Sokomesh Request an audit

The merchant’s side of agentic commerce

The sale was
almost yours.

The product fit. The agent moved on.
Give your store a way to answer
before the decision is made.

Follow one lost sale

We’re building Sokomesh for that moment.

Your store

Still one move to make.

Merchant-approved offers.
A chance to be chosen.
01

The visit you never see

A short story about a missed sale

The buyer is asleep.
The decision isn’t.

At 11:42 PM, an agent arrives with a simple request: a steel dive watch under $500, delivered by Saturday.

Your store has the watch. But at $510 with five-day shipping, it misses two hard limits. The agent moves on.

Your analytics see a visit.
You never see the chance to respond.

Agent activity11:42:08 PM
Buyer request

“A steel dive watch under $500.
Delivered by Saturday.”

01Product matchFound
02Budget · $500$10 over−
03Delivery · SaturdayToo late−
04DecisionNext store
What your analytics see1 page view
Illustrative scenario
02

The same night, with Sokomesh

One request. Another possibility.

A better answer.
The same watch.

You could accept $499. You could deliver on Friday. Sokomesh is being built to turn those approved options into an offer the agent can act on.

The merchant’s side
Your catalogITEM 001

Steel dive watch

The product the buyer is looking for.

Listed price$510
Standard delivery5 days

Your approved options
$495 minimum · Express over $300

The product stays the same.
The response makes the difference.

Illustrative store configuration
Try the same requestIllustrative
Buyer’s limitsUnder $500 By Saturday
Your listed termsPassed over
Price
$510
DeliveryFive days
A good product. Outside the brief.

The agent can only act on the terms it sees.

Your store has room to respond.
$495 price floor · Express over $300

Not a marketplace. Not an ad network. Not a chatbot on your site. Not a checkout.

Agents ask. Your store answers.

A fitting offer earns consideration.
The buyer still makes the final choice.
03

You stay in charge

Flexibility, with a boundary

The offer adapts.
Your limits don’t.

Your products. Your minimum prices. Your delivery options. You set the scope; every proposed offer is checked against it.

AI drafts the answer. Your rules decide whether the offer can leave your store.

Try your price floor
Your offer policy
$495
Express deliveryOrders over $300
Offer expires24 hours · single use
The $499 offer is allowed.

$4 above your floor. Inside your rules.

Move the slider to test the boundary. This changes only the example.

01

AI proposesReads the request. Drafts an answer.

02

Your rules checkEvery term must fit your scope.

03

Your store respondsAn approved offer, ready to verify.

04

What we know so far

Early evidence. Open questions.

The first signal.
The next test.

An early simulation gave us a reason to keep building. The next step is to learn from real stores.

September 2026 · Fictional stores
3 AI models · 24 decisions per version

Original listing

0/ 24

Chosen by the agent
Improved listing data

19/ 24

Chosen by the agent
With a store response

19/ 24

Chosen by the agent
Where an offer made a difference

When facts alone
couldn’t change the terms.

Nine simulated decisions involving a delivery upgrade, a bundle price or an order-specific price.

Improved listing data

4 / 9

With a store response

7 / 9

Three tasks × three models. The data-only condition could add facts and products, but could not change prices; one of its four selections was invalid. These results depend on that setup.

48 acceptance cases passed. The rule-engine prototype was tested using synthetic data.

Both approaches reached 19 selections out of 24 overall. The offer-sensitive subset above shows where their results differed. This is a direction to test, not a live conversion claim.

What this study can—and can’t—tell us
  • Fictional stores. We tested made-up stores and products, not real shops.
  • AI models. We ran three AI models through 24 shopping decisions per version. Nine of those decisions tested whether a store’s offer could make a difference.
  • No merchant interviews yet. The audits are how that starts.
  • No live traffic or sales. An agent choosing a store in a test is not the same as a customer placing an order.

The prototype’s offer rules passed 48 tests with made-up examples. Those tests did not cover live AI models, real traffic or connections to other services.

05

An invitation to build

The founding pilot

Five stores. Built closely.

Your store.
Our next move.

We’re building the first pilot with five merchants who want a say in what comes next.

It starts with a manual audit: five buyer requests your store should win, run on three AI shopping assistants. If we find a missed opportunity, we’ll explore the pilot together.

Tomer ShabatSeven years selling. Nine building software.

A good fit

Shopify or WooCommerce · Physical products
Room to improve price, delivery, availability or fit

  1. 01Your request

    Tell us about your store.

  2. 02The audit

    One page. What happened, and why.

  3. 03The invitation

    If there’s a fit, we build it with you.

Request a store audit

Let’s start
with your store.

Email your store address and a little about what you sell. Tomer will reply personally.

Email Tomer

tomer@sokomesh.com

The submission form is unavailable. Use the address above to send your request.