Which workflows absorb well, and which don't.
A field guide to telling the difference between work that can be done by an AI agent today and work that still needs a person. Most of the noise about "AI for business" ignores this distinction. The economics, and the disappointment, hinge on it.
The shift that's actually happened
For most of the last decade, "use AI" meant use a tool that helps a person do the work faster. The person still did the work. The tool gave them a head start on a draft, surfaced an answer, summarised a document. Useful, but the time savings never quite showed up in the P&L, because the human bottleneck didn't move.
Something different started happening in 2024 and accelerated through 2025. In narrow, well-defined workflows, agents began doing the work themselves: filing routine returns, reconciling accounts, drafting and sending standard correspondence, answering the same five customer questions, onboarding the next tenant. A human still reviews and signs off where it matters, but the production work, the bit that used to eat half your team's week, gets done by software that knows what good looks like.
That shift makes the old framing wrong. The right question isn't "where could AI help my team?" It's "which work in my business could come off my team's desk entirely?" Those are very different questions, and they point to very different workflows.
Five signals, one rule of thumb
We've scored hundreds of workflows now, across accounting, legal, property management, B2B services, and customer success. The workflows that absorb well share a profile. The ones that don't share a different profile. Here's the shape.
1. Repeatability — same shape, different details
The work has a fixed structure. The inputs vary, but the steps don't. If you could write the checklist on a single page and a new junior could follow it, this signal is high. Quarterly VAT filings score high. "Advise the client on a tricky restructuring" scores low.
2. Already outsourced — or already buried in admin
If the work has already been handed to junior staff, a BPO, or a contractor, nobody inside the business treats it as the thing that makes them special. That's a sign it can move further out. The work nobody fights to keep is the work an agent can do.
3. Clear right answer — a tight band of acceptable outputs
A return is correct or it isn't. An invoice matches the PO or it doesn't. The narrower the band of acceptable answers, the more confidently an agent can do the work. Workflows where "good" is contextual, "depends what the client meant," score poorly here.
4. Outcome-measured — paid for the result, not the hours
If the client cares that the filing is on time and correct, not how long it took, you can change how the work gets done without changing what they pay for. That's where the economics open up. Hourly-billed workflows are harder to absorb cleanly: the business model resists the productivity gain.
5. Margin — expensive enough that an order-of-magnitude reset matters
Meaningful staff time, meaningful spend, meaningful turnaround pressure. If the work is a small cost in the corner of the business, the juice isn't worth the squeeze. We only take engagements where the maths is obvious.
Rule of thumb: a workflow worth absorbing scores 4 or 5 on at least four of these five. Three or fewer means it's not ready yet, or it's the wrong shape.
The workflows we keep seeing
Names changed, details blurred, but the shapes recur. These are the kinds of workflows that consistently score 4+ on at least four signals.
- Quarterly VAT and payroll filings for SME clients. Same template every quarter, narrow band of correct answers, already handled by juniors, fixed-fee priced. An agent prepares, a qualified accountant signs off.
- Tenant onboarding for property managers. Six hours of back-and- forth per tenant, mostly chasing documents, sending the same forms, scheduling key handovers. The work is repetitive and outcome-measured ("tenant moved in on time with deposit lodged").
- Tier-one customer success replies. The same five questions answered the same five ways. Margin is high because the cost is the team's attention, which is finite and expensive.
- Document chasing in professional services. "We need your bank statements for Q3" sent seventeen times in slightly different ways. Pure admin, pure repetition, zero strategic content.
- Routine compliance filings in regulated industries. The regulator's template doesn't change. The submission window doesn't change. The qualified human signs off; the agent does the preparation work the qualified human shouldn't be doing.
The workflows that don't absorb well
Just as important: the workflows that look promising and aren't. We turn down engagements where the work falls into one of these shapes.
- Judgment-heavy strategic work. "Advise the client on the right structure for this acquisition." There's no tight band of correct answers, the stakes punish a wrong call, and the value the client buys is the judgment. Layer 3 work doesn't get absorbed — it gets surrounded by agents that handle the execution around it.
- Workflows that exist because of a broken process. Sometimes the "repetitive work eating your team" is repetitive because nobody fixed the form upstream, or the integration that should exist doesn't. The right answer there is to fix the process, not to deploy an agent to absorb the friction.
- Low-margin, low-volume work in a corner of the business. If the workflow takes one person two hours a month, building a production agent to absorb it loses you money even if the agent is perfect. The juice has to be worth the squeeze.
- Workflows where the customer is paying for the human. Some services are explicitly "you're buying my time and attention." White-glove concierge, executive coaching, certain legal work. The customer doesn't want the work done faster by software; they want the relationship. Absorbing it kills the product.
How to use this in your own business
Look at where your team's week actually goes, not where you think it goes. The workflows worth absorbing are usually the ones nobody talks about in strategy meetings, the ones that are too boring to be on the org chart but big enough to be a real chunk of payroll.
Score each one against the five signals. If three or more score 4+, you have a candidate. The 5-minute assessment does this scoring against your business, asks four follow-up questions, and ranks the three most absorbable workflows you have. If you'd rather have the longer version, the Diagnostic Sprint goes deeper.