Skip to main content
← All posts
By Abhay Muttreja, Founder, SolvTreeAI

Choosing the next automation after the first one works

After a first automation works, rank the next one on volume, exceptions, measurable value, dependencies and ownership, not novelty.

·4 min read
  • Automation Strategy
  • Workflow Prioritisation
  • Automation ROI

The short answer

Pick the next workflow by evidence, not excitement. Compare candidates on how often the work happens, how many cases are exceptions, whether the value can be measured, what the workflow depends on, and who will own it. The more impressive candidate is often harder to validate. A smaller, well-measured next step usually teaches more and pays back sooner.

Key points

  • Success with the first automation creates pressure to do something bigger. Resist choosing on novelty.
  • Volume and exception rate matter more than how clever the solution sounds.
  • If you cannot measure the baseline, you cannot show the improvement.
  • Dependencies on other systems, people or data quality add risk and time.
  • Every workflow needs an owner who will use it, review it and report problems.

Why the second choice is harder than the first

The first automation usually picks itself: an obvious, painful, repetitive task. Once it works, the list of ideas grows. Someone suggests an AI tool that does something visible and new. Someone else suggests a dull reconciliation that eats hours each week. The visible one tends to win the meeting.

Choosing well means scoring both against the same questions and being honest about what is uncertain.

Five factors for ranking the next workflow

  1. Volume: how many times a week does this happen? Rare work, however annoying, gives little room to save time and few cases to test on.

  2. Exception rate: what share of cases do not follow the normal path? High exceptions mean more review and slower gains.

  3. Measurable value: can you measure time, errors, cash or speed today, and again afterwards? Include the full running cost: subscriptions, review time and maintenance.

  4. Dependencies: what does it need: clean data, access to another system, a supplier’s cooperation, a decision from someone else?

  5. Ownership: who will use it daily, check its output and report when it breaks?

Score each from 1 (poor) to 3 (strong), and write a one-line reason. The reasons matter more than the totals.

An illustrative example: two candidates at a travel company

An illustrative example with hypothetical figures, not a real client: consider a hypothetical travel company in Mumbai that handles corporate travel and holiday packages. Its first automation, extracting supplier invoice details into its accounting system, has been working for three months.

Two candidates are on the table.

Candidate A: an AI itinerary builder. Sales staff describe a client request; the AI drafts a day-by-day itinerary with hotels and transfers.

  • Volume: about 30 holiday enquiries a week. Score 2.

  • Exceptions: high. Many requests are custom, and live hotel availability and rates sit in supplier systems that are not connected. Score 1.

  • Measurable value: unclear. Staff time per itinerary varies widely and nobody has timed it. Conversion is affected by many other factors. Score 1.

  • Dependencies: supplier rate access, a content library of hotels and activities, and pricing rules. Score 1.

  • Ownership: the sales head is enthusiastic but has no time to review drafts. Score 1.

Candidate B: matching supplier invoices to bookings and client billing. Each supplier invoice is checked against the booking record and the client invoice to catch missed billing and overcharges.

  • Volume: around 400 supplier invoices a month. Score 3.

  • Exceptions: moderate. Amendments and cancellations create mismatches, but they follow known patterns. Score 2.

  • Measurable value: the accounts team already logs time on reconciliation and records missed recoveries. A baseline exists. Score 3.

  • Dependencies: it builds on the first automation’s extracted invoice data and the booking system’s export. Score 3.

  • Ownership: the accounts manager owns reconciliation and will review the exception queue. Score 3.

Candidate B wins clearly, and it is less exciting. It also strengthens the base for later work: once bookings, supplier costs and client invoices are reliably linked, an itinerary tool can draw on real margins rather than guesses.

Candidate A is not rejected. It is narrowed: the company could start by timing itinerary work for a month and building a reusable hotel and activity library, so a later pilot has a baseline and content to work from.

A ranking checklist to bring to your next planning meeting

For each candidate workflow, write down:

  1. Weekly or monthly volume, from a count, not a guess.

  2. Share of exceptions, from a sample of recent cases.

  3. Current baseline: time, errors, cost or delay, and how it was measured.

  4. Full running cost after launch, including human review.

  5. Systems, data and people it depends on, and whether each is ready.

  6. A named owner who agrees to review output for the first month.

  7. The biggest uncertainty, stated plainly.

  8. A smaller first version that could be tested in weeks rather than months.

If two candidates are close, choose the one you can measure. For more on separating savings from capacity and rework, see our note on AI automation ROI.

Where SolvTree fits

This maps to our free operational audit and scoped engagements: we look at the candidates you have, compare them on the same evidence, and scope the next one as a single workflow with agreed success criteria. Sometimes the answer is a process fix or a small integration rather than a new build, and we will say so.

Next step: Bring two candidate workflows to a free 30-minute operational audit and we will rank them with you. The report is yours to keep.

Frequently asked questions

How should a team rank its next automation?
Compare volume, exception rate, measurable value, dependencies and ownership. Use observed counts and a baseline, include full running and review costs, and record the biggest uncertainty for each candidate.
What if an appealing automation has no measurable baseline?
Narrow the idea and first measure the current work or prepare the data it needs. A smaller candidate with known inputs, clear ownership and measurable value may be a better next step.