DAOVA.ai Blog

HubSpot and Pipedrive Just Shipped AI Note-Takers for Your CRM — What to Check Before You Turn Yours On

victor@daova.ai (victor@daova.ai) — Wed, 23 Sep 2026 16:18:10 GMT

On September 16, 2026, two of the CRM platforms small businesses actually use — HubSpot and Pipedrive — announced, on the same day, AI features that do the same basic thing: listen to your sales calls and meetings, write up what happened, and update your CRM records without anyone typing it in. HubSpot unveiled a Mobile Notetaker and a Deal Progression tool as part of its Fall '26 Spotlight, feeding meeting and call details straight into a new self-updating "Smart CRM." Pipedrive shipped Nova, an AI assistant that briefs a rep before a call, transcribes it, and drafts the CRM update afterward — included in every plan at no extra cost.

Neither launch is really a coincidence. Pipedrive's own research, released alongside Nova, found that 42% of sales professionals spend at least 40% of their working day on tasks that don't generate revenue, and nearly half say CRM updates or follow-ups routinely take more than a day to actually get done. That's the real problem this category of feature is aimed at: the gap between when a conversation happens and when it gets written down — which is exactly where small deals quietly go cold and where account history gets lost the moment a rep leaves.

If you run a small sales or account team on either platform — or on Salesforce, Zoho, Close, or anything else likely to ship a similar feature soon — this is worth ten minutes of attention for two reasons. It's a genuinely useful fix for one of the most common, least glamorous problems small businesses have. And "AI now listens to your calls and writes directly into your customer records" is exactly the kind of feature that takes five minutes to turn on and can be genuinely awkward to walk back once it's been running for a month.

What's actually new here

SMBs have had call-recording and transcription add-ons for years. What's different about this wave is where the output goes: instead of producing a transcript you have to read and act on yourself, the AI writes directly into the deal record — updating stage, logging next steps, summarizing the conversation — with a human reviewing after the fact rather than typing it in the first place. That's a meaningful shift in who's making the first pass at your customer data, and it changes what you need to check before you rely on it.

The part worth slowing down for: consent

Recording a phone or video call isn't legally neutral everywhere. A significant number of U.S. states — including California, Florida, Illinois, Pennsylvania, and Washington — require the consent of every party on a call before it can be recorded, not just the person who initiated it. If your team calls customers or prospects in any of those states, or your customers are calling in from one, an AI notetaker that silently starts recording and transcribing is a real exposure, not a theoretical one.

Check what your tool actually discloses

Both HubSpot and Pipedrive build some form of participant notice into their meeting tools, but "some form" is doing a lot of work in that sentence — defaults vary by integration (a direct dial-in behaves differently than a Zoom or Teams call routed through a bot), and settings get missed. Before rolling this out past a pilot, actually join a call with the feature turned on and confirm, as a participant would experience it, that there's a clear verbal or on-screen notice that the call is being recorded and by what.

Don't assume the vendor's default is compliant for your states

A feature can be legal to ship broadly and still not match every state's consent requirement out of the box — that's a configuration question for you, not something the vendor can fully solve on your behalf. If your business calls into two-party consent states regularly, this is a five-minute conversation with your account rep worth having before go-live, not after.

Before you turn it on: a short checklist

1. Decide what it's allowed to write automatically

Most of these tools let you choose between the agent drafting an update for a human to approve versus writing directly into the record. Start with draft-and-approve, especially for anything that changes deal stage, close date, or forecast — the fastest way to lose trust in an automation like this is one wrong auto-logged detail that quietly throws off a pipeline report.

2. Find out who can access the recordings and transcripts

A transcript of a sales call often contains pricing discussions, competitor mentions, and sometimes a customer's own confidential information. Check where recordings live, how long they're retained, and whether access is scoped to the deal team or visible portal-wide by default — this is the same "who can see it" question worth asking of any AI tool touching customer conversations.

3. Pilot on internal meetings first

Turn the notetaker on for your own team standups or internal deal reviews for a week before pointing it at customer calls. You'll get a fast, low-stakes read on transcription accuracy, how well it identifies action items, and whether the CRM updates it drafts actually reflect what was said — all without any consent or customer-data questions attached.

4. Update whatever AI or data policy you already have

If your business has a written AI usage policy — even a short one — add a line covering call recording and consent specifically. Most existing policies were written before "your CRM listens to calls by default" was a real setting your team could flip on, so this is a gap worth closing rather than assuming it's already covered.

The takeaway

This is a good example of AI arriving as a quiet setting inside software you already pay for, rather than a new tool you have to evaluate from scratch — which makes it easy to turn on without really deciding to. The underlying idea is sound: sales teams genuinely lose time and deal context to manual CRM updates, and a well-configured notetaker fixes a real problem. The parts that deserve ten minutes of deliberate attention before go-live are the ones that don't show up in the launch announcement — consent, transcript access, and how much you trust it to write to your records unsupervised. Get those settled with a small pilot, and this is one of the more genuinely useful AI features to land in ordinary business software this year.

DAOVA helps small businesses evaluate new AI features in the tools they already use and build the governance practices to deploy them safely. Explore AI Governance & Operations.

OpenAI's Sponsored Agents Bring Paid Ads Inside ChatGPT — What SMBs Should Know Before Buying In

victor@daova.ai (victor@daova.ai) — Mon, 21 Sep 2026 16:07:08 GMT

On September 16, 2026, OpenAI rolled out "Sponsored Agents" inside ChatGPT — a new ad format where clicking a sponsored placement doesn't just take you to a landing page, it opens a labeled conversation with an AI agent representing that business. In the same announcement, OpenAI shipped its first e-commerce and CRM integrations: US Shopify merchants can now install a ChatGPT Ads app directly from the Shopify App Store, and HubSpot users can create and manage ChatGPT ad campaigns without leaving their CRM. Early partners include Wayfair, Angi, Newegg, Best Buy, Lowe's, and VistaPrint — mostly larger players, but the on-ramp through Shopify and HubSpot is squarely aimed at businesses much smaller than that.

If you sell through Shopify or run your pipeline through HubSpot, this is the first time "advertise inside ChatGPT" has become something you can actually click a button and try, rather than a wait-and-see enterprise pilot. That also means it's worth five minutes of clear thinking before anyone on your team spends real budget there.

What Actually Launched

Ads that talk back

A traditional ChatGPT ad still shows up as a labeled placement in a conversation. What's new is what happens after someone clicks it: instead of just landing on your website, they can start a clearly disclosed conversation with an agent built to represent your business — answering questions, working through objections, and pointing them to your site or booking page when they're ready to move. OpenAI is positioning this as a way to keep a prospective customer engaged during the research phase, rather than losing them the moment they leave the chat.

Shopify and HubSpot are the new front doors

Before this update, getting into ChatGPT Ads meant working through OpenAI's ad tools directly. Now there are two much shorter paths in. Shopify merchants in the US can install the ChatGPT Ads app, sync their product catalog automatically, and launch campaigns using data ChatGPT already has about their store — with international availability planned later this month. HubSpot users can connect a ChatGPT Ads account and build campaigns using their existing contact and deal data, without switching tools. Both are explicitly built for businesses that don't have a dedicated ad-ops team, which is most small businesses.

The money behind the push

This isn't a side experiment for OpenAI. Its advertising business reportedly hit a $1 billion annualized revenue run rate in under 200 days, and the company is targeting $2.5 billion in ad revenue for 2026. That scale of investment usually means the product keeps getting pushed harder and faster — worth knowing before you assume this is a quiet beta you can ignore for a year.

Why This Is Worth Your Attention, Not Just Enterprise's

It's tempting to file this under "another AI feature for big brands" and move on — Wayfair and Best Buy aren't exactly small business comps. But the mechanism that matters here isn't the partner list, it's the distribution: this launched with a Shopify App Store listing and a HubSpot integration, the two platforms an enormous share of small businesses already run on. When a new ad channel ships with a one-click install for tools you already use, the barrier to trying it drops from "hire an agency" to "click connect," and that's exactly the moment a lot of businesses either jump in without thinking or dismiss it without looking. Neither is the right move yet — the right move is to evaluate it deliberately, the same way you would any new ad channel.

Questions to Ask Before You Spend Anything Here

1. Is this actually available to you yet, and at what cost?

Sponsored Agents themselves are still limited to select advertisers, so most small businesses won't be running one on day one — but the underlying ChatGPT Ads platform, including the Shopify and HubSpot on-ramps, is broader. Check your Shopify App Store or HubSpot marketplace directly rather than assuming based on headlines, and get real numbers on minimum spend and pricing model (cost-per-click, per-impression, or something newer) before committing a budget.

2. Who is actually writing what the agent says on your behalf?

A Sponsored Agent conversation is a live, branded interaction with your prospective customer — closer to a chat rep than a static ad. Before you'd let a new hire freelance answers to customer questions, you'd give them a script and boundaries. Ask the same of an AI agent representing your business: what is it allowed to say about pricing, availability, and policies, and who reviews that before it goes live? Get this wrong and the risk isn't wasted ad spend, it's a customer holding your business to something an AI agent said that wasn't true.

3. What's your actual benchmark, not the vendor's pitch?

Every new ad platform launches with the best possible framing. Before spending here, know what you currently pay to acquire a customer through your existing channels — Google, Meta, referrals, whatever it is — so you have a real number to compare against instead of judging this new channel in a vacuum. If you can't answer "what does a customer currently cost me" for your existing channels, that's the more urgent problem to fix first.

4. Can your website close what the conversation starts?

Just like the shift toward AI shopping agents that send traffic to merchants' own sites rather than transacting in-chat, a Sponsored Agent conversation still ends with a handoff to your website or booking page. If that page is slow, confusing, or missing basic information, you're paying to generate interest your own site then loses. Fix your landing experience before you fix your ad spend.

What to Do This Week

If you're on Shopify or HubSpot, take ten minutes to check whether the ChatGPT Ads app or integration is available in your account, and if it is, read the actual terms rather than the press coverage. If you're not ready to spend anything yet, that's a completely reasonable place to be — this is five days old at the time of writing, and pricing, guardrails, and results data are all still thin. The businesses that get the most out of a new ad channel are rarely the ones who move fastest; they're the ones who know their own numbers well enough to recognize a good deal when the channel matures.

DAOVA helps small businesses evaluate new AI tools like this with a clear head instead of a rush. Explore AI Enablement.

Your Business Is Buying AI Faster Than It Can Support It — A Readiness Check

victor@daova.ai (victor@daova.ai) — Fri, 18 Sep 2026 16:06:41 GMT

A story making the rounds this month gets at something a lot of small business owners are feeling but haven't quite named: you can buy AI faster than your business can actually absorb it.

New reporting from CPA Practice Advisor, published September 14, pulls together a few data points worth sitting with. QuickBooks now finds that more than three in four U.S. small and midsize businesses use AI regularly. Clutch's August research found that many of those same AI-using businesses still have real gaps in data readiness, governance, and technology strategy. And Gartner research reported by the Wall Street Journal shows 85% of functional leaders plan to increase AI spending again in 2026.

Put those together and the picture is clear: adoption is outrunning the plumbing. Most businesses aren't struggling to find an AI tool to try. They're struggling with what happens after they turn it on.

Why "It Exposes What Was Already There" Is the Real Headline

The most useful line in the coverage is a simple one: AI can expose weaknesses in a business that were already there. That's not a knock on AI — it's just how amplification works. If your customer data lives in three different places and nobody's sure which one is current, an AI tool doesn't fix that. It just works faster with whichever copy it happens to find, and it does that inconsistently across whichever tools you've connected it to.

Before AI, a messy CRM or a spreadsheet-based process was slow but survivable — a person could quietly compensate for the mess. AI tools don't compensate. They take the input you give them at face value and produce output at speed. Feed a sales assistant three different versions of a customer's contact history, and you get confidently wrong answers delivered faster than a human ever would have delivered them.

A Readiness Check Before Your Next AI Purchase

You don't need a formal audit or a consultant to get ahead of this. A few hours with the following checklist will tell you more about your actual AI readiness than any vendor demo will.

1. Map where your customer data actually lives

List every place a customer's name, contact info, or purchase history is stored — your CRM, your accounting software, a shared spreadsheet, an inbox. If you can't produce this list in fifteen minutes, that's the finding. Any AI tool you connect to "the business" is really only connecting to one of these, and the others will quietly drift out of sync.

2. Check how your systems actually talk to each other

Not how they're supposed to talk to each other — how they actually do. A lot of SMB tech stacks are held together by someone manually re-typing or re-exporting data between two tools that were never properly integrated. That manual step is invisible until an AI tool is added on top, at which point it becomes the bottleneck that determines whether the AI's output is trustworthy.

3. Count the AI tools already running in your business

Not just the ones you approved — the ones your team has quietly started using on their own. If you haven't asked, assume the real number is higher than you think. This is worth doing before you add one more tool to the pile, because a sixth unmanaged AI tool creates more risk than a fifth well-integrated one creates value.

4. Budget for the plumbing, not just the subscription

The obstacles owners cite most often aren't mysterious: software costs, staff training, system upgrades, and uncertainty about implementation. Notice that only one of those four is the AI subscription itself. The other three are the work of making an AI tool fit into how your business actually runs, and they deserve a line item, not an afterthought.

5. Name one person who owns AI tool decisions

Even if that person is you, and even if it's a part-time responsibility, someone needs to be the one who asks "does this need to connect to something we already have?" before a new tool gets approved. Without that, tools accumulate independently, each one solving a narrow problem while adding to the tangle underneath.

A Simple Test Before You Buy the Next Tool

Before subscribing to the next AI tool that promises to save your team time, ask three questions:

Does it need data from a system you already use? If yes, find out exactly how that connection works before you buy, not after.
Who owns that connection once it's built? Integrations quietly break when software updates. Someone needs to notice when they do.
What happens to your data and workflow if this vendor disappears or changes terms in six months? Small AI vendors get acquired, pivot, or shut down features regularly. Don't build a process you can't unwind.

If you can't answer all three in a few minutes, that's not a reason to skip the tool — it's a reason to spend fifteen more minutes finding out before you commit your team's workflow to it.

The Bottom Line

The businesses getting real value from AI right now aren't necessarily the ones using the most tools. They're the ones whose underlying data and systems can actually support what the tools promise. That's not a reason to slow down on AI — 85% of leaders increasing spend next year suggests standing still isn't really an option either. It's a reason to spend a little less time evaluating the next flashy feature and a little more time making sure the last three tools you adopted are actually talking to each other correctly.

Set aside an hour this week to run through the checklist above. It's the cheapest AI investment you'll make all quarter, and it's the one that determines whether everything after it actually works.

Claude for Small Business Just Got 43 Workflows and 27 New Integrations — What's Actually Worth Trying

victor@daova.ai (victor@daova.ai) — Wed, 16 Sep 2026 16:06:43 GMT

On September 15, Anthropic expanded Claude for Small Business, the version of Claude aimed squarely at owner-operators rather than enterprise IT departments. The update added 43 pre-built workflows and 27 new tool integrations — Shopify, Salesforce, Xero, Gusto, Square, Stripe, Zapier, Zoom, Atlassian, and TikTok among them — plus a fall workshop tour running free half-day AI training sessions in ten U.S. cities.

If you run a 5-to-50-person business, this is worth five minutes of attention even if you never touch Claude specifically. It's a useful snapshot of where "AI for SMBs" is actually heading in late 2026, and it comes with a design detail every owner should steal regardless of which AI tool they use.

What Actually Changed

Claude for Small Business runs inside Claude Cowork, a desktop workspace where the AI can see and act on your connected files and apps rather than just chatting in a text box. The September update roughly tripled the number of ready-made workflows available — things like a weekly financial brief pulled from your accounting software, lead qualification from inbound form fills, first-draft proposals, social content calendars, and bookkeeping reconciliation.

The 27 new integrations matter more than the workflow count. Most small businesses don't want a general-purpose AI assistant; they want their existing stack — the Shopify store, the QuickBooks or Xero ledger, the Square terminal, the Gusto payroll run — to get a little smarter without switching tools. Connecting Claude directly to systems you already pay for is the difference between "AI experiment" and "AI that touches real operations."

The Detail Worth Copying: Approval Mode by Default

Every one of these workflows ships defaulting to what Anthropic calls approval mode: Claude drafts and stages the work — the email, the social post, the payment, the reconciliation entry — and waits for a human to click approve before anything actually sends, posts, or pays. Nothing fires automatically on day one.

This is the right default, and it's a useful gut-check for any AI tool you're evaluating, not just this one. If a vendor's "automation" workflow doesn't have an obvious staging or review step before money moves or a message goes out, that's a question to ask before you connect it, not after something goes out wrong.

Which Workflows Are Actually Worth Trying First

You don't need all 43. Pick one recurring task that meets three criteria: it happens weekly or more often, it's tedious rather than judgment-heavy, and it already lives inside a tool you use daily. A few that fit that bar for most SMBs:

Weekly financial brief — a plain-language summary of cash position, AR aging, and burn pulled from your accounting software, so you're not opening QuickBooks or Xero cold every Monday.
Lead qualification — a first pass on inbound form fills or chat inquiries, sorted by fit before a human follows up.
Bookkeeping reconciliation — matching transactions and flagging exceptions for your review, not auto-posting them.
First-draft proposals or social content — a starting draft you edit down, rather than a blank page every time.

Start with exactly one. Run it in approval mode for two to three weeks and actually read what it produces before deciding whether to loosen the leash or drop it.

Questions to Ask Before You Connect Your Stack

Whether it's Claude, a competitor, or a workflow tool your accountant recommended, the integration list is the real decision point — connecting AI to payroll, payments, or customer data is a different risk tier than connecting it to a shared drafts folder. Before you flip on any integration, get straight answers to:

What data does this specific integration read, and can I scope it to less than "everything"?
Is approval-before-action the true default, or something I have to configure myself?
Is there a log of what the AI drafted, approved, and sent, that I can actually pull up later?
What happens to my data if I cancel — is it deleted, retained, or used to train anything?

None of this is unique to Anthropic. It's the standard due-diligence list for connecting any AI tool to a system that touches money or customer data, and it's worth keeping on hand for the next vendor pitch too.

The Free Training, If You'd Rather Not DIY It

Alongside the product update, Anthropic and training partner Tenex are running the Claude SMB Tour again this fall: free half-day, in-person AI workshops for local business owners in Boston, Pittsburgh, Detroit, Minneapolis, Phoenix, Memphis, Savannah, Bentonville, Tampa, and Raleigh, capped at around 100 attendees per stop. A separate webinar series running through November covers individual connectors — Shopify, Xero, and the rest — in more depth, with a live workflow run during each session.

If you're near one of those cities, it's a low-cost way to get hands-on before committing budget, and the fluency you build there — how to scope an integration, how to read an approval queue, what questions to ask a vendor — transfers directly even if you end up choosing a different AI tool entirely.

The Takeaway This Week

You don't need to adopt Claude for Small Business specifically to get value from this update. Three things worth doing regardless of which tools you use:

Pick exactly one tedious, recurring, low-judgment task and see whether an AI workflow can draft it for your review — not run it unsupervised.
Before connecting any AI tool to accounting, payroll, or customer data, get a straight answer on data scope, approval defaults, and audit logs.
If free, local, hands-on AI training is available near you this fall, go. It's cheaper than trial-and-error on your own systems.

The pattern behind this release — more integrations into tools you already own, staged approval instead of full autonomy, and free in-person training to close the skills gap — is likely to keep showing up across other vendors this year. Knowing what to ask for now saves you from having to catch up later.

Microsoft Merged Its Two Copilot Apps and Cut Features — What SMBs on Microsoft 365 Should Check This Week

victor@daova.ai (victor@daova.ai) — Mon, 14 Sep 2026 16:07:20 GMT

If your business runs on Microsoft 365, there's a change worth five minutes of your attention this week. Microsoft has been merging its two separate Copilot apps — the free consumer one and the Microsoft 365 Copilot app your business may be paying for — into a single app with one name, one icon, and one web address. The mobile and web versions switched over starting in mid-August. The Windows and Mac desktop apps are converting this month. Along the way, Microsoft quietly killed off three features and moved a fourth behind a paywall, with little fanfare beyond a support page and a few tech outlets picking it up.

None of that is dramatic on its own. But it's a clean, current example of something every small business using an AI-powered software subscription should get in the habit of watching for: the tool you signed up for six months ago is not a fixed thing. It gets reshaped, quietly, on the vendor's schedule — and if you're not paying attention, you find out what changed only after you needed the feature that's gone.

What Actually Changed

Strip away the branding language and three concrete things happened:

Two apps became one. The consumer Copilot app and the Microsoft 365 Copilot app — previously separate downloads with separate purposes — are merging into a single "Copilot" app. Mobile and web rolled out first; Windows and Mac are converting through mid-September.
Three features were cut outright on August 18. Group Chat (multi-person conversations inside Copilot, including any shared messages and images sitting in them), the AI-generated Podcasts feature, and Mico, the animated on-screen character in voice mode, all stopped working — with existing Group Chat content wiped, not archived.
Deep Research moved behind a paywall. A feature that was previously accessible now requires a paid tier to use.

Microsoft's own framing is that this is a step toward a "Super App" — one place that eventually brings together chat, AI coding tools, its Cowork collaboration features, and new autonomous "AutoPilot" agents. That's a reasonable long-term direction for Microsoft. It's also, from where a small business sits, a reminder that the roadmap is being written by the vendor, not by you.

Why This Isn't Just a Microsoft Story

If you don't use Microsoft 365 Copilot, the specific feature list above doesn't matter to you. The pattern underneath it does, because it will repeat with whatever AI tools you do rely on — an accounting AI feature, a marketing assistant, a scheduling agent. Subscription software that includes an "AI" component is being iterated on faster and more aggressively than the software you're used to, because every vendor is racing to ship agent features and consolidate their products around them. That means:

Feature availability is less stable than it used to be

Traditional software changes slowly — a new version once a year, clearly documented. AI features are being added, removed, merged, and repriced on a rolling basis, sometimes with a single support-page update as the only notice. A capability your team built a habit around can disappear in a release you never saw announced.

"Free" and "included" are moving targets

Deep Research wasn't a paid add-on until it was. This is a version of the same pattern seen in QuickBooks bundling a price increase with a new AI Payments Agent: features that ship as part of your subscription today can get walled off tomorrow, and the only way you find out is by hitting the wall.

Shared content living inside an AI tool is not automatically safe

The Group Chat wipe is the detail worth sitting with. If your team used it to collaborate — pasting drafts, sharing generated images, working through a problem together — that content is gone, not exported somewhere for you to retrieve. Any AI tool your team uses for real work should be treated the same way you'd treat a shared drive: assume it can change format or disappear, and don't let it become the only copy of something you need.

What to Do This Week

1. Check what your team actually uses in Copilot

If anyone on your team has Microsoft 365 Copilot, ask them directly: did you use Group Chat, Podcasts, or Deep Research? If the answer is yes, find out now whether they lost something, rather than discovering it mid-project. This is a two-minute Slack message, not a research project.

2. Re-run the "is this still worth the seat price" question

Microsoft 365 Copilot is a per-seat add-on most businesses pay extra for on top of their existing 365 subscription. A consolidation like this is a natural trigger to ask whether it's still earning that cost for the people using it — not because anything is necessarily wrong, but because "we bought it and forgot to check" is how software budgets quietly bloat. If Deep Research was the main reason someone had a license and it's now paywalled at a higher tier, that's worth a real look rather than an automatic renewal.

3. Stop treating any AI tool as safe long-term storage

Wherever your business uses AI chat or collaboration features — Copilot, ChatGPT, Claude, Gemini, or anything else — make sure anything that actually matters (a finished draft, a decision, a summary someone will need later) gets copied into a real document or system of record. Treat the AI tool as where the work happens, not where it lives.

4. Put a quarterly "what changed" check on the calendar

You don't need to track every AI vendor's release notes in real time. You do need a recurring 15-minute slot — quarterly is enough — where someone checks the tools your business actually pays for and asks: what got added, what got removed, and what got moved behind a paywall since we last looked. That single habit would have caught this Copilot change before it caught your team by surprise.

The Bottom Line

Microsoft merging two apps into one isn't itself a big deal. The bigger deal is what it represents: AI features embedded in the software you already pay for are being reshaped faster than most businesses are used to tracking, sometimes with real consequences — lost content, a feature quietly repriced, a capability your team depended on gone. You don't need to chase every vendor announcement. You do need a habit of checking in on the tools you're actually paying for, on a schedule you control rather than one dictated by whichever change catches you off guard next.

DAOVA helps small businesses build the habits and governance to manage a fast-changing AI toolset without getting blindsided by it. Explore AI Enablement.

Mastercard's Agent Connect Lets AI Shopping Agents Pay You Directly — Here's What to Ask Before You Turn It On

victor@daova.ai (victor@daova.ai) — Fri, 11 Sep 2026 16:06:40 GMT

On September 9, 2026, Mastercard launched Agent Connect, a single integration that lets AI shopping assistants — the kind built into ChatGPT, Gemini, Perplexity, and similar tools — discover a merchant's products, build a cart, and complete a customer-approved purchase, all without that merchant having to build a separate technical connection for every AI platform that wants to sell on its behalf. Mastercard paired it with an upgraded Agent Suite for Merchants and a new partnership with Anthropic, bringing Claude models into post-purchase tasks like order tracking and refunds. Visa has been running a similar product, Intelligent Commerce Connect, in pilot since April. Put simply: the card networks just started building the plumbing for AI agents to pay merchants directly, not just recommend them.

What Mastercard Actually Launched

One integration, not a project for every AI platform

Before Agent Connect, a merchant wanting to accept a purchase initiated by an AI shopping agent would, in theory, need to work out a separate technical relationship with each AI platform — one for ChatGPT, a different one for Gemini, another for whatever comes next. Agent Connect is Mastercard's attempt to collapse that into a single connection point: plug in once, and the purchase flow works across participating AI agents and platforms. The initial U.S. network listed alongside the launch includes payment and technology partners like Global Payments, Getnet, Network International, Nexi, and Samsung, with Fiserv, FreedomPay, Moneris, and Worldline also named among the broader Agent Suite for Merchants partners.

This is infrastructure your processor plugs into — not something you build

Notice who's actually named in that list: payment processors, gateways, and point-of-sale providers. Not individual small businesses. That's the important detail for an SMB owner reading the headline. You are not expected to integrate with Mastercard's agent commerce API yourself, the same way you've never had to integrate with Visa's card network directly — your payment processor or e-commerce platform does that, and you inherit the capability (or don't) based on whether they've signed on. The practical question for most SMBs isn't "how do we build this," it's "does our processor even offer this yet."

The Risk Surface Nobody's Fully Answered Yet

"Customer-approved" is carrying a lot of weight in that sentence

The pitch is that these are "customer-approved payments" — the shopper authorizes the agent to buy on their behalf, similar to approving a payment on a phone. But the dispute mechanics of that authorization are still being worked out in public, in real time, across a brand-new type of transaction. If an agent misreads a customer's intent, selects the wrong size or quantity, or gets talked into an upsell the customer didn't actually want, who absorbs that: the merchant, the card network, or the AI platform that ran the agent? Traditional chargeback rules were written for a person clicking "buy," not for a delegated AI intermediary acting on loosely specified instructions. Mastercard and Visa are building the rails first; the liability case law is going to be argued out afterward, one dispute at a time.

Fraud patterns will move faster than the safeguards

Any new payment channel attracts new fraud attempts before the rules catch up to close the gaps — that's been true of every payment innovation from card-not-present transactions to mobile wallets, and there's no reason agentic commerce will be different. A compromised AI agent account, a manipulated shopping assistant, or a scripted flood of "agent-initiated" test purchases are all plausible early abuse patterns. None of this means the technology is unsafe to eventually use — it means it's early, and early is exactly when a small business should be asking questions rather than opting in by default.

What to Actually Do This Month

1. Ask your payment processor or POS provider directly

Whether you use Square, Stripe, Shopify Payments, Clover, or a regional processor, a five-minute email or support call will tell you whether they're part of Mastercard's Agent Connect network or Visa's Intelligent Commerce Connect, and what their rollout timeline looks like. Most SMBs will find the honest answer today is "not yet" — which is useful information on its own.

2. Don't rush to opt in the moment it's offered

When your processor does enable this, you'll likely get a setting to turn on. Treat it the way you'd treat any new payment method: read what it actually changes about your dispute and chargeback exposure before flipping it on, rather than accepting the default because it sounds forward-thinking.

3. Get specific with your processor about dispute handling

Ask directly: if a customer disputes a purchase their AI agent made, whose policy governs the resolution — the card network's standard chargeback process, or something new and agent-specific? If your processor can't answer that clearly yet, that's a reason to wait, not a reason to assume it's fine.

4. Keep your product and pricing data clean either way

Whether or not you ever accept an agent-initiated payment, AI shopping tools are already reading your product listings, hours, and pricing to make recommendations. Accurate, current data reduces the odds of a mismatch between what an agent thinks it's buying and what you actually sell — which matters for disputes regardless of which payment rail the purchase runs through.

5. Put a 90-day check-in on your calendar

This space will look different by December than it does today. Set a reminder to check back with your processor once a quarter rather than trying to track every announcement from Mastercard, Visa, and the AI platforms in real time — the practical details that matter to you will settle out over the next few rollout cycles.

The Bottom Line

Agent Connect is a genuinely new piece of financial infrastructure, not another AI feature announcement — it changes who can initiate a payment at your business, and the rules for what happens when something goes wrong aren't fully written yet. For most SMBs, there's nothing to build and nothing urgent to decide this week. There is, however, a short list of good questions worth asking your payment processor now, so that when this capability does show up as a setting in your dashboard, you're deciding with information instead of just clicking "enable."

DAOVA helps small businesses cut through AI hype like this and focus on the practical steps that actually move the needle. Explore AI Enablement.

The EU AI Act's Chatbot Disclosure Rule Just Took Effect — Does It Apply to Your Business?

victor@daova.ai (victor@daova.ai) — Wed, 09 Sep 2026 16:09:04 GMT

On August 2, 2026, a new transparency rule under the EU AI Act quietly became enforceable — and it's one most small business owners in the United States have never heard of, even though it may already apply to them. Article 50 of the Act requires that anyone interacting with an AI chatbot be told, clearly and up front, that they're talking to a machine. It also requires AI-generated images, audio, and video to be labeled. If your website has a chat widget and any visitors from Europe, this law may already cover you — regardless of where your business is headquartered or how many people you employ.

What Article 50 actually requires

The rule itself is narrower than "AI regulation" headlines make it sound. It doesn't restrict what your chatbot can say, require a license, or demand a government filing for most small businesses. It requires two specific things:

Chatbot disclosure: A person has to be told they're interacting with AI, no later than their first message, unless it's already obvious from context. A line buried in your terms and conditions doesn't count — the disclosure has to be visible inside the conversation itself.
AI content labeling: Images, audio, or video that are AI-generated or AI-manipulated need to be marked as such when you publish them, particularly anything that could otherwise pass as authentic.

For most small businesses, the chatbot piece is the one that matters day to day — because a website chat widget, a support bot, or an AI-powered booking assistant is exactly the kind of tool a lot of SMBs added over the past two years without thinking of it as a compliance question.

Why this applies to you, even if you've never sold to Europe

The EU AI Act follows the same "long-arm" logic as GDPR: it doesn't matter where your company is registered, it matters who's on the other end of the interaction. If a visitor based in the EU lands on your site and chats with your AI assistant, that interaction falls inside the Act's scope — even if your business is a five-person shop in Ohio that has never shipped a product overseas.

This is easier to trigger than it sounds. A services firm with one client in Germany. An e-commerce store that ships internationally, or simply doesn't block EU traffic. A SaaS tool with a handful of European sign-ups from an ad campaign you ran once. A consulting practice whose website chatbot answers questions from anyone who finds it through search, EU-based or not. None of these businesses think of themselves as "operating in Europe," but their chatbot might already be.

Provider vs. deployer: which one are you?

Most SMBs aren't building their own AI models — they're using a chatbot from a vendor (a website chat platform, a helpdesk tool, a booking assistant). In the Act's language, the vendor is usually the "provider" and you're the "deployer." Deployers still carry responsibility, especially if you've configured or customized the bot for your own use, which nearly every business does. Practically, that means you can't assume your vendor has this handled — you need to confirm it.

What happens if you ignore it

The headline fine for Article 50 violations is up to €15 million or 3% of global annual turnover, whichever is higher. That number is aimed at large companies. A separate provision, Article 99(6), caps penalties for small businesses and startups at whichever of the two figures is lower — a meaningful difference, though still not something worth testing. Realistically, enforcement against a small business is far more likely to start with a customer complaint or a routine audit than a proactive sweep. But the exposure is real, it's new, and it's the kind of thing that's cheap to fix now and expensive to explain later.

A same-week fix: the 4-step checklist

1. Inventory your AI touchpoints

List every place AI talks to a customer on your behalf: website chat widget, AI phone or voice assistant, automated email responder, booking or scheduling bot. Most businesses find two or three; some find more than they expected once they actually look.

2. Add a clear, upfront disclosure

No certification or specific legal wording is required. A first message along the lines of "You're chatting with our AI assistant" or a persistent label like "AI Assistant" next to the chat window satisfies the requirement — as long as it's visible in the interaction itself, not tucked into a privacy policy.

3. Check your vendor's settings, don't assume

If you're using a white-label or third-party chat platform, log in and look for a disclosure setting. Many platforms added one this year in response to this exact rule; some still default it to off. Turning it on is usually a five-minute fix once you find it.

4. Keep a one-page record

Write down what AI tools you use, where they're deployed, and what disclosure each one shows. This isn't a legal filing — it's a plain internal note that proves you thought about it, dated, and kept somewhere you can find it if anyone ever asks.

Part of a bigger pattern, not a one-off rule

Article 50 isn't an isolated EU quirk. More than a dozen U.S. states have passed or advanced their own AI chatbot disclosure laws in 2026, built around the same basic idea: customers shouldn't have to guess whether they're talking to a person or a program. A few of those state laws go further and let consumers sue directly, with statutory damages per violation. The specific legal triggers differ state to state and from the EU rule, but the underlying expectation is converging fast — disclose that it's AI, plainly, at the start of the conversation. Treating that as a baseline practice for every AI touchpoint you run, not just the ones a lawyer flags, is the simplest way to stay ahead of wherever the next version of this rule shows up.

The practical takeaway

None of this requires a compliance department or outside counsel for the vast majority of small businesses. It requires about an hour: list your AI touchpoints, check what each one discloses, fix the ones that don't, and write down what you did. The businesses that get caught off guard by rules like this usually aren't the ones running risky AI — they're the ones who never realized a rule like this existed until it was already a problem.

DAOVA helps small businesses build practical, right-sized AI governance practices that keep pace with a fast-moving regulatory landscape. Explore AI Governance & Operations.

OpenAI Cut Cursor Off From Its AI Models — Here's the Vendor-Risk Lesson for Your Business

victor@daova.ai (victor@daova.ai) — Mon, 07 Sep 2026 16:09:06 GMT

On August 14, 2026, SpaceX closed a $60 billion acquisition of Anysphere, the company behind Cursor — one of the most widely used AI coding tools, and by most accounts one of the fastest-growing software companies in history. Two weeks later, OpenAI told Cursor it was walking away. OpenAI notified SpaceX that it intends to wind down the contract supplying its models to Cursor, with a shutoff date of November 12, 2026. The reason wasn't a service failure or a missed payment. It was a change-of-control clause: the contract let OpenAI exit if Cursor's ownership changed hands in a way OpenAI didn't like, and OpenAI has said it doesn't trust that SpaceX will use its technology in line with its terms of service.

If you don't write code and have never heard of Cursor, it's tempting to file this under "tech industry drama." That would be a mistake. The mechanism here — a vendor cutting off access to the AI your product runs on, on contractual notice, for reasons that have nothing to do with how well the product works — isn't unique to Cursor. It's a risk sitting quietly underneath a lot of the AI tools small businesses have adopted over the past two years, most of which nobody has ever stopped to think about.

Why this isn't just a Cursor problem

Anysphere wasn't a struggling startup that lost leverage. It was valued in the tens of billions of dollars, growing faster than almost any software company on record, with deep pockets and a sophisticated legal team. And it still couldn't stop its core AI supplier from pulling the plug on about 90 days' notice, over a dispute about who now owns the company — not over anything Cursor itself did wrong.

Scale down from a $60 billion acquisition to the reality of a small business, and the same structural risk is still there, just with far less leverage to negotiate around it. Most SMBs don't have a contract with OpenAI, Anthropic, or Google at all. They have a contract with a CRM, an accounting platform, a scheduling tool, or a customer-service inbox that happens to run an AI feature underneath — and that vendor's relationship with its own model supplier is invisible to you until something changes it.

The dependency you probably haven't mapped

Most of the AI capability small businesses use today doesn't come from a direct account with an AI lab. It comes bundled inside software you already pay for: an AI drafting feature in your email tool, an AI agent in your accounting software, an AI assistant built into your CRM. Each of those features is a small business (the vendor) that itself depends on a model provider, a hosting arrangement, and a contract that can change. When that underlying arrangement shifts — a price increase, a policy change, an acquisition, a shutoff — the change reaches you as a surprise, because you were never party to the relationship that broke.

This is the flip side of how convenient AI has become. Because these features arrive already integrated, it's easy to treat them as a permanent part of the tool rather than what they actually are: a layered dependency that can be renegotiated, restricted, or removed by parties you have no relationship with.

A practical vendor-continuity checklist

None of this means avoiding AI tools or over-engineering a small business's tech stack. It means treating AI-dependent tools the way you'd treat any other single point of failure — with a basic continuity habit, not a legal department.

1. Know what's underneath your critical tools

For the two or three AI-powered tools your business genuinely couldn't operate without, ask the vendor directly: which model provider powers this feature, and is it exclusive or swappable on your end? Most vendors will answer a direct question like this, and the answer tells you how exposed you are if that one relationship changes.

2. Don't let one vendor's AI feature become the only place a workflow lives

If an AI feature drafts your customer replies, generates your invoices, or triages your inbox, make sure the underlying process — the templates, the rules, the logic — also exists somewhere you control, not only inside that one tool's AI layer. That way, if the AI feature disappears or degrades, the workflow keeps running manually or on a different tool while you sort out a replacement.

3. Keep your own copy of the data and outputs

Export or back up what an AI tool produces for you — customer summaries, generated content, automated reports — on a regular schedule, not just when something breaks. A vendor cutoff rarely deletes your historical data overnight, but access can be interrupted with far less notice than you'd like, and having your own copy means a bad week doesn't become a bad quarter.

4. Read the termination and change-of-control language before you sign, not after

Most SMBs never read past the pricing page on a software contract. For any AI tool that becomes core to how you operate, it's worth five minutes to check what happens if the vendor is acquired, discontinues the AI feature, or changes its underlying model provider. You're not trying to predict the future — you're trying to know your notice period if it happens.

5. Write down a "day after" plan for your one or two most critical AI tools

Pick the single AI tool your business would feel the most pain from losing with 90 days' notice. Spend fifteen minutes writing down what you'd do: which alternative tool you'd trial first, what would revert to manual, who on your team would own the transition. You don't need this plan for every tool you use — just the one or two that would actually hurt.

The bigger picture

The OpenAI-Cursor story is a useful jolt precisely because it happened to a company with every advantage a small business doesn't have — money, lawyers, and leverage — and it still couldn't prevent a 90-day notice to rebuild a core dependency. That's not a reason to slow down AI adoption; the productivity gains are real and worth pursuing. It's a reason to treat "which AI tools would hurt if they disappeared" as a question worth answering once, in a calm moment, rather than for the first time during a scramble.

DAOVA helps small businesses build practical AI governance that accounts for vendor and continuity risk, not just data security. Explore AI Governance & Operations.

AI Mode Is Changing How Customers Find Local Businesses — Here's How to Stay Visible

victor@daova.ai (victor@daova.ai) — Fri, 04 Sep 2026 16:06:58 GMT

Type a question like "best plumber near me" or "does [business] deliver on weekends" into Google today, and there's a good chance you won't see the familiar list of blue links at all. You'll see an AI-written answer, sometimes naming one or two businesses by name, sometimes naming none. New research on Google's AI Mode, published through 2026, puts numbers behind a shift small business owners have likely already felt: roughly 93% of AI Mode sessions end without a click to any external website, and AI-generated local results surface only about 32% as many unique businesses as the traditional map pack did. Search hasn't gone away. But for a growing share of searches, the AI is now the whole interaction — and it either mentions your business or it doesn't.

What's Actually Changing

AI answers are replacing the click, not just sitting above it

AI Overviews and AI Mode aren't a new box at the top of a familiar results page anymore — for a large share of "what," "who," and "which one" questions, they're becoming the entire answer. A behavioral study tracking tens of thousands of real Google searches found people click a traditional result about half as often when an AI summary appears. In AI Mode specifically, the vast majority of sessions never reach an external site at all. The customer gets their answer, and your website — even a well-built one — may never enter the picture.

Fewer businesses get named, and the criteria are different

A traditional map pack could show three businesses plus a "more" link to a dozen others. AI Mode often names one, or two, by whatever criteria it's weighing that moment. A separate industry study found that brands the AI actually cites earn meaningfully more of both organic and paid clicks than businesses it doesn't mention for the same query. In other words: getting cited isn't a nice-to-have anymore. For a real slice of your local customers, it's the new front door, and there isn't a "page two" to fall back on if you're not on it.

Google Business Profile signals still matter most — but they're not the whole story

According to the latest local search ranking factors research, a well-maintained Google Business Profile remains the single most heavily weighted signal for local visibility, AI Mode included. But AI systems are also pulling from a wider net than the map pack ever did — your website's own content, review platforms, directories, and community discussion — looking for consistent, specific, and current information it can confidently repeat.

Why This Matters for a Business Like Yours

This isn't really a marketing-budget problem

It's tempting to read "AI Mode changed local search" as a signal to spend more on ads. That's mostly the wrong lever. What's actually being rewarded here is accurate, specific, consistent information the AI can confidently repeat without checking with you first — a data and content discipline problem, not a spending problem. A business with modest marketing spend but clean, complete, consistent listings will out-perform a business throwing money at ads while its hours, service list, or address are wrong in three different places online.

"We rank well" and "the AI recommends us" are no longer the same question

You can still show up reasonably in traditional results while being invisible in an AI answer, because the AI is optimizing for something slightly different: it wants to give a confident, defensible answer, and it favors sources it can verify and trust. If your information is thin, inconsistent, or scattered, the AI doesn't call you to check — it just recommends whichever competitor gave it a clearer, more confident answer to work with.

Zero-click doesn't mean zero-value, but it does mean new metrics

If a customer gets your business's hours and a strong recommendation directly from an AI answer and simply shows up or calls, that's a good outcome for you even though it won't show up as a website click. The uncomfortable part is that most small businesses are still only measuring website traffic, so this kind of win is currently invisible in their own reporting.

What to Actually Do About It

1. Ask the AI what it currently says about you

Open Google (in AI Mode), ChatGPT, and Gemini, and ask each one plain questions a customer would ask: your hours, whether you offer a specific service, your pricing, your location. This takes about fifteen minutes and tells you immediately whether you have a visibility problem, an accuracy problem, or neither.

2. Make your Google Business Profile complete, not just claimed

Fill in every relevant field — categories, services, attributes, Q&A, current hours including holiday exceptions — and keep it current. This remains the top-weighted signal for local AI visibility, and it's also the field most small businesses set up once and never revisit.

3. Keep your name, address, and phone number identical everywhere

Your website, Google Business Profile, Yelp, industry directories, and any other listing should say exactly the same thing. Inconsistency is one of the clearest signals that tells an AI system your information isn't reliable enough to repeat confidently.

4. Write your website like it's answering a direct question

Pages that clearly state what you offer, to whom, at what price, and where — in plain sentences, not just a photo gallery or a PDF menu — are far easier for an AI system to extract and repeat accurately. If a fact about your business only exists as an image or a scanned document, treat that as a gap to close.

5. Collect reviews and mentions on other people's platforms, not just your own

AI systems weigh third-party evidence — reviews, directory listings, local coverage — alongside what your own website says. A steady stream of specific, recent reviews mentioning what you actually do gives the AI independent confirmation, which is exactly what it's looking for before it's willing to recommend you by name.

The Bottom Line

None of this requires a new AI tool, a redesigned website, or a bigger ad budget. It requires treating your business's basic information — hours, services, pricing, location, and what customers say about you — as infrastructure that AI systems now depend on to make a recommendation, not paperwork to fill out once and forget. The businesses that show up when a customer asks an AI for a recommendation won't be the flashiest ones. They'll be the ones whose information was accurate, consistent, and easy to verify when it mattered.

DAOVA helps small businesses get their fundamentals right for an AI-mediated world — from how they're found to how they operate. Explore AI Enablement.

QuickBooks Raised Prices and Bundled In an AI Payments Agent — Here's How to Decide If It's Worth It

victor@daova.ai (victor@daova.ai) — Wed, 02 Sep 2026 16:08:13 GMT

On August 1, 2026, QuickBooks Online raised prices across its Essentials, Plus, and Advanced plans — Plus went from $115 to $140 a month, Advanced jumped from $275 to $340. Intuit's explanation wasn't just "costs went up." It was bundled with new capabilities, most notably an AI-powered Payments Agent that watches your invoices, predicts which customers are likely to pay late, drafts personalized reminder emails before they do, and suggests customizable late fees based on each customer's payment history. Intuit says it gets users paid roughly five days faster on average.

If you use QuickBooks, this is worth five minutes of attention on its own. But the more useful story here isn't really about QuickBooks. It's about a pattern that's about to become routine: software vendors bundling price increases with "AI-powered" features, banking on the AI label making the higher bill feel justified without anyone actually running the math. This is the first clean example of that pattern landing squarely in small business software. It won't be the last.

What actually changed

Strip away the marketing and the QuickBooks update has three real pieces:

A price increase that applies at renewal — Essentials, Plus, and Advanced all went up, while the Free, Lite, Ledger, and Simple Start tiers were left alone.
A new AI Payments Agent that works your accounts receivable in the background: flagging invoices at risk of going late, drafting (not just sending) reminder language, and recommending late-fee terms based on a customer's own history.
Other bundled additions at the higher tiers — expanded bill pay, reconciliation tools, and industry-specific features — that have nothing to do with AI at all.

Notice that the AI feature is one line item inside a broader bundle. That's the part worth paying attention to, because it's also the part most owners will skip past. When a renewal notice says "new AI capabilities," it's easy to read that as the whole story and either shrug ("fine, everyone's doing AI now") or bristle ("another AI tax") without actually asking whether the specific feature is worth the specific price increase for your business.

The question that actually matters

"Is AI worth paying for" isn't a real question — it's too broad to answer. The question that's actually answerable is: is this specific feature worth this specific price increase, for my specific business, this year? That's a math problem, not a philosophy debate, and it's one every owner can run in about fifteen minutes.

1. Put a number on the vendor's own claim

Intuit's pitch is "get paid five days faster." Don't take that as a vibe — price it. If you invoice $60,000 a month and typically wait 35 days to collect, shaving five days off that cycle frees up roughly $10,000 of cash that would otherwise be sitting in accounts receivable. For a business tight on working capital, that's real money and probably worth more than a $25/month plan bump. For a business that already collects on time and pays its own bills from a comfortable cash cushion, five faster days is nice but not decision-driving. Do this arithmetic with your own invoicing volume and average days-to-pay before you decide anything.

2. Check whether you can get most of the value for free

A meaningful share of what "AI payment reminders" do — nudging customers before they're late, offering a one-click pay link, applying a consistent late-fee policy — can be approximated with a scheduled reminder email and a written late-payment policy you enforce yourself. It won't be as personalized or as automatic, but if your invoice volume is low (a dozen invoices a month, say), the manual version may get you 80% of the benefit at 0% of the price increase. The AI feature earns its cost at volume, where personalized, automatic follow-up starts saving real staff time — not necessarily at every business size.

3. Pilot it for one billing cycle before assuming you need the higher tier

You don't have to commit to a year of the pricier plan to find out if the feature works for you. Turn it on, run it through one full invoicing cycle, and look at what actually happened: did flagged invoices genuinely get paid faster, or did the agent flag the same handful of slow-paying customers you already knew about? A short, deliberate trial beats assuming the sales page description matches your experience.

4. Set a renewal review date instead of letting it auto-apply

The easiest mistake here is doing nothing — letting the higher price take effect automatically because reviewing it felt like a hassle. Put your software renewal dates on a calendar with a 15-minute review attached: what changed, what it costs now, and whether the new capability solves a problem you actually have. This will keep happening across your other subscriptions too, not just accounting software, so the habit is worth building once rather than relearning every time a renewal email lands.

What to do this week

If you use QuickBooks, check which plan you're on and whether your next renewal falls after August 1, then confirm the new rate for that plan directly in your account. If you're not sure the Payments Agent applies to your tier, it's worth a quick look in the app rather than assuming.

More broadly, whatever accounting, CRM, or scheduling software your business runs on, expect a version of this same announcement sometime in the next year: a price increase, wrapped around one or two genuinely useful AI features and a handful of unrelated additions you may never touch. When it arrives, resist answering "is AI worth it" in the abstract. Pull up your own numbers — your invoice volume, your average collection time, your actual pain point — and answer the narrower question instead. That fifteen minutes of arithmetic is the difference between paying for a tool that fixes something real and paying a quiet AI premium because the alternative felt like more effort than it was worth.