Solomozone

The Every Day Digital

Recent Posts

  • Mastering AEO: Your 90-Day Optimization Plan

    Mastering AEO: Your 90-Day Optimization Plan

    A 90-day plan turns answer engine optimisation (AEO) from a vague worry into three phases: find out where you stand, fix what is wrong and publish what is missing, then measure and repeat. It needs no new budget, only an owner, a list of customer questions and the discipline to check the results.

    The reason to plan is that AEO is not a one-off fix. You are trying to become a source that AI answers can draw on, and that depends on many small things being right at the same time: clear pages, consistent facts, visible expertise and a reputation outside your own site. Doing one of them well does little. Doing all of them over a quarter starts to show.

    This post sets out the plan, a week-by-week schedule, how to measure it, and what to watch for in financial services. Where the evidence is thin, which is often, I say so.

    What is known, and what is guesswork

    Google’s own guidance is the firmest ground, and it is plainer than most AEO advice. For its AI Overviews and AI Mode, Google tells site owners to:

    • create original content with a unique point of view and first-hand expertise, and “don’t just recycle what others on the internet have already said”;
    • keep content people-first, clearly structured, and technically crawlable and indexable;
    • keep business and product data current in Merchant Center and Google Business Profile;
    • measure with the Generative AI performance report in Search Console.

    Google also lists what it says you do not need: new machine-readable or AI-specific files, special markup, rewriting content into tiny chunks, or special phrasing. It says structured data is not required for generative AI search, that an llms.txt file neither helps nor harms visibility in Google Search, and that “no third-party tool has access to our internal ranking or AI systems” (Google Search Central). An earlier Google post made the same point that clicks from AI Overviews tend to be higher quality, and that success should be judged on outcomes such as sales and sign-ups, not only clicks (Search Engine Journal, May 2025).

    That covers Google. ChatGPT, Perplexity, Gemini and others choose sources in ways they have not published. Most “AEO checklists” mix Google’s advice with practitioner opinion. One widely read playbook, for example, recommends a 40 to 80 word opening answer, set refresh schedules and a 30/60/90-day plan, but none of that is backed by cited data (ALM Corp). I use that kind of advice below as a sensible working method, not as proven rules.

    So the plan rests on one idea. The things most likely to help are the things that make you the best, clearest and most consistent source on your topic. If a tactic only makes sense as a trick, skip it.

    Days 1-30: find out where you stand

    The plans below are my working method, built on Google’s guidance and common practice. They are not a standard. Use home loans as the running example: swap in your own product.

    The first month is about evidence. Do not rewrite anything until you know what is wrong.

    1. Name an owner and pick one topic cluster. AEO touches content, product, compliance and web. One person needs to hold it together. Start with a single cluster, such as home loans, not the whole site.
    2. Build the question list. Collect 20 questions real customers ask before choosing: eligibility, rates, fees, documents, how long approval takes, how you compare with a bank or another lender. Use sales and service call notes, search queries, and your own site’s search box.
    3. Run the baseline. Ask each question in the AI tools your customers use (ChatGPT, Gemini, Perplexity, Google’s AI answers). Record for each: whether you appear, how you are described, who is cited, and what is wrong. Run each question more than once, since answers vary.
    4. Audit your facts. Check that rates, fees, eligibility rules and product names are identical across your website, app store listings, partner pages, social profiles and directories. Every mismatch is a risk, because a model may pick the wrong version.
    5. Check the technical basics. Pages must be crawlable, indexable and fast, and nothing important should sit behind a login or a script that blocks reading. Google says a page must be indexed and eligible to show a snippet to appear in its AI features (Google Search Central).

    Output of the month: a scorecard with your 20 questions, how you appear on each, and a ranked list of fixes.

    Days 31-60: fix and publish

    The second month turns the scorecard into changes. Fix the inconsistencies first, then fill the gaps.

    1. Lead with the answer. On each priority page, put a direct, plain-language answer near the top, then the detail. For “how long does home loan approval take?”, say it in the first lines, with the conditions that change it. Write in short paragraphs that each cover one idea.
    2. Say something original. Google’s advice is to avoid recycling what is already online. Add what only you know: your real approval process, your own data on how long things take, how a customer’s situation changes the answer.
    3. Show who is behind it. Add a named author or reviewer, their credentials, and a last-reviewed date. Make substantive updates, not cosmetic date changes.
    4. Close the gaps. Publish clear pages for the questions your baseline showed you missing on. Keep each one focused on a single question or decision.
    5. Use the same words everywhere. Pick the primary name for each product and term, and use it consistently, adding alternatives only as plain synonyms.
    6. Build credibility off your site. Contribute expert commentary, publish original research or data where you have it, and make sure reputable third parties describe you accurately. Google says seeking inauthentic mentions is not helpful, so this is about earning coverage, not manufacturing it (Google Search Central).
    7. Keep business data current. Update Google Business Profile and any product feeds you use.

    You do not need to add special markup for AI answers. Google says structured data is not required for its generative AI features. Standard markup that matches what is visible on the page is still fine for other reasons.

    Output of the month: revised priority pages, new pages for the gaps, and one fact sheet of approved product details that everyone in the company uses.

    Days 61-90: measure and repeat

    The third month shows what moved and sets the habit that keeps it moving.

    1. Re-run the baseline. Ask the same 20 questions, the same way, in the same tools. Compare with day 1: did you appear more often, was the description correct, who else was cited?
    2. Read the Search Console report. Google points site owners to the Generative AI performance report in Search Console (Google Search Central). Check which pages show up and for which queries.
    3. Look at the business outcome. Compare leads, applications and branded search for the optimised cluster against a similar cluster you did not touch.
    4. Keep what worked, fix what did not. Pages that earned citations show you the pattern. Pages that did not need a clearer answer, better proof or a fix to a conflicting fact elsewhere.
    5. Extend to the next cluster and set a rhythm. Make the question check a monthly habit and review the fact sheet every quarter.

    Output of the month: a before-and-after report, a decision on the next cluster, and a standing monthly review.

    The plan at a glance

    [embed: node/c3d05eaf-6c78]

    Each phase has one output and one gate. Do not start the next phase until the gate is met, because fixes made on top of an unmeasured baseline cannot be judged. Assign one person to own the plan, and name a reviewer from compliance for the gates in months two and three.

    How to measure when clicks are not the whole story

    An AI answer can shape a decision without sending a visit, so clicks alone understate the effect. Google’s own advice is to judge success on outcomes such as sales and sign-ups, not only clicks (Search Engine Journal). A practical set, adapted from a common four-layer approach (ALM Corp), is:

    • Presence. How often you appear for your 20 questions, how you are described and who appears alongside you. Track it by hand each month, or with a tool if you use one, remembering that answers vary from run to run.
    • Search signals. Impressions and queries for the optimised cluster in Search Console, including the Generative AI performance report Google points to, plus the trend in branded search.
    • Referral behaviour. Visits that arrive from AI tools, which pages they land on, and whether they convert differently from other traffic.
    • Business outcome. Qualified leads and applications for the optimised cluster compared with a similar cluster you left alone.

    Two cautions. First, tools that promise an “AEO score” are working without access to the platforms’ systems, as Google itself points out, so treat any score as an indicator you cannot audit. Second, with one quarter of data and a small question set, you are looking for direction, not proof. Keep the question list fixed so the comparison stays fair.

    Considerations for BFSI and fintech

    In financial services a wrong answer is not just a missed click. A model that quotes an outdated rate, the wrong eligibility rule or a fee you no longer charge creates a customer expectation you then have to meet or correct. Four habits help:

    1. Make one fact sheet the single source of truth. Rates, fees, eligibility, documents and timelines, with an owner and a review date. Every page, listing and partner profile draws from it.
    2. Date everything that can change. A visible last-reviewed date and a clear statement of what the figures apply to give both readers and machines a way to tell current from stale.
    3. Put compliance in the workflow. Have compliance review answer-first pages once, then use approved wording and disclosures as templates, so you can publish quickly without reopening each claim.
    4. Monitor for wrong answers, not just missing ones. When your baseline finds an AI answer that misstates your product, log it, fix the conflicting source if there is one, and report it through the platform’s feedback route where one exists.

    Be cautious about promising returns or outcomes in answer-style content. Read your regulator’s advertising rules and each platform’s policies for financial content before publishing at scale. I have not reviewed them for this post.

    Common mistakes

    • Chasing hacks. Google says you do not need new files, special markup, tiny content chunks or special phrasing, and that an llms.txt file does not affect Google Search. Spend the time on the content instead (Google Search Central).
    • Mass-producing pages. Google calls churning out pages to manipulate rankings or AI responses scaled content abuse. A few strong pages beat hundreds of thin ones.
    • Rewriting without a baseline. If you do not record where you started, you cannot tell what worked.
    • Fixing the site and ignoring everything else. Conflicting facts on partner pages, listings and old PDFs can undo the work.
    • Treating it as a project with an end date. Answers and sources keep changing. The monthly question check is the part that lasts.

    The takeaway

    AEO is not a new trick to learn. It is the old discipline of being the clearest, most accurate and most credible source on your topic, applied in a way you can check. Ninety days is enough to find out where you stand, fix what is wrong and see whether it moved. Start with one cluster, one owner and 20 questions. The brands that do this early will have the habit in place when the rules harden.

    Sources

  • Shifts in Performance Marketing: Embracing AI

    Shifts in Performance Marketing: Embracing AI

    AI is not making performance marketing easier. It is moving the work. The platforms now handle bidding, audience targeting and increasingly the creative itself, so the old levers (keywords, audiences, bid tweaks) matter less every quarter.

    What replaces them is a new set of levers: the quality of the creative you feed the machine, the quality of the conversion data you send back, and how honestly you measure what actually worked. The teams that treat those three as their job will keep an edge. The teams that still optimise campaign settings are optimising the part the platform has already taken.

    This post walks through five shifts: automated buying, creative as the main lever, strained measurement, changing search, and agents.

    1. Buying is being automated, and control is being traded for scale

    The two biggest ad platforms have moved to the same model: you set a goal and a budget, and the system decides who sees what, where, and at what price.

    • Meta. Meta reported that Advantage+ sales campaigns improved return on ad spend by an average of 22% across 2025, and that automated Advantage+ structures became the default for sales, leads and app objectives, with manual controls still available (PPC Land, reporting Meta’s Q4 2025 results of 28 January 2026). The Wall Street Journal reported in June 2025 that Meta aims to let an advertiser supply a product image or URL and a budget, with AI generating the creative and choosing the audience by the end of 2026. That is a reported goal, not a confirmed product (Pixis, a vendor blog citing the WSJ).
    • Google. AI Max for Search adds broader search-term matching, text customisation and final-URL expansion, with controls for brands, locations and URLs. Performance Max has added channel-level reporting and asset segmentation. Ads can also show in AI Overviews, but advertisers cannot target that placement directly (ALM Corp, a secondary summary that does not link Google’s own announcements).

    The results are real but self-reported. Meta’s 22% is Meta’s own average. A vendor blog cites 640 incrementality tests by Haus that reportedly found Advantage+ campaigns underperforming manual ones over time despite strong platform-reported ROAS (Pixis).

    The practical point does not depend on who is right. When the platform marks its own homework, the advertiser’s job shifts from adjusting settings to checking whether the results are incremental.

    2. Creative becomes the main lever

    When targeting is automated, the creative does the targeting. Different images, hooks and messages attract different people, and the algorithm learns who responds to which. That makes the volume and range of creative the input you can still control.

    The platforms are moving fast to make that input cheap:

    • Meta says its video generation tools reached a $10 billion combined revenue run-rate (PPC Land, reporting Meta’s Q4 2025 results).
    • At Advertising Week on 6 October 2026, Meta announced that Advantage+ Creative’s tool for turning static images into video ads is generally available, that Ads Creative Studio now has broader access, and that its AI business assistant can take in your brand-voice and strategy documents (TechJuice).

    Cheap creative changes the job in three ways. This part is my analysis.

    1. Testing moves from A/B pairs to many variants. The scarce skill is no longer producing the next ad, but deciding which ideas deserve a test.
    2. The brief becomes the product. A model fed a vague brief produces the same average ad as everyone else’s model. A sharp insight, a distinct tone and clear claims are what make the output different.
    3. Brand risk rises. Generated ads can drift from your voice or make claims you cannot support. Guardrails (approved claims, tone rules, banned phrases) have to exist before volume goes up, not after.

    One commonly quoted statistic says 70 to 80% of ad performance comes from creative quality.

    3. Measurement is under strain

    Automation makes measurement harder in two ways. The platform decides who sees the ad, so you can no longer infer much from the audience you picked. And the same platform reports the results, so credit is awarded by the party being judged.

    Marketers are responding. A July 2025 EMARKETER and TransUnion survey found 52% of US brand and agency marketers use incrementality testing and experiments, and 36.2% plan to spend more on it in the next 12 months. But adoption is ahead of maturity: an IAB and BWG Global study found three in four marketers say their measurement does not deliver the speed, accuracy or trust they need (EMARKETER).

    Platforms are adapting too. Meta says its incremental attribution product reached a multi-billion-dollar annual run-rate about seven months after launch, and that it delivered 24% more incremental conversions than standard attribution (PPC Land, quoting Meta’s Q4 2025 call).

    The measurement stack I would aim for, which is my view and not a standard:

    1. Platform reporting for daily optimisation, read with scepticism.
    2. Incrementality tests (geo or holdout) on your biggest channels, to check that platform-reported results are real lift.
    3. Marketing mix modelling to see the whole budget, including offline and brand spend, and to calibrate the other two. I covered this in my post on why MMM is back.

    No single layer is enough. The discipline is to let each one check the others.

    4. Search is changing under your feet

    Search is the backbone of most performance programmes, and AI answers are reducing the clicks available. Two pieces of evidence, both with limits:

    • Seer Interactive studied 3,119 informational queries across 42 client organisations from June 2024 to September 2025. On queries showing an AI Overview, organic click-through rate fell from 1.76% to 0.61% (a 61% drop) and paid click-through rate fell from 19.7% to 6.34% (a 68% drop). Brands cited in the overview had higher click-through rates than those not cited, 91% higher on paid, though the researchers say they cannot prove citation is the cause (Seer Interactive via PPC Land). The sample is agency clients, informational queries only, and the paid sample is small.
    • SparkToro estimated that about 68% of US Google searches from January to April 2026 ended without a click, up from about 60% in 2024. Only about 0.34% of searches had moved into AI Mode in that period (SparkToro via TechWyse). The data is panel-based with assumptions, and the rise cannot be cleanly pinned on AI Overviews.

    For a performance marketer, the lesson is narrower than “search is dead”. The clearest click losses are on informational queries. Neither study isolates queries with clear buying intent, so I would not assume the same drop applies there, and I would not assume it does not. Plan for fewer cheap clicks at the top, and for competing to be the cited source, which is the same visibility work as answer engine optimisation.

    Google’s ad response is to place ads in and around AI answers, where advertisers cannot target the placement directly (ALM Corp). That makes broad match, strong assets and good conversion signals more important, not less.

    5. Agents arrive on both sides of the click

    Agents are showing up as media buyers and, increasingly, as shoppers.

    Agents as media buyers. Meta opened its Ads MCP server to any developer with a Meta app (published 16 July 2026, updated 22 September 2026). Agents connected to it can create, edit and delete campaigns, ad sets and ads, manage audiences and catalogs, pull performance reports, check signal quality and run A/B tests and conversion lift studies. Meta lets you set rules that limit what an agent can do on an account, such as read-only access (Meta for Developers). Google’s Ads MCP server, per one integration guide dated 14 June 2026, is read-only for now (ContentGrip).

    Agents as shoppers. Meta is testing direct checkout inside Meta AI with Adyen, PayPal, Shopify and Stripe, and its Business Agent on Messenger already supports in-chat checkout (TechJuice). When the purchase happens inside an assistant, the ad click that your funnel depends on may never occur.

    What this means, in my view:

    • Permissions are now a marketing decision. Decide what an agent may change (budgets, creative, audiences) and what it may only read, and who signs off. Meta’s rule system is built for this.
    • Routine work gets automated first. Reporting, pacing alerts and catalogue fixes are the obvious candidates. Judgment about brand, claims and budget allocation should stay with people.
    • Your product data and site become the interface. If an agent compares you with a rival, clean, accurate and structured product information decides who gets picked.

    What to do in the next 90 days

    • Audit the conversion events you send to Google and Meta. Add a downstream, value-bearing event and remove weak ones.
    • Run one incrementality test (geo or holdout) on your largest paid channel, and compare it with the platform-reported result.
    • Write a creative brief template with the insight, the proof, the approved claims and the banned phrases, and use it for every AI-generated batch.
    • Build a swipe library of approved claims and disclosures that generative tools must draw from.
    • Track click-through rate and conversions on informational and high-intent queries separately, so you can see where AI answers are eating clicks.
    • Start an answer-engine visibility check: ask the ten questions customers ask and note who is cited.
    • Pilot a read-only agent for reporting and alerts, with written rules on what it may never change.
    • Review brand and performance budgets together once a quarter.

    The takeaway

    Performance marketing is not being replaced by AI. It is being split in two. The platform takes the mechanical work of buying, and the marketer keeps the parts a model cannot supply: a real insight, a distinct brand, honest measurement and the judgment to say no. The marketers who do well will be the ones who get better inputs and better proof, not the ones who push more buttons.

    Sources

    Pages opened for this post, as of 19 October 2026. Secondary and vendor sources are flagged in the text.

  • Understanding Agentic AI: Transforming Customer Journeys

    Understanding Agentic AI: Transforming Customer Journeys

    Until now, every customer journey has had a person at every step. A person searched, compared, filled in the form and clicked buy. Marketing was built to win that person’s attention.

    Agentic AI starts to change that. Instead of only answering a question, an AI agent can take a task, such as finding a better insurance quote, renewing a policy or comparing loans, and carry it out across several steps with limited supervision. In some of those steps, the one reading your website, weighing your offer or checking your terms will be the customer’s software, not the customer.

    The choices brands make now about how clear, consistent and trustworthy their information is will help decide whether those agents pick them. This article looks at what agentic AI is, how it reshapes each stage of the journey, and what marketers in trust-based sectors should do about it.

    What agentic AI is, and what it isn’t

    A chatbot answers. An agent acts. Agentic AI takes a goal from a person, then plans and carries out steps across tools and services to meet it, with limited supervision.

    FinTech Futures draws the line clearly: AI that recommends versus AI that acts. In finance, it says, agents can initiate payments, switch products and rebalance portfolios, which is a change in who, or what, is acting (FinTech Futures, July 2026). In retail, one guide describes an agent given a goal such as finding trail shoes under $150 that compares options and completes the purchase (Fin).

    It helps to be clear about what this is not.

    • It is not full autonomy. The customer sets the goal and the limits. Spending limits and approval thresholds are how people stay in control.
    • It is not evenly spread. The clearest early activity is in retail. In January 2026, Google announced the Universal Commerce Protocol, an open standard for discovery, buying and post-purchase support, developed with retailers including Shopify, Etsy, Wayfair, Target and Walmart, with agentic checkout for eligible U.S. retailers (Google). Financial services is moving more cautiously, and regulators are still working out the rules.
    • It is not yet the norm. One study reported by Fin found that 45% of consumers use AI for at least part of the buying journey (IBM Institute for Business Value, 2026). Using AI to help is a long way from handing it your wallet, so treat adoption figures with care.

    How each stage of the journey changes

    An agent does not remove the customer journey. It changes who walks each stage, and what that stage rewards.

    StageTodayWith an agent in the loopWhere marketers should focus
    IntentThe customer searches, or sees an adThe customer states a goal to an agent, or the agent spots a need from saved preferencesBe the brand that answers real questions clearly
    Discovery and comparisonThe customer browses sites, reviews and comparison pagesThe agent queries catalogues and reviews and compares specifications and pricesComplete, structured, consistent facts that a machine can read
    DecisionThe customer weighs brand feel, offers and adviceThe agent applies the customer’s rules on price, features and limits, and shortlistsEarn a place on the shortlist on facts, and earn trust before the rules are set
    Purchase and onboardingThe customer fills in forms and completes verificationThe agent completes steps within agreed limits, and the human approvesSimple, secure journeys with clear consent and identity checks
    Use and serviceThe customer calls or chatsThe agent tracks status, raises issues and can start a return or a switchReliable service channels that software can use, and clear complaint routes
    Renewal and loyaltyThe customer renews out of habit or after a reminderThe agent compares options at every renewal and may recommend a switchEarn renewal on value and trust, not inertia

    Two things stand out. First, inertia, the quiet reason many customers stay, weakens when software re-checks the market every cycle. Second, the brand does not disappear. It moves earlier in the journey, because it shapes the rules a customer gives their agent and which agents and providers they trust at all.

    Who you are marketing to now

    You now have two audiences. The customer sets the goal, the limits and the level of trust. The agent reads the facts and acts within those limits.

    The agent does not respond to a clever headline or a beautiful hero image. It responds to information it can read, compare and rely on. For retail, the guidance is concrete: provide complete, structured product attributes, keep inventory and pricing real time, and connect to emerging protocols, because stale data erodes trust in both the agent and the merchant (Fin). The same logic applies to a loan, an insurance policy or an investment product. Rates, fees, eligibility and terms need to be complete, consistent and current wherever an agent might look.

    Brands will also have agents of their own. Google’s January 2026 announcement included a branded chat agent that answers product questions in a retailer’s voice, and a pilot in which advertisers set offers and Google’s AI decides when to show them (Google). Your own agent will increasingly talk to the customer’s.

    The human audience matters just as much, in a different way. A customer deciding which providers their agent may use is making a trust decision, and trust is built by the brand work that happened long before the agent arrived.

    What it means for financial services

    Financial services raises the stakes, because an agent’s mistake can move money.

    • A wrong action can happen before anyone notices. The historic risk was a poor recommendation. With agents, a wrong payment or trade may be completed before a human sees it (FinTech Futures).
    • Authorisation needs defining. Did the customer approve a specific transaction, or only the agent’s authority to act? For how long, and up to what limit? The same article argues customers need easy ways to pause or withdraw an agent’s authority at any time.
    • Accountability can blur. When the firm, a model provider, a platform and the customer’s own agent all play a part, the customer may not know who is responsible. The UK’s Mills Review, published in July 2026, calls for auditable records, clear complaint routes and the ability to explain outcomes (Addleshaw Goddard summary).
    • Persuasion draws scrutiny. The review warns that firms could use AI to tailor persuasive prompts and friction in real time, and advises firms to ask whether personalisation serves consumers or mainly drives conversion. It also advises keeping non-AI routes available.

    The Mills Review is a UK document, and rules differ by country. The principles of clear authority, accountability, auditability and fair persuasion travel well, but your compliance team should confirm what applies to you.

    What marketers should do now

    You do not need an agent strategy for every scenario. You need strong foundations that serve both audiences. The examples are illustrative.

    1. Make your product facts consistent and machine-readable. Keep one source of truth for rates, fees, eligibility and terms, and make every channel match it.
      • Example: A home loan’s rate range, processing fee and eligibility criteria are held in one governed record, and the website, app and partner pages all draw from it.
    2. Keep information current. Stale facts cost you trust with both people and software.
      • Example: A rate change is published at source the same day, with a visible last-updated date.
    3. Answer real questions clearly. The practices behind answer engine optimisation, such as question-led headings and direct answers, also help agents find and use your information.
      • Example: A plain-language page on “What happens to my EMI if rates change?” that leads with the answer.
    4. Design consent and control into delegated actions. Show customers exactly what an agent may do, for how long and up to what amount, and make it easy to pause or revoke.
      • Example: A standing-instruction page listing each permission, its limit and expiry, with a one-click way to withdraw it.
    5. Make service channels software-friendly. Status checks, changes and complaints should not depend on a phone call.
      • Example: A documented route for tracking an application and raising a complaint that works from an app or a secure link.
    6. Keep building the brand. Trust decides which providers a customer permits their agent to use.
      • Example: A reputation for clear communication and fair treatment that customers name when they say which providers they trust.
    7. Assign accountability and keep records. Name an owner for each agent-facing journey, and keep an audit trail of instructions, data used and actions taken.
      • Example: One named executive owns agent-facing journeys, and every automated action can be traced back to a mandate.
    8. Start small and test. Ask today’s AI assistants the questions your customers ask, and see what they say about you.
      • Example: Run your 25 most important customer questions through several assistants each month and log accuracy errors for correction.

    Risks and open questions

    • Adoption is early and uneven. Retail is ahead, and financial services is slower. Plan for a gradual shift, not an overnight one.
    • Gatekeepers may emerge. Agents and platforms can sit between you and the customer. Google’s pilot, in which its AI decides when to show advertiser offers labelled as sponsored, shows that paid placement is likely to appear inside agent answers.
    • Access may be unequal. The Mills Review warns of a possible two-tier market in which some consumers benefit from better AI and others do not.
    • Persuasion can cross a line. Real-time, AI-tailored nudges could exploit consumers. Decide in advance what you will not do.
    • Liability is unsettled. Who is responsible when a firm, a platform and a customer’s agent all play a part is still an open question.
    • Measurement is unclear. If an agent chooses on the customer’s behalf, today’s funnel metrics will not capture why you won or lost.
    • Standards are still forming. Protocols such as the Universal Commerce Protocol are young, and the picture for financial products may look different.

    The bottom line

    Agentic AI does not make marketing less important. It moves the work. Part of the audience becomes software that reads facts, and part of the job becomes earning the trust that decides which software is allowed to act for the customer.

    The brands that do well will have clear, consistent, current information, journeys that give customers real control, and a reputation worth delegating to. In financial services, where an agent’s action can move money, every automated step must still lead back to an accountable human institution.

    A question for you: if a customer’s agent compared you with your three closest competitors tomorrow, would your facts, your service and your reputation hold up?

    Sources

  • Why Marketing Mix Modelling is Essential for Marketers

    Why Marketing Mix Modelling is Essential for Marketers

    For years, the fashionable measurement question in marketing was which click led to which sale. Dashboards promised to trace every conversion back to an ad, and budgets followed the numbers.

    Now a method that has been around for decades is back on the CMO’s agenda: marketing mix modelling, or MMM. It does not track individual people. It looks at the whole picture, including spend by channel, prices, promotions, seasonality and sales over time, and estimates what each part contributed.

    The reasons for the comeback are practical. Tracking individuals has become harder, platforms mark their own homework, and finance teams are asking tougher questions about where the money goes. This article explains what MMM is, why it is returning, how it compares with other methods and how to start, with a focus on financial services.

    What marketing mix modelling is

    MMM is a statistical method that estimates how drivers such as price, promotion, distribution and media affect outcomes like sales or new customers. It separates the incremental effect of each from the baseline trend and the noise, and it covers non-media factors as well as paid media (Measured).

    It works on aggregate data, such as weekly spend by channel and weekly sales, not on individual people. That is why it needs no cookies or personal identifiers.

    In plain terms, it asks: if we had spent more on one channel and less on another last quarter, what would probably have happened to sales? A typical model uses regression with two ideas built in: adstock, meaning the effect of advertising carries over time, and saturation, meaning each extra rupee of spend does a little less than the last (Digital Applied).

    It typically produces three things:

    • Contribution by channel: how much of the result each channel and non-media factor drove.
    • Response curves: where extra spend stops paying back.
    • Scenarios and budget recommendations: what might happen if you move money around. Google’s open-source tool, Meridian, produces budget optimisation recommendations and scenario analyses (Google).

    One caution belongs at the start. MMM finds relationships in historical data. It is not inherently causal unless it is calibrated with experiments (Measured).

    Why it is back

    Four things have pushed MMM back up the agenda.

    1. Tracking individuals got harder. Privacy rules and the loss of tracking signals have weakened person-level measurement. Google says Meridian needs no individual identifiers or cookies, which it calls privacy-safe (Google).
    2. Platform reporting got more opaque. As buying becomes automated, marketers see less about where ads run. One travel-brand analyst quoted by AdExchanger values MMM for moving “from a microscope to a telescope”, crediting channels such as TV, radio, out-of-home and search with more control than the placement reporting in automated campaign types gives (AdExchanger, July 2026).
    3. Offline and brand effects need credit. MMM covers all channels, including offline and brand. Attribution sees only the trackable slice, so it tends to over-credit digital channels (Digital Applied).
    4. Free, open tools lowered the barrier. Meta’s Robyn arrived in 2020, and Google describes Meridian as launched at the start of 2025. Both let teams build a model without buying a platform. Meridian can also use the results of incrementality experiments to calibrate its estimates (Google).

    The appetite shows up in surveys, too. As reported by Digital Applied, eMarketer’s 2026 data found 27.6% of US marketers rating MMM the most reliable method, against 19.4% for multi-touch attribution. That is a preference, not proof of accuracy.

    The comeback comes with caveats, covered under pitfalls below.

    How MMM compares with attribution and experiments

    MMM is not a replacement for everything else. Each method answers a different question, and the strongest set-ups use all three.

    Marketing mix modellingMulti-touch attributionIncrementality tests
    What it doesRegresses aggregate spend and other drivers against salesTracks individuals across touchpoints and splits the creditCompares exposed regions or audiences with a control group
    StrengthsCovers every channel, including offline and brand. Needs no personal data. Supports budget allocationFast and granular, and useful for weekly digital tweaksGives causal evidence and settles disputes between the other two
    WeaknessesSlow. Can give confident but wrong answers if inputs are incomplete or history is shortSees only the trackable slice. Identity coverage has fallenHard to run for national media. Covers only the channels tested
    Data neededRoughly two years or more of clean weekly spend and sales dataUser-level journey data with identity resolutionA designed split of regions or audiences
    SpeedQuarterly or rollingNear real time to weeklyA test period, then analysis
    Best used whenSpend includes offline media, or sales cycles run beyond about 30 daysSales cycles are short and conversion volumes are highThe other methods disagree, or you need to validate a model

    The table draws on Digital Applied, with the data requirement also reflecting Google’s guidance that Meridian needs at least two years of history.

    The layered approach it recommends is simple. Use MMM to set quarterly budget envelopes, attribution for weekly digital tactics inside them, and experiments to validate the model and resolve conflicts.

    What it means in financial services

    This section is my analysis, not a sourced finding, and the examples are illustrative. Financial services has several features that make MMM both useful and tricky.

    • Long, lumpy journeys. A loan, an insurance policy or an investment is rarely bought on first click. Long consideration cycles are exactly where click-based attribution struggles and MMM is said to fit best.
    • Offline and relationship influence. Branches, relationship managers, call centres and partners shape outcomes. MMM can include them as drivers, where click tracking cannot.
    • Brand and trust effects. Brand spend builds the trust that makes performance campaigns work later. MMM can estimate that lagged effect, which is hard to see in last-click reports.
    • Strong seasonality. Tax-saving season, festival offers, salary cycles and rate changes move demand regardless of marketing. A good model separates these from media effects.
    • Thin data for niche products. Low-volume products, such as a specialised bond offering, may not have enough observations for a stable model. Model at a higher level, such as a product family, or rely more on experiments.
    • Choose the outcome carefully. Model disbursed loans, activated cards or first investments, not leads. Otherwise the model repeats the cheap-lead trap.

    Example: A lender models weekly disbursals against spend on search, social, TV, outdoor and branch promotions, with variables for rate changes and festival weeks. The model suggests that search is near saturation while a regional TV campaign is under-funded. The lender then tests that finding with a geo experiment before moving a large budget.

    How to start

    You do not need a large programme. You need a clear question, decent data and the discipline to test what the model tells you. The examples are illustrative.

    1. Define the outcome that matters. Pick the result the business cares about, not the easiest one to count.
      • Example: Activated customers or disbursed loans, not leads or clicks.
    2. Gather at least two years of weekly data. Include spend by channel, prices and rates, promotions, seasonality and relevant external events. Google says Meridian needs at least two years of history.
      • Example: Weekly spend for search, social, TV, outdoor and branch campaigns, alongside weekly disbursals, rate changes and festival weeks.
    3. Capture all the spend. Leaving channels out can distort the results for the ones you include.
      • Example: Add sponsorships, partner marketing and agency-managed spend, not just the platform dashboards.
    4. Choose your route and your people. Open-source tools are free to download, but Google notes that most teams need a data scientist or a partner to get the most from Meridian.
      • Example: A small in-house analyst paired with an independent consultant for the first build and review.
    5. Calibrate with experiments. Run a geo or holdout test on one or two channels and use the results to check and tune the model.
      • Example: Pause one channel in a few matched regions for six weeks and compare results with the model’s prediction.
    6. Move a slice of the budget first. Act on findings gradually and measure what happens.
      • Example: Shift around 10% of the budget the model flags as saturated into the channel it flags as under-funded, then compare outcomes.
    7. Refresh regularly. Models go stale in volatile markets (Measured).
      • Example: Re-run the model every quarter and after any major pricing or product change.
    8. Set up governance. Decide who owns the model, document its assumptions and have it reviewed independently.
      • Example: A one-page assumptions log signed off by marketing, finance and analytics.

    Pitfalls to avoid

    • Treating the output as truth. MMM is correlational unless calibrated with experiments. Treat its answers as strong hypotheses to test.
    • Accepting default settings. AdExchanger quotes Mutinex’s CEO warning that many advertisers use Meridian’s off-the-shelf settings, when the model should be customised for each business (AdExchanger).
    • Treating free as cheap. The software is free, but the data work, the people and the review are not. The same article likens it to a puppy, not a free beer.
    • Ignoring conflicts of interest. A platform-built model can be deeply tied to that platform’s ecosystem. Open code makes a method inspectable, but it does not make the model independent. Ask for independent review.
    • Betting on one tool. AdExchanger reports, citing unnamed sources, that Meta is winding down Robyn. Whatever the outcome, keep your data and assumptions portable so you can switch tools.
    • Over-fitting a thin history. With short or messy data, a model can fit the past beautifully and predict the future badly (Measured).
    • Believing big vendor promises. One vendor guide claims 10 to 30% efficiency gains and cites no source for the figure (Measured, a vendor). Ask for independent evidence.

    The bottom line

    Marketing mix modelling is back because the click-based picture became less complete, not because it is magic. It sees what clicks cannot, including offline media, brand effects and the slow build of trust, and it works without tracking individuals.

    Use it as one layer of three. Let MMM guide the big budget decisions, attribution guide the weekly tactics, and experiments keep both honest. Choose outcomes the business cares about, calibrate the model, review it independently and refresh it regularly.

    A question for you: if you knew the true contribution of each channel to your last quarter’s results, which budget would you move first, and how sure are you of that answer today?

    Sources

Create your own website, free on WordPress.com.Create your own website.
Start free