We Sell What We Know: How to Test an AI Partner's "We Use It Ourselves"

In March 2024 the US Securities and Exchange Commission (SEC) announced settled charges against two investment advisers, one of them based in Toronto, over false and misleading statements about their use of AI. The penalties came to $400,000 in total.¹ In June 2025 the research firm Gartner estimated that only about 130 of the thousands of agentic AI vendors (vendors selling AI agents, software that does part of a task on its own) are real, and called the rest "agent washing".² Be exact about what those two facts are. One is a regulator's finding about investment advisers, the other an analyst's estimate about software vendors, and neither is about consultancies as a group. Together they show why any claim about AI use is worth checking, including the one an AI consulting firm makes to you about itself.

"We use it ourselves" is a claim, and a claim can be tested. By "use" we mean a task done on working days, by an AI agent, with a person accountable for what goes out, and that definition is ours, not a standard. If you are the owner or CEO, the COO or head of operations, or the CFO or controller, and a provider has said it to you, this article gives four requests that test it, then answers them for Houseblend as far as we honestly can.

If you would rather put the four requests to us directly, a free 30-minute Strategy Call with an AI adoption specialist is a good place to do it, and it ends with the first two or three workflows worth automating.

01. Why "we use it ourselves" is easy to say

The scene. Three providers in a row tell you they run on what they sell. All three sound the same, and nothing in a pitch tells you which one means it. Maybe you bought AI licences and few people use them, or a board asked for an AI plan, and now you are choosing who helps.

What you need from the choice depends on your role:

• Owner or CEO: the question is what to say to the board.
• COO: the question is who will do the daily work.
• CFO or controller: the question is what is being represented to you, and what happens if it is wrong.

If you have wondered whether your AI partner uses AI itself, this is the section for you.

What regulators have done with the claim. Regulators call overstating AI use "AI washing": describing a product or service as using AI, or as doing more with AI than it does. The SEC's March 2024 release says the Toronto-based adviser made false and misleading statements from 2019 to 2023, in SEC filings, a press release and on its website, about its use of AI and machine learning on client data, and that the other adviser falsely claimed in 2023 to be the "first regulated AI financial advisor". Both settled without admitting or denying the findings.¹ Those are investment advisers, a regulated setting, and they show one principle: "we use AI" is a statement of fact, so it can be false.

The pattern is not limited to finance. In September 2024 the US Federal Trade Commission (FTC) announced actions against deceptive AI claims. In one, the complaint alleges that a company offering a legal-services chatbot did not test whether its output equalled a human lawyer's and did not hire or retain any attorneys. The proposed order would require a $193,000 payment and notices to subscribers about the service's limitations.³ That is a consumer product and an allegation in a complaint, not a finding. Gartner's name for the vendor version of the same problem is agent washing: relabelling existing products, such as AI assistants, robotic process automation (software that repeats fixed steps) and chatbots, as AI agents without substantial agentic capabilities.²

In Canada. The Competition Act is the text to know. Its section 74.01 makes it reviewable conduct (a civil, not criminal, matter) to make a representation to the public that is false or misleading in a material respect, and to make a statement about how well a product performs that is "not based on an adequate and proper test", with the proof on the person making it. The Act says a product includes a service.⁴

The Competition Bureau's own pages say the Act already prohibits representations to the public promoting a product that are false or misleading in a material respect, and report broad agreement among consultation respondents that AI can amplify traditional deceptive marketing practices.⁵⁶ In December 2024 the Canadian Securities Administrators warned investment fund managers against "vague and unsubstantiated statements that incorporate jargon in order to attract investors".⁷ A 2024 note from one Canadian law firm said the Act could reach AI washing,⁸ and a September 2026 note from another says Canadian regulators have not yet taken similar action to the SEC's.⁹ This is general information, not legal advice, and whether a particular claim about a consulting service falls under the Act is a question for a lawyer.

The turn. None of this says providers are lying. It says the sentence is cheap, which is why it is worth testing. Our view: a claim with no evidence trail is a pitch, not a credential.

02. What a buyer can and cannot learn from it

What buyers do with claims. A March 2026 survey of 1,202 US business decision makers, run by SurveyMonkey and Reddit, found that 55% find it difficult to identify trustworthy information sources and 46% find it hard to parse seller or vendor-provided information, and that 73% trust peer insights above vendor websites (55%).¹⁰ Reddit sells advertising and SurveyMonkey sells a survey platform, and the figures come from their press release, not the full study. It describes a buyer's problem more than it proves anything about AI providers.

The provider side of the same gap. The Thomson Reuters Institute's 2026 report, drawing on more than 1,500 professionals in legal, tax, accounting, risk, fraud and government, says two-thirds of corporate respondents want their outside firms to use AI, yet fewer than 20% mandate it, and most believe firms should initiate clearer AI conversations.¹¹ That is professional services in legal and tax settings, not management consultancies, and Thomson Reuters sells AI products to those professions. Buyers want providers to use AI, and few require it.

The research gap. We looked for a study showing that buyers choose, or get better results from, a provider that uses AI itself, and we did not find one. The nearest evidence is the two items above, so this article does not claim that use is a credential. It claims that use is testable, and that the test tells you more than the claim. For the owner or CEO, a test is something you can show a board, and a claim is not.

What a provider's own use can and cannot tell you. Our view: it can tell you the provider has met the failure modes of AI in its own daily work. It cannot tell you the tool fits your data, your documents or your sign-off rules, because a provider's own tickets and documents are not yours.

Gartner's August 2026 note on buying AI makes the buyer's side plain: organizations acquire AI through software, cloud platforms, infrastructure and consulting services, among other routes, and the challenges it names include inconsistent data rights protections and limited pricing transparency. It recommends company-wide rules for AI purchasing, governance and data rights.¹² If you are the CFO or controller, that is your vocabulary. If you are the COO, it is the reason to ask to see the work, not the slide. The four requests follow in section 05.

03. What we run on our own company

Where it started. Read this section as an example of the specificity you can ask any provider for. In our own words: companies were buying AI licences and nothing changed. People tried it a few times, got generic answers, and went back to doing the work by hand.

The tool was never the problem. Nobody had connected it to the company's own data, files and way of working. We had already done that for our own delivery, so we started doing it for clients. That is one of the reasons behind what makes our approach different.

Meetings to tickets. Houseblend runs its own company on AI agents, which are pieces of software that do a part of the work on their own while a person reviews the result. The clearest example is one we run daily.

Each morning an agent reads the previous day's client meeting transcripts, emails and time entries, drafts or updates the matching tickets with estimates, and posts a recap for each client to the account lead, the person who looks after that client. The ticket is the transcript's receipt: what was said on a call ends up as a ticket, not in someone's memory. Three rules come with it:

• The agent looks back over a rolling seven days, so a note that turns up late is still caught.
• A person signs off anything that goes out.
• Anything ambiguous is flagged, not written.

For the COO, that is the point: the daily rhythm has a person standing in it. For a professional services firm, the gap between what was said on a call and what is in the project tool will be familiar, and it is slow, knowledge-heavy manual work of the kind AI agents for business can take on.

No counts of accounts or agents, hours saved or accuracy figures are published for it, and none are claimed here. On your side, the question is who looks after each client, because that person should be the one who gets the recap.

The rest of what we run, one line each.

• We use a company knowledge assistant daily on our own documentation (what a company knowledge assistant is).
• We run the weekly client status report with agents, and a person reviews and sends it (how we run weekly client status reports).
• We generate proposals and design documents from templates, which a person reviews (document and reporting automation).

A short list, on purpose. That is what "we use it" means here. You can hold us to it in section 06.

04. What we know from running it, and what we do not claim

What we know, as lessons and not results. As we run these agents, trust takes three things:

• A look-back long enough to catch late items.
• A person who signs off before anything goes out.
• A default of flagging what is unclear instead of writing it.

Human sign-off is a working rule for us. Our view of why the third matters: an agent that fills a gap with a guess is the failure to design against. If you are choosing a provider, ask whether it can name its own version of these three.

A fourth lesson belongs to estimates. A known bias in an AI estimate is logged in the ticket, where the next person will see it, as we told in our introduction to AI adoption.

The same method on client work. We have used AI-assisted analysis, with a consultant reviewing it, to audit a stalled ERP implementation for a client. A consultant validates each finding before it goes to the client, and access is set up only after a non-disclosure agreement is in place.

In work with a mechanical contractor, the starting point was the contractor's own material: the drawings, the specifications, an estimating spreadsheet and a past proposal. The estimator reviews everything before it goes out. These are client work, so they show what we know, not what we use.

What we do not claim. We do not publish hours saved, accuracy rates or error counts for these agents, and we do not claim they are always right. The figures in our sample workflows are estimates, labelled typical, not measurements.

We also do not claim that our daily use proves your use case will work, since your data and sign-off rules are not ours, and what we offer is part of what we do every day. Our view: a provider who cannot tell you what it does not claim has not thought about it. For the CFO or controller, a claim with no figures attached can look like an omission, and here it is deliberate. For the owner or CEO, a provider that says what it does not know is easier to hold to account.

05. Four requests that test any AI partner's claim

A working method, not a standard. If you are working out how to choose an AI consultant, or how to vet an AI partner already on your shortlist, these are four requests you can send to any provider, in an email, before a call. They are a light form of AI vendor due diligence, and they are ours. Each is paired with what public buying guidance and risk frameworks already ask for. Those are models, not obligations: the US federal memo binds US federal agencies, and the NIST and OECD papers are voluntary guidance or written for governments.

Request 1: show me a day of it. Ask: COO or head of operations. Ask what runs on working days, and ask to see one day's output: the artifacts, not a slide.

The US Office of Management and Budget's April 2025 memo on acquiring AI says that, where practicable, federal agencies should seek detailed demonstrations and tests "in scenarios that closely reflect the intended real-world operating environment", and that when evaluating proposals they must, to the greatest extent practicable, test proposed solutions to understand the capabilities and limitations of an offered AI system or service.¹³ It is written for US federal agencies, not for your company, and it is a good model for the request. What a good answer sounds like, in our view: specific artifacts, dates and a named person, not adjectives.

Request 2: who signs off, and what can they reject. Ask: owner or CEO. Ask who approves what goes out, and whether that person can say no. Accountability lands on the owner.

If the answer is always the one senior person, you have found the bottleneck again, only now it is inside the AI. Which tasks need a person at all is the subject of which tasks AI drafts, which it runs, and which always need a person, so it is not repeated here. Then ask the plain question: has a reviewer ever rejected something, and what happened next?

Request 3: what went wrong, and where is the correction logged. Ask: CFO or controller. Ask for one thing that went wrong and where the fix is written down. NIST's AI Risk Management Framework says, in its MANAGE 4.3 subcategory, that incidents and errors are communicated to relevant AI actors (the parties involved with the system), and that processes for tracking, responding to and recovering from them are followed and documented.¹⁴

In February 2025 the OECD published a common framework for reporting AI incidents, with 29 criteria.¹⁵ That is a policy paper aimed at policymakers, not a template for a consultancy, and NIST's framework is voluntary. Our view: a provider with no story about a mistake has either not run it daily or is not telling you.

Request 4: whose account, and whose data. Ask: CFO or controller, with the COO. This is where data and security worries sit. Ask whether what is built runs on your own account, with your data and your licences, and what happens to it if you leave.

NIST's GOVERN 6.1 subcategory asks for policies and procedures that address AI risks associated with third-party entities, meaning providers outside your company.¹⁴ The US memo says agency processes should ensure contracts permanently prohibit using nonpublic inputted agency data to further train publicly or commercially available AI. It also says agencies must include terms against vendor lock-in (being stuck with a provider because leaving costs too much), such as knowledge transfer and data and model portability. And it says agencies must determine whether to require disclosure of AI use in a solicitation.¹³ Access and permissions have their own article: why we connect AI read-only first.

06. Our answers to the four, and where they stop

A day. Meetings to tickets (section 03), the knowledge assistant on our own documentation, the weekly client status report, and proposals and design documents from templates. If you want to see what a day of it looks like, ask us.

Who signs off. AI drafts and does the work, and nothing goes out or goes live without someone's sign-off. On meetings to tickets, a person signs off anything that goes out.

What went wrong. The one correction we have already told is in section 04. We do not list further errors here, so ask us.

Whose account. For client work, it runs on your own AI account, with your data and your licences, and your data and account stay yours. That is how we build for clients. It is not a statement about our internal tools.

Where the answers stop. They stop where our evidence stops. What happens after you sign is the job of how a monthly AI support retainer works. The owner or CEO can bring the sign-off question to the call, the COO the day of work, and the CFO or controller the account and the data.

On NetSuite? Houseblend's NetSuite practice is separate; see its agentic delivery page.

Where this fits in the series. This is the eighth article in our AI adoption series. The first, our introduction to AI adoption, explains Plan, Build and Run, and the seventh, how a monthly AI support retainer works, covers what happens after launch.

Where to read next. How we run weekly client status reports with AI agents walks through a report that reads two company systems and is sent by a person.

This content may be AI-generated and may contain inaccuracies. It is not professional, legal, or financial advice. Verify independently before relying on it.