Answer
Should you buy an AI tool or build the workflow yourself?
Buy where the problem is common and the vendor owns the hard part. Build where the workflow encodes your own decisions.
Buy where the problem is the same everywhere and the vendor maintains the difficult part. Build where the workflow encodes decisions specific to your business, because those are what you would spend the configuration effort recreating anyway.
The test is whether the value is in the problem or in your particular version of it. Transcription, scheduling, document extraction, categorising inbound messages: these are the same problem in every business, someone maintains a product for it, and building your own means reproducing work that is already done and then keeping it working. Buying is straightforwardly right, and the decision is between products rather than between approaches.
The other side is the workflow that encodes your decisions. Which enquiries are worth pursuing and why, how your quotes are structured, what your escalation rules are, what constitutes a complete job in your business. A product cannot know these, so buying one means configuring it until it approximates them, and configuration effort is build effort with less control over the result. Where a product requires extensive configuration to fit, that is a signal that the fit was never there.
Two costs are consistently underweighted on the buy side. Lock-in of data and process — what happens when the price changes or the product is discontinued, and whether your history comes with you. And integration debt, since a bought tool that does not connect to what you already run creates a copying step that someone performs forever. Both are worth asking about before purchase and neither appears in a comparison of features.
One cost is underweighted on the build side, and it is the same one that appears everywhere in this subject: maintenance. A built workflow is a permanent obligation, and it belongs to a business whose staff will change. A bought product is maintained by someone whose business it is, which is worth a considerable amount and is the main argument for buying even when building looks cheaper on the first pass.
The arrangement that works most often is neither: buy the components and assemble the workflow. Use a product for the hard general part — the extraction, the transcription, the classification — and keep the decisions in your own thin layer of rules. That way the difficult technology is maintained by someone else and the part that is specific to your business is yours, editable, and does not depend on a vendor's roadmap.
One practical note on evaluating products. The question that separates them is not what they can do but what happens when they get something wrong: whether you can see it, correct it, and stop it. Products with no answer to that are being sold on capability and will be operated on trust, which is the arrangement this whole subject argues against.
Buy the part everybody has, build the part that is the reason customers choose you.
Siddharth Sharma, Context Theory
Related questions
How do you judge a product demonstration?
Bring your own awkward cases. A demonstration uses examples chosen to work, and the useful information comes from a handful of your real inputs including the messy ones. A vendor who welcomes this is telling you something; one who deflects it is telling you something else, and both answers are worth more than the demonstration.
Is it cheaper to build now that models do most of the work?
The building is cheaper and the maintaining is not, and maintaining is the larger lifetime cost. Lower build cost genuinely shifts some decisions towards building, particularly for small internal workflows, and it does nothing about the recurring obligation, which is what determines whether the thing still exists in two years.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| US SMB retainer, focused one-to-two-service engagement | $1,500–$4,000 | Category-wide |
| Sub-15-minute compliance — automated routing vs manual only | 62.5% vs 39.1% | Category-wide |
2026 agency pricing survey · per month · verified
2026 speed-to-lead benchmark · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Procurement | Where a product requires extensive configuration to fit a business's process, the configuration effort is build effort with less control over the result, which makes the extent of required configuration a signal about fit rather than a setup cost. | Estimating the configuration required for a candidate product against the effort to build the equivalent rules. |
| Buying behaviour | Data and process lock-in and integration debt are the two costs consistently omitted from product comparisons, because neither appears in a feature list and both are incurred after purchase. | Asking a vendor what happens to accumulated data on cancellation and what connects the product to existing systems. |
| Workflow | The hybrid arrangement — buying the general capability and keeping business-specific decisions in an owned rules layer — places the difficult technology with a maintainer and the particular logic under the business's control. | Separating a candidate workflow into its general capability and its business-specific decisions. |
| Constraint | The distinguishing question between products is what happens when one is wrong — whether the error is visible, correctable and stoppable — rather than what the product can do, since capability without those is operated on trust. | Asking a vendor to demonstrate how an incorrect output is surfaced and corrected. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one