Skip to content
upshift

The AI Exposure Index

The Index is a census, one edition per sector, each dated and sourced. Every edition asks one question of every line of billable work: can your client get the finished deliverable without you? It names the companies selling that work today, then rules the line replaced, repriced, or defensible.

What the three verdicts mean

Replaced
Your client can buy the finished deliverable without you. The fee is attached to work they can now produce, so the offering gets rebuilt rather than repriced.
Repriced
Your client still needs somebody to run the AI and check its output. You keep the engagement and bill fewer hours for it, which makes it a rate conversation.
Defensible
Nothing on the market does the billed work. The hours underneath still compress, so whether the saving stays with the firm depends on what the engagement letter quotes.

Reading across the editions

The editions disagree most on the repriced band: it covers 8 of marketing’s 16 service lines and 1 of accounting’s 17. They agree on the defensible band, where one feature runs through every edition’s short list.

How the census is built

Every company in an edition sells a product a buyer can pay for today, and earns its place only with a sourced account of what it sells. We place a company on a line when it targets the work that line bills for, not a task beside it. We drop anything we can’t tie to a primary source. In the accounting edition we reopened every cited source to confirm it names the company and backs the figure, and that took out six entries.

Attaching a company to a line is a judgment call, so each edition publishes it with a source on every entry and you can check the call yourself. Treat the counts as a floor: a sweep only finds the companies that publish. Each edition reads shipped product as of its review date, so it names who is coming for the work rather than billing that has already moved.

Every edition publishes its full census as JSON and CSV on its own page, free under a Creative Commons BY-ND 4.0 license. Cite the edition and its review date.

Questions

The AI Exposure Index is original research from Upshift: a dated, sourced census of the funded AI companies and AI-native firms selling into professional services work. It runs one edition per sector, and every edition names each line of billable work and gives it a verdict of replaced, repriced, or defensible. Every company named carries a source link, so a reader can check the call against the company.

Each edition asks one question of each line of billable work: can the client get the finished deliverable without the firm? A line is replaced when the client can. It’s repriced when the client still needs somebody to run the AI and check its output, so the firm keeps the work and bills fewer hours. It’s defensible when nothing on the market does the billed work at all. The verdict sits on the line, never on a company.

The lines that came out defensible in each of the four editions share one feature: a named person answers for the result afterward, and in the licensed lines signs for it. In law, the defensible lines are white-collar defense and internal investigations; public-company M&A; public-company reporting and securities compliance; and trial practice. In marketing, the defensible line is brand strategy and identity. In accounting, the defensible lines are financial statement audit; forensic accounting and litigation support; employee benefit plan audit; and reviews and compilations. In consulting, the defensible lines are board and C-suite strategic advisory; and change management facilitation.

Yes. Every edition publishes a JSON and a CSV file of its full census under a Creative Commons BY-ND 4.0 license, linked from that edition’s own page beside the method that produced it. Cite the edition and its review date, since each edition is re-reviewed rather than rebuilt.

Each edition censuses a different market. Law’s companies sell into legal practice areas and marketing’s sell into agency service lines, and those two fields are not one population. Adding them would produce a combined figure that no census supports. A managing partner reads the edition for their own profession and compares lines inside it.