercial risk that no price justifies. Get ownership in writing.
Pressure to sign quickly. Legitimate firms have pipelines and can wait a week. Artificial urgency in a services sale is almost always a tactic.
Structuring the engagement so it can survive problems
Prefer milestones over a single deliverable. Each milestone should produce something functional you can evaluate — not a status report. This gives you an exit point and gives them a forcing function.
Agree what 'done' means before starting, in measurable terms. Performance targets, browser support, accessibility standard, what's in scope and explicitly what isn't. Ambiguity here is where most disputes originate.
Start small if you can. A discovery phase or a single-module build tells you more about whether you can work together than any number of reference calls, and it caps your exposure if the answer is no.
Finally, weigh communication quality heavily. Over a three-month project you'll spend more time in conversation than looking at deliverables. A technically excellent team you can't get a straight answer from is a worse experience than a merely good team who tell you the truth promptly.