How to Evaluate a B2B Agency: Capability, Team, Process and Terms

Capability: what the agency can actually do this quarter

Agency websites describe ambitions as much as capabilities. A service list is a statement of what the agency is willing to sell, not proof of what it has recently delivered. The distinction matters most in fast-moving disciplines, where a capability that was real two years ago may now rest on one person who has since left.

The most reliable question is narrow: which of these services did you deliver in the last quarter, for how many clients, and who did the work? Recent, repeated delivery is a far better predictor than a decade-old flagship case study. Follow it with a request to see something live — a site, a dashboard, a campaign structure, a repository — rather than a slide about it.

Be equally alert to breadth that looks impressive. An agency of twenty people advertising fifteen disciplines is subcontracting most of them. Subcontracting is not a problem in itself, but it changes accountability, margin and communication speed, and you should know which parts of your engagement it applies to before you sign.

Evidence: what counts and what does not

Case studies are marketing artefacts. That does not make them worthless, but it does mean you should read them for the parts that are hard to fake: the starting position, the constraints, the specific interventions and the timeline. A case study that reports a percentage uplift with no baseline, no timeframe and no named client is decoration.

Team: who does the work, and how much of them you get

The most common source of disappointment in agency relationships is not incompetence. It is the gap between the team that pitched and the team that delivers. Senior people win the work; mid-level people deliver it. That is a normal and often sensible model, but only if you know the ratio before you commit.

Process: how work actually gets done

Process is where agencies differ most and describe themselves least clearly. Every agency claims to be collaborative, agile and transparent. Ask instead for the mechanics, and listen for whether the answer sounds like a description of something that exists or an improvisation.

• Onboarding: what the first thirty days look like, and what they need from you in week one
• Planning: how priorities are set, who signs them off and how often they are revisited
• Reporting: format, frequency, who attends and what happens when a number is bad
• Quality control: the review steps between a specialist finishing work and you seeing it
• Escalation: the named path when something slips, and the response time attached to it
• Tooling: where work is tracked, and whether you get access or a summary

References: ask better questions

Agencies supply their happiest clients as references, which is expected. You can still extract signal by asking questions that a coached reference cannot answer with a generality. Focus on change, difficulty and recovery.

1. 1What did the agency change about how your business operates?
2. 2What was the hardest part of working together in the first three months?
3. 3Tell me about a time they got something wrong — how did they handle it?
4. 4Did the people who pitched stay on your account for the full engagement?
5. 5How did reporting evolve once the initial enthusiasm wore off?
6. 6Would you hire them again for the same scope at the same price?

Commercials, ownership and exit terms

Contract terms are where a good relationship is protected and a bad one is survived. The core items are ownership of work product, ownership of accounts and data, notice periods, and the handover obligation at the end. Every one of them is cheap to agree at the start and expensive to argue about later.

Insist that advertising accounts, analytics properties, cloud environments, domains and repositories are created in your name and merely granted to the agency. This single decision removes most of the leverage that makes ending an agency relationship painful. Good agencies are entirely comfortable with it; hesitation here is itself a finding.

Finally, agree a review point. A six-month checkpoint with pre-agreed criteria turns an uncomfortable conversation into a scheduled one, and gives both sides an honest opportunity to adjust scope rather than quietly drift.