Use Case
Interchange AI GatewayMove AI out of SG&A.
Put it on the invoice.
If your business bills clients for costs incurred — legal, construction, manufacturing, engineering, agencies — AI is a line item, not overhead. Stop letting AI spend silently erode your margin and start treating it the same way you already treat every other direct cost on an engagement.
Who this is for
Any business that bills clients for incurred costs.
- Law Firms
- Construction
- Manufacturing
- Engineering
- Architecture
- Accounting
- Marketing Agencies
- Consulting
- IT Services
The reframe
AI flows through your P&L either way. The question is which line.
When AI lives in SG&A — buried under Software & Subscriptions alongside your email, your CRM, and your Zoom seats — it eats your margin invisibly. The bill arrives, accounting pays it, and nobody connects the spend to the engagements that generated it. Every dollar of AI usage is a dollar of profit you no longer keep.
When AI is attributed at the moment of use — to the matter, the project, the job, the client — it becomes a recoverable cost. The leak in SG&A goes away, EBIT improves by the amount you were quietly absorbing, and any standard pass-through markup on the billed-back cost becomes new gross margin on top. The recovery is not break-even. It is accretive.
The precedent is everywhere. Law firms have billed back Westlaw and LexisNexis usage to clients as soft costs for forty years. Construction GCs already pass direct costs through with markup on cost-plus and T&M contracts. Manufacturers absorb consumables into COGS per production run. Engineering firms have reimbursable-expense clauses in standard AIA and ACEC contracts. Agencies bill third-party tool costs and stock licensing through SOWs. The accounting infrastructure for treating AI this way has existed for decades — what was missing was the per-query attribution to make it work.
Illustrative example
Same AI spend. Two very different outcomes for EBIT.
A small professional-services firm consumes ~$8,400/year in AI tokens across active client engagements. In one scenario it absorbs the cost in overhead. In the other it bills the cost back as a reimbursable direct cost with a standard pass-through markup — and improves operating income by roughly $10,000.
Before
Margin erosionAI on the firm credit card. Lumped into the monthly software run-rate. No attribution to any matter or client.
AI is invisible inside firm overhead. Every dollar of AI usage on a client's matter is a dollar of EBIT the firm absorbs.
After
Margin contributionAI attributed to the matter at the moment of use. Billed back to clients as a reimbursable direct cost with a standard 20% pass-through markup.
AI moves from an overhead drag to a margin contributor. EBIT swings by ~$10,080 versus the prior treatment — $8,400 because the cost no longer falls on the firm, plus $1,680 of new gross margin on the standard pass-through markup.
Illustrative example. Markup percentages and recovery rates vary by industry, contract structure, and client agreement. A 20% pass-through markup is within the standard range for direct-cost reimbursables in construction, agencies, manufacturing, and many professional-services arrangements; law firms typically bill AI at actual cost in line with ethics guidance.
What it takes
Four things have to be true before AI becomes a recoverable cost.
Per-user attribution
Every query is tied to the employee who ran it. Without this, there is no defensible chain of custody for client billing.
Per-project / per-matter / per-job tagging
Each request is associated with a client, matter, project, or job code at the moment it happens — not reconstructed later from memory.
Audit-ready logs
Token counts, model used, timestamp, and cost are recorded for every interaction. Clients and auditors can verify the math.
Itemized line items
Spend exports cleanly into your invoicing or accounting system — one line per matter, project, or engagement.
The bridge
This is exactly what Interchange does.
The four requirements above aren't theoretical — they map directly onto how the Interchange AI Gateway is built. Every employee authenticates through an internal subdomain, every request carries identity and project metadata, every interaction is logged locally with full token and cost detail, and analytics export aligns with how your billing system already structures data.
-
Per-user attribution
Employees authenticate through an internal subdomain. Every request carries their identity.
-
Per-project tagging
Request metadata captures the client / matter / job code at submission time.
-
Audit-ready logs
Every interaction is logged locally — model, tokens, cost, timestamp, user, project.
-
Itemized line items
Analytics export aligns with the columns your billing system already uses.
Related reading
How this plays out in specific industries.
AI cost recovery for small law firms
The modern Westlaw model. ABA Model Rule 1.5, soft-cost precedent, and the numbers for a four-attorney practice.
AI as a project cost code
Cost-plus and T&M contracts already pass direct costs through. AI joins the existing cost-code structure.
AI as cost of goods sold
For custom manufacturers, AI absorbs into COGS per job — not SG&A. A cleaner accounting story than services.
Project-billed AI for engineering firms
AIA and ACEC reimbursable-expense clauses already cover this. AI joins printing, plotting, and specialty software.
AI as a line item for agencies
The leak is biggest where AI use is heaviest. Agencies subsidizing client AI use are giving away margin every month.
FAQ
Common questions.
Can I really bill clients for AI usage on engagements?
Yes, when your engagement letter or contract authorizes it. Professional-services firms have billed back research databases, printing, and specialty software for decades. AI fits the same pattern — what matters is clear disclosure, attribution to the matter, and billing at actual cost.How is this different from a service markup?
A service markup raises the rate on labor you were already going to bill. Cost recovery is separate — it itemizes the actual AI tokens consumed on a specific matter and passes them through to the client, typically with a 15–25% pass-through markup on direct costs (the same kind of markup that already applies to subcontractor invoices, stock licensing, and other reimbursables). Many firms do both: bill labor at their professional rate, and bill AI as a recoverable direct cost on top.What about flat-fee or fixed-bid work?
Cost recovery still works, but the mechanics differ. Fixed-fee firms typically budget AI into the bid (the cost stays internal but is no longer invisible — you know it exists when pricing the engagement). Some firms quote a base fee plus pass-through AI costs as a separate clause.Is there ethics guidance for law firms specifically?
ABA Model Rule 1.5 has long permitted reasonable cost recovery with proper client disclosure, and several state bars have published AI-specific opinions in the past two years. The common ground: bill AI at actual cost (not marked up as legal services), disclose the practice in the engagement letter, and itemize on the invoice. Check your state's most recent opinion.Do clients need to consent or be notified?
Notified, yes — typically through the engagement letter, SOW, or master services agreement. Active consent depends on jurisdiction and industry. The defensible practice is plain-English disclosure up front, itemized line items on invoices, and a clear request rate that matches the underlying token cost.