What "digital marketing company" actually means in 2026
The category label covers three structurally different things, and conflating them is the main source of buyer disappointment. Knowing which you are actually shortlisting shapes every other question on the list.
| Model | What you are buying | AI's effect on value | When it fits |
|---|---|---|---|
| Full-service agency | Execution headcount: campaign setup, creative production, reporting, account management | AI automates much of what the team billed for; margin pressure is compressing the model | You need a managed service with clear SLAs and do not have internal capability |
| Performance marketing agency | Paid channel execution and optimisation, often ROAS (return on ad spend)-linked | Platform AI (Google PMax, Meta Advantage+) now runs optimisation loops agencies used to run manually | You have a direct-response goal and a budget large enough to exit platform learning phases |
| AI-native marketing consultancy | Diagnostic judgment, AI operating model design, and execution sequencing; AI handles the mechanical work | AI is the engine room, not the competition; the consultancy's value is in what the AI cannot supply | You want the thinking, the measurement, and the outcome architecture, not just a managed retainer |
The "growth marketing agency" label is typically the performance-marketing model above, positioned at startups and scale-ups that want customer acquisition cost (CAC) and lifetime value (LTV) as primary metrics rather than reach or impressions. The mechanics are the same; the vocabulary and the client base differ.
leapbuzz occupies the third row deliberately. The founding thesis is that the time of the agency is functionally over for AI-era marketing problems: Google's Performance Max (PMax) and Meta's Advantage+ do the campaign assembly work; the platform's AI reads the feed, picks the creative variant, and sets the bid. What platforms cannot do is tell you which brief to write, how to measure the result independently of their own reporting, or when to override their recommendation. That is where the consultancy work starts. For more on the agency-versus-consultancy framing, see the companion post on Google Ads agency vs AI consultancy.
Five questions that separate strong candidates from shortlist-padding
Most pitches are structurally similar: platform logos, case study slides with large percentage figures, a team slide, and a proposed retainer. The following five questions cut through that surface and reveal whether the firm has genuine operating depth or is mostly a reseller of platform features.
- Who specifically will own the thinking day-to-day, and what is their background? You want a name, a tenure, and a description of the last three accounts they ran, not a role title like "Senior Performance Manager." The named person's experience should be visible on LinkedIn and should be plausible for your account size. A 50-person agency that assigns a junior account executive to a mid-market account is not delivering the senior judgment it sells in the pitch.
- How do you measure outcomes independently of the platform's reporting? This is the single most differentiating question. Every platform reports its own results in its own attribution model. Google's last-click attribution flatters Google. Meta's view-through attribution flatters Meta. A firm that relies entirely on platform dashboards is not doing the measurement work required in 2026. Look for: Marketing Mix Modelling (MMM), incrementality testing, geo holdout experiments, or at minimum a clear position on which attribution model they use and why it is more conservative than the platform default.
- What happens when the platform's AI recommendation conflicts with your judgment? Google AI Max and Meta Advantage+ make autonomous decisions. Sometimes those decisions are wrong for a specific account or brand. Does the firm have a documented process for overriding platform recommendations? Do they know when to accept the platform's learning and when to intervene? "We trust the platform" is not a sufficient answer. "We apply the platform recommendation when X conditions are met, and we override when Y signal appears" is.
- What is your escalation process when something goes wrong? Response time when a campaign spends incorrectly or a creative goes live in error. Who gets called. What the remediation process looks like. A firm with no documented escalation path is operating on goodwill. That is fine until an error costs you a significant budget day.
- What does a realistic 90-day outcome look like for an account at our stage? The answer should name specific platform learning phases (Google's Smart Bidding requires approximately 30 to 50 conversions to exit the learning phase per ad group), realistic cost per lead (CPL) ranges for your sector and markets, and what the measurement baseline will look like after 30 days. "It depends" as an answer to this question signals the firm either does not know your sector or is hedging on a commitment it cannot make.
These five questions apply equally to a traditional digital marketing agency, a growth marketing agency, and an AI-native consultancy. The quality of the answers is the differentiator. For a broader view of the AI strategy work these conversations should connect to, the services section covers the diagnostic-to-execution arc.
Multi-market signals: what to look for in SG, US, CA, AU, MY
leapbuzz operates across five markets: Singapore, the US, Canada, Australia, and Malaysia. Buyers in those markets often assume a firm that works in one jurisdiction can extend cleanly to another. The assumption is wrong on two counts: regulatory and operational.
Regulatory differences that affect campaign setup directly. Singapore's Personal Data Protection Act (PDPA) governs how you collect and use personal data for marketing, including retargeting audience rules. Malaysia operates under its own PDPA 2010 with different consent mechanics. Canada's Anti-Spam Legislation (CASL) is one of the strictest consent-based frameworks globally: it requires express consent for most commercial electronic messages, not just opt-out. Australia's Spam Act 2003 and Privacy Act 1988 impose requirements on electronic marketing that differ from Singapore's framework in meaningful ways. A firm claiming to manage multi-market campaigns without explicitly naming these frameworks in the pitch has not actually thought through the compliance exposure.
| Market | Primary framework | Consent model | Key implication for paid media |
|---|---|---|---|
| Singapore | PDPA (PDPC.gov.sg) | Opt-out default for most marketing | Custom audience uploads require consent documentation; server-side tagging recommended |
| Canada | CASL (CRTC.gc.ca) | Express consent required for most commercial messages | Lead-nurture email sequences require double opt-in; CRM imports need consent records |
| Australia | Spam Act 2003 (ACMA) | Consent required; unsubscribe mandatory | Remarketing lists sourced from email require consent audit; ACMA enforcement is active |
| Malaysia | PDPA 2010 | Consent required for sensitive data; notice required otherwise | Cross-border data transfers to Singapore require data subject consent in most cases |
| United States | CAN-SPAM; state-level CCPA/CPRA (California) | Opt-out for most commercial email; opt-in for California data sales | California audiences require separate consent handling; lead gen campaigns need clear identification |
Operational differences that affect account performance. Platform ad auction dynamics differ by market. Cost per click (CPC) benchmarks in Singapore are structurally higher than Malaysia for most B2C verticals due to market size and advertiser competition density. Australian and Canadian audiences have different device-mix patterns and different peak conversion windows. A firm running one audience strategy across all five markets is leaving performance on the table. Ask to see how they have structured audiences differently by market in past accounts. The analytics and insights work that supports multi-market measurement is a distinct capability, not a default extension of single-market reporting.
Red flags and walk-away signals
The following patterns appear consistently in pitches from firms that will disappoint. None of them are obvious in a 30-minute meeting without specific questions. They become visible in the details.
- Guaranteed ROAS with no qualifying conditions. Return on ad spend is a platform-reported metric. No external firm can guarantee what a platform will report, because the platform controls the attribution model and the measurement window. A guaranteed ROAS figure in a proposal is either a sign the firm does not understand measurement, or a sign they plan to report a metric that is favourable to them regardless of your actual business outcome. Walk away.
- Platform-badge-led pitches without independent measurement evidence. Google Premier Partner, Meta Business Partner, TikTok Marketing Partner badges indicate platform certification and spend thresholds. They do not indicate the firm measures results independently of those platforms. Ask what independent measurement they use. If the answer is "we use the platform's reporting," the badge is the entire credential.
- No named delivery team in the proposal. If you cannot find the name of the person who will manage your account on LinkedIn with a plausible history, you are buying a staffing allocation, not a named practitioner. The person you meet in the pitch is often not the person who runs the account after onboarding.
- Case studies with outcomes but no methodology. "We grew revenue by 300%" means nothing without knowing the baseline, the attribution model, the time period, and whether the growth was incremental or just measured differently after the engagement started. Ask for the methodology behind any case study number. A firm confident in its work explains the method. A firm hiding behind the number changes the subject.
- Scope creep as the first conversation. A firm that immediately tries to add services to the initial scope before demonstrating anything in the primary scope is prioritising its own revenue over your problem. Start with the defined scope. Expand when there is evidence of performance.
- AI-powered as the primary pitch. "AI-powered" is a Tier-2 slop term with no operational meaning without specifics. Ask which specific AI features are active, what they optimise for, and what the override criteria are. If the answer is a product demonstration rather than an operating position, the AI is a marketing label, not a capability.
The 10-point evaluation scorecard and selection flowchart
Use the scorecard below during or immediately after a pitch. Tick each criterion the candidate firm meets. The tally gives you a qualitative band, not a pass/fail score: no single point is disqualifying by itself, but patterns of low scores in adjacent areas are signals worth discussing before signing.
The flowchart below maps the same 10 criteria into a binary decision path. If you are evaluating more than one candidate simultaneously, run each through the flowchart and compare the exit points.
The flowchart covers the five binary gates. The full 10-point scorecard above adds nuance within each gate: a candidate who passes all five binary gates may still score 6 or 7 out of 10 on the scorecard, which is a reason to proceed with specific written clarifications, not a reason to walk away. The two instruments work together, not as substitutes for each other.
The related post on marketing technology stack decisions covers the infrastructure side of the same evaluation: what tools the firm connects to, how data flows, and what independence you retain over your own measurement environment.
