Incrementality Testing for Influencer Marketing — 2026
·9 min read·By the Anga team
Many brands in Nairobi and across Kenya run influencer campaigns and measure vanity metrics — views, likes, or reach — then hope sales follow. Incrementality testing (also called holdout testing) answers a clearer question: did the influencer activity cause additional sales or conversions, above what would have happened anyway?
Who this guide is for
Marketing managers and founders running e-commerce, retail, or service campaigns in Kenya and Africa who want to measure true sales lift from creators. You will get step-by-step instructions, sample-size examples using Kenyan numbers (KES), tracking options that work with M-Pesa and local shops, analysis approaches that are realistic for SMEs, and practical next steps.
Quick overview: what is incrementality testing?
Incrementality testing compares outcomes between a treatment group exposed to your influencer campaign and a control (holdout) group that is not exposed. The difference in conversions (sales, leads, sign-ups) estimates the campaign's causal impact — the incremental lift.
Step 1 — Define your goal and primary KPI
Pick a single primary KPI: e.g., paid online purchases, catalogue orders, or store visits. For Nairobi e-commerce use paid conversions; for FMCG test redemption of unique codes at Naivas or local kiosks.
Decide the minimum uplift you'll act on (Minimum Detectable Effect, MDE). Example: a KES 200 product with 1% baseline conversion — you might set MDE = 20% relative lift (0.2 percentage points absolute).
Set the test window (commonly 2–6 weeks). Longer windows help capture delayed purchases but risk contamination.
Step 2 — Choose your test type
Pick one that fits your campaign and operations. Common types:
Creator-level (randomized) holdout: Randomly assign creators to treatment or control. Control creators do not post campaign content. Best when you control creator activation (useful on marketplaces like Anga).
Audience-level holdout: Use ad platforms to exclude a percent of matched audiences from seeing paid amplification of creator content (works when you run creator content as ads).
Geo holdout: Run the campaign in some counties/townships and hold out others. Good if you can localize creator selection and distribution (e.g., run in Nairobi & Mombasa, hold out Kisumu).
Code or link holdout: Give unique coupon codes or UTM-tagged links to treatment creators but not control. Track redemptions or conversions by code/link.
Each has pros and cons. Geo holdouts are operationally easy but have fewer clusters (counties) so statistical power can suffer. Creator-level holds are efficient if you can recruit many creators — a strength of marketplaces like join Anga.
Approximate sample per arm — tens of thousands of users. A realistic back-of-envelope: ~80,000 visitors per group (so 160,000 total).
Practical Kenyan example: if your paid site conversion is 1% and your campaign reaches 200,000 people with an average click-through rate producing 40,000 visits, you may be underpowered. Instead, increase conversions (improve landing page), increase reach (activate more creators on Anga), or aim for a larger MDE (e.g., 50% uplift) to lower required sample sizes.
If you need a calculator, use Google Sheets or simple online power calculators and input your p0, p1, α and power. For help on tagging and UTM structure that aids analysis, see our step-by-step guide to UTM parameters for influencer marketing 2026.
Step 4 — Tracking setup (the backbone)
Choose tracking methods that match sales channels used by your Kenyan customers:
Online purchases: Use GA4 e-commerce events, platform events (Shopify/WooCommerce), and server-side tracking where possible. Fire browser pixels and Conversions API events for Facebook and TikTok to reduce data loss on mobile networks common in Kenya.
Coupon codes: Issue unique codes per creator or per test arm. Example: NAIVAS30-OG or NAIVAS30-CNTRL. Codes work offline and online and are easy to reconcile at POS or on Jumia.
Payment references (M-Pesa): For direct sales where customers pay with M-Pesa Paybill, ask customers to use a specific reference or include a short code to trace campaign conversions.
UTM links: Ensure every creator uses UTM parameters. Follow our guide: UTM parameters for influencer marketing 2026. Use medium=influencer, source=creatorname, campaign=campaignName.
Phone & WhatsApp leads: Track via unique contact numbers or WhatsApp click-to-chat links with campaign UTM querystrings; export leads to a CRM or spreadsheet for matching.
Step 5 — Running the test (operational steps)
Randomize and assign creators or geos to treatment/control before briefing. If you use creator-level holdout, recruit at least N creators per arm (N depends on sample size and average reach per creator).
Brief creators in treatment to use specific creative and CTAs; control creators either do not post or post a neutral content that doesn't promote the offer (depends on test type).
Distribute unique coupon codes/links to creators. Store codes in a single spreadsheet and track redemptions daily.
Launch concurrently. Staggering can bias results due to seasonality (weekend vs weekday behaviour in Nairobi). Coordinate posts via WhatsApp threads or Anga's messaging — creators on Anga prefer WhatsApp-first communication and mobile-friendly briefs.
Monitor early signals but wait until pre-registered sample and time-window complete before analysis.
Step 6 — Analysis (measure the lift)
Primary approach: compare conversion rates between treatment and control. Use confidence intervals and statistical tests appropriate for proportions.
Estimate absolute and relative lift: (p_treatment - p_control) and (p_treatment/p_control - 1).
Use a two-proportion z-test or chi-square test for significance. For clustered designs (geo/creator clusters) use cluster-robust standard errors or a difference-in-differences regression.
Report: sample sizes, conversion rates, absolute lift, relative lift, p-value, and 95% confidence interval for lift.
Example result: control conversion 1.00%, treatment 1.30%. Absolute lift +0.30 p.p., relative lift +30%, p = 0.04 — statistically significant at 5%.
If you have low power, consider Bayesian analysis or present the effect size with CIs and treat as directional evidence for further testing.
Step 7 — Action steps after the test
If lift is positive & significant: scale the creator mix that drove the effect. Use Anga to activate more nano and micro creators in the same regions or similar audiences — everyday creators often outperform a single celebrity because of authenticity and cost efficiency.
If lift is null but confidence intervals include meaningful increases: iterate. Improve landing pages, creative CTAs, or test larger MDEs.
If negative or harmful: stop the campaign, investigate contamination (control saw content), creative fatigue, or mis-specified tracking.
Practical Kenyan tips and pitfalls
Mobile data and load times matter. Design landing pages that load quickly on 2G/3G. Large video-heavy pages reduce conversions.
M-Pesa is central. Make payment flows simple and trackable with clear payment references.
Watch out for coupon sharing across test arms. Use per-creator codes or single-use codes and reconcile regularly.
For retail activations (Naivas or local kiosks), train POS staff to ask where the customer heard about the offer, and capture that in a simple tally sheet or form.
Local festivals, match days, or Safaricom promotions can confound results; avoid running tests during major national events unless you stratify by them.
Tools & resources (that work in Kenya)
Google Analytics 4 (e-commerce events).
Facebook/Meta Conversions API and TikTok pixel (server-side helps on mobile).
Shopify, WooCommerce, Jumia seller dashboards for sales data.
Spreadsheet-based trackers and simple pivot tables — many Kenyan SMEs rely on them.
Marketplaces like join Anga to recruit many verified local creators, manage payments (escrow), and scale holdouts by controlling which creators you activate. Anga supports mobile-money payouts (M-Pesa) and WhatsApp-first coordination.
Example: a Nairobi skincare brand sells a face oil at KES 1,200 (~USD 8). Baseline conversion is 0.8%. They want at least 40% relative uplift. Using creator-level holdout with 80 creators (40 treatment, 40 control) and issuing each creator a unique code redeemable online, they collected 120 conversions in treatment and 75 in control over 4 weeks. Treatment conversion = 120/8,000 = 1.5%; control = 75/8,000 = 0.94%; absolute lift +0.56 p.p., relative lift +60% — significant. They scaled by onboarding 200 more creators via Anga and increased ad amplification on top-performing creator videos.
Wrap-up
Incrementality testing for influencer marketing in Kenya is doable and vital for sensible budget allocation. Plan your KPI, choose a practical holdout design, size for the effect you care about, set robust tracking (M-Pesa and UTM-friendly), and analyse with transparent statistics. Use local operational workflows — WhatsApp coordination, mobile-first landing pages, and marketplaces like Anga to scale creator activation and payment securely.
Ready to recruit verified Kenyan creators and run a proper holdout? join Anga to brief creators, manage escrows, and scale tests across counties with M-Pesa payouts.
Short motivating CTA
Stop guessing and measure lift. Use this guide on your next campaign and join Anga to activate many local creators fast — everyday creators, honest results, and secure M-Pesa payments.
Frequently Asked Questions
What is incrementality testing for influencer marketing?
Incrementality testing (holdout testing) compares outcomes between a treatment group exposed to influencer activity and a control group that is not, to estimate the causal sales or conversion lift attributable to the campaign.
Which test type should I use: geo holdout or creator-level holdout?
Choose creator-level holdouts when you can recruit many creators (better statistical power). Choose geo holdouts when you must limit creator changes by region. Creator-level is often easiest if you use a marketplace like Anga to manage many local creators.
How large must my sample be to detect uplift?
Sample size depends on baseline conversion, minimum detectable effect (MDE), and chosen power/significance. Low baseline rates (e.g., 1%) require tens of thousands of visitors per arm to detect small lifts. If you have limited traffic, aim for larger MDE or improve conversion mechanics.
How do I track offline redemptions at Kenyan retail stores?
Use unique coupon codes per creator or per test arm, train retail POS staff to record redemptions, and reconcile daily. Codes work at Naivas or local kiosks and are easy to match to creators.
Can I combine influencer content with paid social and still run a holdout?
Yes. Use audience-level holdouts in ad platforms or exclude part of your matched audience from paid amplification. Make sure you avoid cross-exposure between treatment and control and track exposures carefully with server-side events and UTMs.
What common mistakes should I avoid in incrementality tests?
Common mistakes: underpowered tests, cross-contamination between test arms, poor tracking (no UTMs or codes), running tests during market noise (major events), and ignoring mobile/M-Pesa payment flows in Kenya.
Do I need expensive tools to run these tests in Kenya?
No. Many Kenyan brands run valid tests with GA4, UTM links, coupon codes, spreadsheets, and marketplaces like Anga for creator management and M-Pesa payouts. Server-side pixels improve accuracy but aren't mandatory for small pilots.
How can Anga help with incrementality testing?
Anga helps recruit and verify many local creators, manage briefs and escrow payments, and scale creator activations across counties. That makes it easier to set up creator-level holdouts and maintain clean test/control assignments.
