Alex Wittenberg writes about finance operations, financial modeling, and practical uses of AI in finance at airCFO, which manages back-office operations for tech startups.
Alex Wittenberg is useful for finance AI content that explains context, process design, and model trust
Best for
A fintech selling AI into CFO processes, content collaboration or advisory on practical adoption
Not for
Security vendors targeting software engineers seeking paid sponsorships, rates, conversions, or short-form video
Key takeaways
Connecting Claude to accounting and banking surfaced numbers but not the business story
People should define finance processes and decide where AI belongs, not let agents figure it out
Financial models need trust, traceability to assumptions, and access across a business
Check first: Ask for audience survey, newsletter subscriber breakdown, and recent post-level results before committing budget
Main ideas
AI is only as useful as the context you give it · Define the process first, then put AI inside it · Spreadsheets fail on trust, traceability and usability — and that is an architecture problem
Alex Wittenberg writes about finance operations, financial modeling, and practical AI use at airCFO, which manages back-office operations for tech startups. Across these posts, they argue that AI is more useful in finance when it has business context and supports workflows people have designed, rather than making decisions on its own.
Wittenberg connects technical choices to the problems finance teams face.
In a post about building a finance “team brain,” connecting Claude to accounting and banking systems surfaced the numbers but not the business story behind them.
On AI autonomy, they argue that people should define the process and decide where AI belongs, rather than telling it to “figure it out” in core finance operations (why finance processes need human direction).
A third post applies the same practical lens to financial models, focusing on tests for trust, traceability back to assumptions, and access for people across a business (the case for hybrid financial modeling).
Readers looking to put AI and finance tools to work in day-to-day operations will find concrete design questions to consider.
Alex Wittenberg writes about putting AI to work in finance at airCFO, which runs back-office operations for tech startups.
Read more
The posts argue that AI’s value in finance comes from the context and process a human team provides, not from the model’s autonomy: give a model the mechanics of a business and it can play back the numbers; give it the narrative behind them and it can identify which numbers matter.
Wittenberg applies that argument to financial modeling, saying spreadsheets have structural limits — no real test suite, hard-to-trace numbers and all-or-nothing permissions.
The proposed fix is a model engine with validation and a UI layer on top.
AI is only as useful as the context you give it
Alex Wittenberg argues that disappointing AI output on financials is usually a context problem, not a model problem.
Read more
A model can repeat the numbers it receives, but without the story behind them, it cannot judge which figures matter. The fix is to build a deliberate context layer: connect the systems of record and have the model write up what it can establish, then add the financial model and interview the operator to capture what the systems never held.
To build a “finance team brain,” Alex Wittenberg connected Claude to airCFO’s QuickBooks Online, Ramp and Mercury accounts and had it draft everything it could establish with certainty.
The result was a thorough mechanical picture — a self-graded B+ — but it missed the semantic layer.
Adding the financial model and having Claude interview them brought the work to what they call an A-level understanding, in roughly an hour.
A colleague said the finished profile read like something a competent auditor might have produced.
Alex Wittenberg distinguishes AI agents from AI used within a human-defined process.
Read more
An agent, in this view, means handing over the objective and telling the model to figure out how to achieve it — a job Wittenberg argues belongs to a human. airCFO has detailed playbooks from more than a decade of helping startups fundraise and scale, so the team sets the process and uses AI in specific steps rather than replacing it.
airCFO manages back-office operations for 300-plus tech startups, and Alex Wittenberg writes that it will be a long time, if ever, before the firm relies on AI agents for core workflows.
The post recounts a conversation with Daniel Kang, who works on deploying AI at Mercury.
As Wittenberg paraphrases it, Kang’s view is that deciding where AI can and should not be used in finance is the CFO’s job.
Wittenberg asks readers whether they have useful policies for drawing that line.
Alex Wittenberg frames this as a position on airCFO’s own operations, not a claim that agentic automation cannot work elsewhere.
Spreadsheets fail on trust, traceability and usability — and that is an architecture problem
Alex Wittenberg identifies three problems that have dogged financial modeling for decades.
Read more
Trust: spreadsheets make it hard to build a full test suite, so checks remain ad hoc.
Traceability: as a workbook grows, it gets harder to trace any number back to its source.
Usability: spreadsheet permissions are all-or-nothing, so giving the wider team access risks breakage.
The proposed solution is a hybrid platform: a model-based calculation engine that validates results against hundreds of tests at build time and on every refresh, paired with a UI that lets non-finance colleagues work safely with the model.
In this approach, finance stops being a bottleneck and enables faster, more aligned decisions.
Alex Wittenberg points to a clear payoff: when an investor questions the assumptions behind growth projections, a traceable model lets the CFO answer in seconds. Executives can make updates without risking breakage, while team leads can see their slice of the business.
Publish from inside the work, and send readers deeper
A recurring priority across the posts is a two-tier publishing habit: a short LinkedIn post states the problem and outcome in practitioner terms, then points readers to the AI CFO newsletter for the step-by-step build or full blueprint.
Read more
The posts draw on work in progress at airCFO rather than commentary on other people’s news. Readers will find the details they need to copy the approach in the newsletter.
The finance team brain post promises a step-by-step account of building a finance context layer in the following week’s AI CFO; the modeling post says the next AI CFO essay will recap the path to the architecture and share a detailed spec.
Answers how to make AI useful in finance beyond simple number playback.
Read more
Start here for a concrete, step-by-step example of improving AI outputs with business context.
Alex Wittenberg describes building a 'finance team brain' by connecting Claude to airCFO's QuickBooks Online, Ramp, and Mercury, then adding the company's financial model and a self-interview to supply narrative context.
Without that context, Claude can play back numbers but cannot tell which ones matter.
The final profile read like a competent auditor report, and the build took about an hour.
For brands planning a campaign
For a fintech company promoting AI in CFO tools, this post is a ready-made example of how to apply AI in finance. It shows how adoption can move beyond basic analysis and could anchor a case study, webinar, or advisory content about context-aware AI.
Answers where AI fits in finance operations and where it does not.
Read more
Start here for a clear argument against full automation and a human-defined approach.
Alex Wittenberg writes that airCFO manages back-office operations for over 300 tech startups but will not rely on AI agents for core workflows.
The post distinguishes AI agents—which imply autonomy—from AI injected into workflows defined by humans.
They quote Daniel Kang of Mercury, who says it is the CFO's job to determine where AI should and should not be used.
For brands planning a campaign
For a fintech company promoting AI adoption, this post highlights the importance of human oversight and policy. It could support content on responsible AI implementation or advisory services that help CFOs set AI boundaries.
Answers what a modern financial modeling platform improves over spreadsheets.
Read more
Start here for a clear product vision and the problems it solves.
Alex Wittenberg outlines three long-standing problems with spreadsheet-based financial modeling: trust (hard to build test suites), traceability (hard to trace a number back to its source), and usability (spreadsheets have all-or-nothing permissions). The post describes airCFO's new hybrid platform, which validates results against hundreds of tests, traces numbers to sources, and adds a UI so anyone in the business can access the model without risking breakages.
For brands planning a campaign
This post details a product architecture that fintech companies could reference in content about next-generation FP&A tools. It offers a concrete example of how to frame AI-enabled modeling for CFOs and finance teams.
Turning AI into a usable finance analyst, not a number playback tool1
Where AI should and should not take over finance workflows1
Rebuilding the financial model itself: trust, traceability and usability1
Turning AI into a usable finance analyst, not a number playback tool
Wittenberg writes about why generic AI models give weak finance answers: they can recite numbers but miss the narrative that makes them matter.
Read more
Their fix is a structured context layer: connect Claude to airCFO's QuickBooks Online, Ramp and Mercury accounts, let it draft what it can establish from the systems alone (which Wittenberg says self-graded as a B+ on the mechanical figures), then add the financial model and a model-led interview to close the gap to what they describe as an A-level understanding. The build reportedly took roughly an hour end to end and was shown internally, where a colleague compared the output to a competent auditor's report.
Connect tools first, extract what the systems can prove, then add the model and the human narrative the systems cannot capture.
For brands planning a campaign
For a fintech selling AI into CFO workflows, this is the most directly relevant topic: Wittenberg is publishing a build-along account of context engineering, not a vendor pitch. That suits co-created explainer content, a walkthrough or a sponsored teardown of what connectors deliver versus what still requires human input.
Where AI should and should not take over finance workflows
Wittenberg draws a line between autonomous AI agents and AI inside human-defined processes.
Read more
Their argument: an agent implies handing over the objective and letting the model figure out how to reach it, but airCFO already has playbooks from more than a decade of helping startups fundraise and scale, and those define the workflow. In the same post, they relay a conversation with Daniel Kang about AI deployment at Mercury, framing the CFO's job as deciding where AI belongs in finance and where it does not, and asking readers whether anyone has useful policy for that split.
Keep process ownership with the human team and use AI in steps the team has already specified.
Wittenberg frames this as a position on airCFO's own operations, not a claim that agentic automation is impossible elsewhere.
For brands planning a campaign
A fintech positioning AI CFO tools can use Wittenberg for governance and policy content. The hard question of scoping AI in finance is where buyers stall, and this topic gives a collaborator a credible, opinionated voice rather than a feature tour.
Rebuilding the financial model itself: trust, traceability and usability
A third topic is the architecture of financial modeling.
Read more
Wittenberg names three long-standing problems: spreadsheets make it hard to run a real test suite, trace a single number back to its source as a workbook grows, and manage permissions beyond all-or-nothing access, which blocks the wider team from using the model. Their answer is a hybrid platform where a model engine validates its results against hundreds of tests at build and on every refresh, while a UI layer lets executives and team leads work with the model without risking breakage — moving finance from bottleneck to enabler.
Explain the architecture through the operational failure it removes, with the investor-question example as the payoff.
For brands planning a campaign
This is the strongest hook for a product-adjacent collaboration: a fintech can sponsor or co-produce a technical blueprint on model architecture, validation and access control for FP&A and CFO readers who already feel these three problems.
Across the posts, the recurring format is a LinkedIn post that opens a deeper build or blueprint and points readers to the AI CFO newsletter for the full account — the finance team brain write-up and the recap of the path to the modeling architecture.
Read more
The posts read as practitioner notes from inside the work, not commentary on other people's news.
A short public post states the problem and the outcome; the newsletter carries the step-by-step detail.
For brands planning a campaign
The newsletter is the substantive channel: a placement or co-authored essay reaches readers who opted in for implementation detail, while a LinkedIn post can bring them to that content.
Alex Wittenberg's posts are written from inside airCFO, a firm that runs back-office operations for tech startups.
Read more
They show one employer relationship and one named brand mention.
airCFO
Employment
Wittenberg posts as an airCFO insider, referring to "our President Justin" and saying "we manage back-office operations for 300+ tech startups."
Read more
Their posts cover airCFO's marketing-built finance "team brain," the QuickBooks Online, Ramp, and Mercury systems connected to it, airCFO's new hybrid financial modeling platform, and the company's AI CFO newsletter. The posts show a working affiliation but do not state a job title; ask Wittenberg to clarify their role.
Wittenberg names Mercury as one of the systems airCFO connected to Claude to build its finance context profile, and relays a conversation with Daniel Kang, described as deploying AI at Mercury, about where finance teams should and should not use AI.
Read more
Ask whether Wittenberg has a commercial relationship with Mercury.
Alex Wittenberg's posts tackle one question: how to make AI useful inside a finance function, beyond the demo.
Read more
The clearest entry point is a pair of build-alongs: one assembles a “finance team brain” by connecting a model to the company's accounting, spend and banking systems, then adding the narrative those systems never captured; the other explains the architecture behind a hybrid financial modeling platform.
A third post draws a boundary: agents get objectives, people own the process.
Together, the posts argue that context and process design—not model choice—determine whether AI improves finance work.
Each post points to the AI CFO newsletter for step-by-step builds and full specifications.
Both write from inside operational B2B service businesses and treat tooling as something that must keep pace with growth.
Read more
Zafar covers recruitment agencies building international teams and taking their stacks global; Wittenberg writes about finance teams that have outgrown spreadsheets and AI with too little context.
Zafar covers recruitment agency operations, global hiring and the technology behind them; Wittenberg focuses on AI in finance, context layers and modeling architecture. They publish longer technical build-alongs, while Zafar shares short operational observations and questions for readers.
Both explore the build-versus-buy question for growing businesses and turn a platform perspective into practical guidance.
Read more
Cliff writes about vibe-coded CRMs and why “Salesforce is the CRM you graduate to” is the wrong frame; Wittenberg explains why a spreadsheet cannot simply be retrofitted with the trust, traceability and permissions a real financial model needs.
Cliff covers CRM and small-business software adoption, often making the case for a named platform. Wittenberg writes about finance workflows and AI from an outsourced finance-services firm, not a software vendor.
Both are finance practitioners who write in detail about how AI is used inside a finance function.
Read more
Hu covers pointing an agent at every contract to build a revenue recognition layer, and what a new CFO should not outsource to a chatbot; Wittenberg writes about building a finance context layer and distinguishing autonomous agents from AI placed inside human-defined workflows.
Hu writes from operating finance roles inside startups, with a strong focus on accounts receivable, billing and revenue recognition. Wittenberg writes from an outsourced finance-services firm serving hundreds of startups, focusing more on context engineering and model architecture.
For a campaign aimed at CFOs considering AI adoption in CFO tools, the posts point to a narrow, senior finance readership rather than a broad one.
Read more
The evidence is thin, so treat the placement as a bet on topic alignment, not proven reach.
The only published audience number is Wittenberg’s LinkedIn following: 4,529 accounts, observed October 8, 2026.
The posts do not establish where those accounts are based or what jobs they hold or how senior they are.
The published work indicates who it is written for.
Three posts address people who run finance at venture-backed companies.
One describes building a finance context layer for airCFO in Claude and reporting that the output read to the company president like a competent auditor’s report; one lays out a hybrid financial modeling platform around three problems — trust, traceability, usability — explicitly for FP&A professionals and CFOs; one argues that back-office operations should stay human-defined while AI is injected into specific, bounded steps, and quotes Mercury’s AI lead on where finance should and should not use AI.
The vocabulary is practitioner vocabulary: FP&A, model refresh, permissions, QBO, Ramp, Mercury, growth projections for investors.
That signals an intended readership, not a measured one.
Before committing budget, ask Wittenberg for any audience survey, newsletter subscriber breakdown, or platform audience data describing followers’ job titles, company size, or seniority, as well as recent post-level results — impressions, comments, reshares — for posts on the topics and in the format you would sponsor.
Follower count is the only audience data available, and it does not show whether people who choose and buy CFO tooling are in that audience.
LinkedIn followers
4,529
Public follower count on the LinkedIn profile linked from this record · 2026-10-08
Alex Wittenberg's posts do not identify a sponsorship program, rate card, media kit, or partnership contact.
Read more
What they promote is airCFO's own output: the AI CFO newsletter, which carries technical essays and platform blueprints, and the company's hybrid financial modeling platform.
Subscribe links are pointed to the newsletter through post comments rather than through any advertised collaboration route.
The posts do not state terms, deliverables, pricing, review process, or an inquiry address, so a buyer should treat a paid placement here as unconfirmed rather than assume a documented offer.
The practical starting point is the same channel the posts use to distribute the newsletter and platform material — direct contact through the LinkedIn account and the company — not an open sponsorship portal.
If a collaboration is pursued, expect the terms to be negotiated from scratch with no published format to plan against.
A fintech selling AI into CFO processes wants a content collaboration or advisory partner to move finance teams past basic AI use.
Alex Wittenberg is a specific fit for this need. All three posts address the problem the fintech wants to solve. In one build-along, Wittenberg connected Claude to airCFO's QuickBooks Online, Ramp and Mercury accounts, asked it to report what it could establish from the systems alone, then added the financial model and a self-interview to supply the narrative the systems could not capture. The mechanical answer graded out as a B+ until that extra context was added. For a vendor arguing that AI in finance fails without the right context, this demonstrates the point rather than merely describing it. In a second post, Wittenberg takes the opposite position to much of AI marketing: an agent takes an objective and lets the model decide how to achieve it. airCFO will not use that approach for core processes, Wittenberg says, because a decade of playbooks already defines them. The third post names three failures of spreadsheet modeling — trust, traceability and usability — and describes the architecture built to address each, with an investor-question example as the payoff.
The subject is the buyer's category, explained in practitioner language (FP&A, model refresh, all-or-nothing permissions). The format is a short public post that leads to a longer technical essay in the AI CFO newsletter. A co-produced piece could therefore meet readers where they already go for implementation detail, rather than serve as a one-off placement. The tradeoff: Wittenberg publishes as a named individual inside the company channel, not through an independent media property. Frame any collaboration as co-produced work with airCFO, not as a neutral review.
Before budgeting, ask for audience details and partnership terms. The LinkedIn account has 4,529 followers, but the posts provide no demographic information, so reach among people who choose or buy CFO tools is unproven. The case rests on topic alignment and the newsletter, not measured reach. The posts also mention no sponsorship program, rate card or inquiry route, so terms would need to be negotiated from scratch.
When to choose a specialist
Read more
A security vendor targeting software engineers wants a paid sponsorship series with published rates, tracked conversions and short-form video.
The posts do not support this need. They mention no rate card, media kit or partnership contact, and none announces a sponsorship; they promote airCFO's own newsletter and platform. A buyer planning against published formats would be starting from zero, with no indication that a paid placement series of this kind is available.
The format is also a mismatch. The posts are written text with links to a newsletter; they include no video or per-post performance data. There is no basis for pricing a short-form video series or projecting conversions. The readership is wrong for an engineering campaign, too: the posts address finance professionals who run the numbers — FP&A professionals, CFOs and executives who need to use a model without breaking it — not engineers evaluating a developer or security tool.
A different offer aimed at the finance audience could work, such as co-produced explainer content on where AI belongs in a finance process and where it does not. As written, the request calls for a commercial product the posts do not describe and an audience Wittenberg does not publish to.
A worked campaign example
Campaign angle for a fintech selling AI into CFO processes: a co-produced build-along showing what an AI connector delivers — and where a human still has to step in.
The angle comes from the strongest example in Wittenberg's posts: connecting Claude to the company's accounting, spend and banking systems, letting it draft what the systems could establish, then finding the output incomplete until the financial model and a human interview were added. That gap turns a vague claim about context into a scoped, checkable exercise for buyers of AI tools for CFO processes.
Shape of the collaboration:
1. A LinkedIn post from Wittenberg's account stating the problem and outcome in practitioner terms: the systems produced a solid mechanical picture, a human supplied the narrative, and here is what the difference looked like. This is the entry point and matches the format already in use. 2. A longer essay in the AI CFO newsletter covering the steps: which systems were connected, what the model could establish alone, which questions only the human interview answered, and what was still missing. The vendor's product can feature as the connector or modeling layer, described through the job it does rather than as a feature list. 3. A short advisory companion for finance leads: a written position on where AI should and should not take over, based on Wittenberg's distinction between autonomy and AI injected into human-defined steps. The Mercury conversation in the posts signals interest in this topic; a policy checklist finance teams can adopt would make it practical.
This approach builds on a demonstration Wittenberg has already run rather than a scripted testimonial. It gives finance readers something to take back to their teams and puts the vendor in the part of the process the audience is trying to solve. The next step is to approach Wittenberg through the LinkedIn account and the company, then scope deliverables, ownership of the co-produced essay, disclosure of the relationship and the measures the vendor wants reported. No performance baseline exists, so agree on how success will be measured. Negotiate the arrangement from scratch.
Choose Hu over Wittenberg when a campaign aimed at CFOs needs a finance-side voice speaking from the operator seat about where AI belongs in a CFO's first 90 days.
Read more
Hu writes directly from that seat: an ERP on QuickBooks, a model living in a Google Sheet, no monthly close — then pointing an agent at every contract in company history to parse into JSON and building a revenue recognition layer that runs its own process alongside the accounting team's.
Hu also names what not to do, warning against feeding AI the same questions and passing slop documents between teams, and drawing the line between moving faster and becoming unable to function without AI.
The Ramp post adds a second useful dimension: writing credibly about a fintech platform entering the AR space from the perspective of someone who helped build the early requirements.
If the collaboration is tied to a product launch into invoicing, billing or AR, Hu has firsthand material for it in a way their posts do not show.
Hu publishes as CFO of Stuut, an accounts receivable software company, so a paid collaboration puts the content on vendor territory, with a direct interest in the same market.
Scope the competing interest and disclosure before agreeing on deliverables.
There is no newsletter, content program or publishing channel behind these posts, so a co-produced asset would be built from scratch rather than placed before an existing audience.
The posts are arguments and build notes, not teaching walkthroughs; a step-by-step resource would need more editorial shaping than their build-along format.
Audience size and make-up are undocumented, so ask for those figures before making a reach case beyond topic and role.
Cliff is the better choice if the campaign's audience widens from CFO tooling to the small-business software buyer deciding whether to build or buy with AI.
Read more
Cliff makes the structural argument a campaign about AI and software buying needs — AI has made building easier, and buyer expectations have shifted — but lands it on CRM: the decades-old build-versus-buy debate, why a fast-growing business may be better off growing than vibe-coding something that already exists, and what a company takes on when it maintains, secures and scales its own system.
The second post reverses the usual upgrade logic, arguing that small businesses should not start on a cheaper tool and migrate later because customer data and processes get scattered in between.
For a campaign aimed at SMB decision-makers evaluating AI features, that is the same decision framed for a different buyer, and Cliff shows how to position a platform as the alternative to building.
Cliff's posts close on Salesforce Starter Suite and a signup link, so co-produced content reads as the vendor's own case from the outset.
The product Clifford is advocating sits in CRM, not finance.
The examples cover CRM adoption, migration and scattered customer data, so a brief about AI in CFO workflows would need finance-specific evidence from elsewhere; Clifford cannot supply close, FP&A or modeling detail.
Compared with Wittenberg, there is no equivalent finance-practitioner demonstration to build a campaign angle on — no build-along, no named failures of spreadsheet modeling — leaving less material for a technical co-production.
Zafar runs a company selling into the operational problem it solves and writes about it in concrete terms.
Read more
The strongest example is the headcount argument: going from 10 to 30 people can be harder than 30 to 50, because at 10 the founder knows every client, candidate and task, and by 20 to 30 communication breaks down, processes go inconsistent, CRM data turns messy and the founder becomes the bottleneck.
Zafar's conclusion is that the fix is not always another manager but better process, and that agencies which scale are the ones where good people can perform consistently.
The second post extends this to international teams — recruiters based anywhere, and a tech stack that has to work across borders.
If the campaign includes a second audience of recruitment or staffing agency owners, or wants a partner who can argue that growth problems are process problems before they are software problems, that framing transfers and gives a campaign angle to build on.
For a campaign aimed at finance teams and CFOs evaluating fintech and AI adoption, Zafar reaches the wrong room.
The audience is recruitment agency owners, and the examples are agency scaling, cross-border hiring and recruitment technology; none addresses finance teams, FP&A professionals, CFOs or AI adoption in finance.
A collaboration would not put the fintech in front of the people who choose CFO tooling.
Reusing the operational-problem framing would mean rebuilding it around a finance workflow, at which point the creator's own examples stop doing the work.
The posts are short and each closes on the product name.
Ask for source material before committing to a co-produced technical piece, since the posts do not show how much depth is available.
Alex Wittenberg's own published posts cover three topics: a build-along on connecting Claude to airCFO's accounting, spend and banking systems to create a finance context layer, a position on defining finance workflows before injecting AI into them, and an account of a hybrid financial modeling platform and the trust, traceability and usability problems it addresses.
Each post points readers to the AI CFO newsletter for the step-by-step detail, which is not included in the posts here.
Similar-creator comparisons elsewhere on the page are based on the visible text of posts by three other finance and software writers.
Claims about audience size rest on a public follower count only; no sponsorship formats, rate card, partnership contact or per-post performance data appeared in the posts here.
